{"source": "wikipedia", "title": "Artificial intelligence", "url": "https://en.wikipedia.org/wiki/Artificial_intelligence", "summary": "Artificial intelligence (AI) is the capability of computational systems to perform tasks typically associated with human intelligence, such as learning, reasoning, problem-solving, perception, and decision-making. It is a field of research in engineering, mathematics, and computer science that develops and studies methods and software that enable machines to perceive their environment and use learning and intelligence to take actions that maximise their chances of achieving defined goals.", "content": "Artificial intelligence (AI) is the capability of computational systems to perform tasks typically associated with human intelligence, such as learning, reasoning, problem-solving, perception, and decision-making. It is a field of research in engineering, mathematics, and computer science that develops and studies methods and software that enable machines to perceive their environment and use learning and intelligence to take actions that maximise their chances of achieving defined goals.\nHigh-profile applications of AI include advanced web search engines, chatbots, virtual assistants, autonomous vehicles, play and analysis in strategy games (e.g., chess and Go), and content generation (e.g. images, audio, and videos).\nThe traditional goals of AI research include learning, reasoning, knowledge representation, planning, natural language processing, and perception, as well as support for robotics. To reach these goals, AI researchers use techniques including state space search and mathematical optimisation, formal logic, artificial neural networks, and methods based on statistics, operations research, and economics. AI also draws upon psychology, linguistics, philosophy, neuroscience, and other fields. Some companies, such as OpenAI, Google DeepMind, and Meta, aim to create artificial general intelligence (AGI)—AI that can complete nearly any cognitive task at least as well as a human.\nArtificial intelligence was founded as an academic discipline in 1956. The field went through multiple cycles of optimism throughout its history, followed by periods of disappointment and loss of funding, known as AI winters. Funding and interest increased substantially after 2012, when graphics processing units (GPUs) started being used to accelerate neural networks, and deep learning outperformed previous AI techniques. This growth accelerated further after 2017 with the transformer architecture. In the 2020s, an AI boom coincided with advances in generative AI, which became widespread and allowed for the creation and modification of media. In addition to AI safety and unintended consequences and harms from the use of AI, ethical concerns, AI's long-term effects, environmental effects, and potential existential risks have prompted discussions of AI regulation.\n\n\n== Goals ==\nThe general problem of simulating (or creating) intelligence has been broken down into subproblems. These consist of specific traits or capabilities that researchers expect an intelligent system to display. The traits described below have received the most attention and cover the scope of AI research.\n\n\n=== Reasoning and problem-solving ===\nEarly researchers developed algorithms that imitated step-by-step reasoning that humans use when solving puzzles or making logical deductions. By the late 1980s and 1990s, methods were developed for dealing with uncertain or incomplete information, employing concepts from probability and economics.\nMany of these algorithms were insufficient for solving large reasoning problems because they experienced a \"combinatorial explosion\", meaning they become exponentially slower as the problems grow. Even humans rarely use the step-by-step deduction that early AI research could model. Humans solve most of their problems using fast, intuitive judgments.\nReasoning models, a type of large language model (LLM) trained to generate intermediate chains-of-thought, emerged in 2024 and allowed improved performance on complex problems in mathematics and coding. These models can produce incorrect outputs or \"hallucinations,\" unlike symbolic reasoning systems.\n\n\n=== Knowledge representation ===\n\nAI programs use knowledge to answer questions intelligently and make deductions about real-world facts.\nFormal knowledge representation and knowledge engineering use symbols to represent words, concepts and things in the world. A knowledge base is a body of knowledge represented in a form that can be used by a program. An ontology is the set of objects, relations, concepts, and properties used by a particular domain of knowledge. Formal knowledge has been studied extensively since the 1970s and researchers have developed formalisms for a wide variety of difficult domains.\nThe symbolic approach has difficulty with several problems: the breadth of commonsense knowledge (the set of atomic facts the average person knows is enormous), the sub-symbolic form of most commonsense knowledge (much of what people know is not represented as \"facts\" or \"statements\" they can express verbally), and knowledge acquisition (the problem of obtaining knowledge for AI applications).\nLarge language models (and some other AI programs developed since 2012) do not require explicit, symbolic knowledge. They acquire knowledge by being trained on the combined text of millions of books and billions of websites. Modern AI can also learn about a domain by running experiments (as when AlphaZero learns game strategy by playing against itself). Machine learning solves the problems of general knowledge, commonsense knowledge and knowledge acquisition, however, this approach has struggled with accurate recall and valid reasoning.\n\n\n=== Planning and decision-making ===\nAn \"agent\" is any entity (artificial or not) that perceives and takes actions in the world. A rational agent has goals or preferences and takes actions to make them happen. In automated planning, the agent has a specific goal. In automated decision-making, the agent has preferences—there are some situations it would prefer to be in, and some situations it is trying to avoid. The decision-making agent assigns a number to each situation (called its \"utility\") that measures how much the agent prefers it. For each possible action, it can calculate the \"expected utility\": the utility of all possible outcomes of the action, weighted by the probability that the outcome will occur. It can then choose the action with the maximum expected utility.\nIn classical planning, the agent knows exactly what the effect of any action will be. In most real-world problems, however, the agent may not be certain about the situation it is in (it is \"unknown\" or \"unobservable\") and it may not know for certain what will happen after each possible action (it is not \"deterministic\"). It must choose an action by making a probabilistic guess and then reassess the situation to see if the action worked. \nAlongside thorough testing and improvement based on previous decisions, having an explanation for why the agent took certain decisions is a way to build trust, especially when the decisions have to be relied upon.\nIn some problems, the agent's preferences may be uncertain, especially if there are other agents or humans involved. These preferences may be learned (e.g., with inverse reinforcement learning), or the agent can seek information to improve them. Information value theory can be used to weigh the value of exploratory or experimental actions. The space of possible future actions and situations is typically intractably large, so the agents must take actions and evaluate situations while being uncertain of the outcome.\nA Markov decision process has a transition model that describes the probability that a particular action will change the state in a particular way and a reward function that supplies the utility of each state and the cost of each action. A policy associates a decision with each possible state. The policy could be calculated (e.g., by policy iteration), determined by a heuristic, or learned.\nGame theory describes the rational behaviour of multiple interacting agents and is used in AI programs that make decisions involving other agents.\n\n\n=== Learning ===\nMachine learning is the study of programs that can improve their performance on a given task automatically. It has been a part of AI from the beginning.\n\nThere are several kinds of machine learning: \n\nUnsupervised learning analyses a stream of data, finds patterns, and makes predictions without any other guidance.\nSupervised learning requires labelling the training data with the expected answers, and comes in two main varieties: classification (where the program must learn to predict what category the input belongs in) and regression (where the program must deduce a numeric function based on numeric input).\nReinforcement learning is when the agent is rewarded for good responses and punished for bad ones. The agent learns to choose responses that are classified as \"good\".\nTransfer learning is when the knowledge gained from one problem is applied to a new problem.\nDeep learning is a type of machine learning that runs inputs through biologically inspired artificial neural networks for all of these types of learning.\nComputational learning theory can assess learners by computational complexity, sample complexity (how much data is required), or other notions of optimisation.\n\n\n=== Natural language processing ===\nNatural language processing (NLP) allows programs to read, write, and communicate in human languages. Specific problems include speech recognition, speech synthesis, machine translation, information extraction, information retrieval, and question answering.\nEarly work, based on Noam Chomsky's generative grammar and semantic networks, had difficulty with word-sense disambiguation unless restricted to small domains called \"micro-worlds\" (due to the common sense knowledge problem). British linguist and philosopher Margaret Masterman believed it was meaning and not grammar that was the key to understanding languages, and that dictionaries and especially thesauri should be the basis of computational language structure.\nModern deep learning techniques for NLP include word embedding (representing words, typically as vectors encoding their meaning), transformers (a deep learning architecture using an attention mechanism), and others. In 2019, generative pre-trained transformer (or \"GPT\") language models began to generate coherent text. By 2023, these models were able to get human-level scores on the bar exam, SAT (Scholastic Assessment Test), GRE (Graduate Record Examination), and many other real-world applications.\n\n\n=== Perception ===\nMachine perception is the ability to use input from sensors (such as cameras, microphones, wireless signals, active lidar, sonar, radar, and tactile sensors) to deduce aspects of the world. Computer vision is the ability to analyse visual input.\nThe field includes speech recognition, image classification, facial recognition, object recognition, object tracking, and robotic perception.\n\n\n=== Social intelligence ===\n\nAffective computing is a field that comprises systems that recognise, interpret, process, or simulate human affect (feeling, emotion, and mood). For example, some virtual assistants are programmed to speak conversationally or even banter humorously; it makes them appear more sensitive to the emotional dynamics of human interaction, or to otherwise facilitate human–computer interaction.\nHowever, this tends to give naïve users an unrealistic conception of the intelligence of existing computer agents. Moderate successes related to affective computing include textual sentiment analysis and, more recently, multimodal sentiment analysis, wherein AI classifies the effects displayed by a videotaped subject.\n\n\n=== General intelligence ===\nA machine with artificial general intelligence (AGI) would be able to solve a wide variety of problems with breadth and versatility similar to human intelligence.\n\n\n== Techniques ==\nAI research uses a wide variety of techniques to accomplish the goals above.\n\n\n=== Search and optimisation ===\nThere are two different kinds of search used in AI: state space search and local search:\n\n\n==== State space search ====\nState space search searches through a tree of possible states to try to find a goal state. For example, planning algorithms search through trees of goals and subgoals, attempting to find a path to a target goal, a process called means-ends analysis.\nSimple exhaustive searches are rarely sufficient for most real-world problems: the search space (the number of places to search) quickly grows to astronomical numbers. The result is a search that is too slow or never completes. \"Heuristics\" or \"rules of thumb\" can help prioritise choices that are more likely to reach a goal.\nAdversarial search is used for game-playing programs, such as chess or Go. It searches through a tree of possible moves and countermoves, looking for a winning position.\n\n\n==== Local search ====\n\nLocal search uses mathematical optimisation to find a solution to a problem. It begins with some form of guess and refines it incrementally.\nGradient descent is a type of local search that optimises a set of numerical parameters by incrementally adjusting them to minimise a loss function. Variants of gradient descent are commonly used to train neural networks, through the backpropagation algorithm.\nAnother type of local search is evolutionary computation, which aims to iteratively improve a set of candidate solutions by \"mutating\" and \"recombining\" them, selecting only the fittest to survive each generation.\nDistributed search processes can coordinate via swarm intelligence algorithms. Two popular swarm algorithms used in search are particle swarm optimisation (inspired by bird flocking) and ant colony optimisation (inspired by ant trails).\n\n\n=== Logic ===\nFormal logic is used for reasoning and knowledge representation.\nFormal logic comes in two main forms: propositional logic (which operates on statements that are true or false and uses logical connectives such as \"and\", \"or\", \"not\" and \"implies\") and predicate logic (which also operates on objects, predicates and relations and uses quantifiers such as \"Every X is a Y\" and \"There are some Xs that are Ys\").\nDeductive reasoning in logic is the process of proving a new statement (conclusion) from other statements that are given and assumed to be true (the premises). Proofs can be structured as proof trees, in which nodes are labelled by sentences, and children nodes are connected to parent nodes by inference rules.\nGiven a problem and a set of premises, problem-solving reduces to searching for a proof tree whose root node is labelled by a solution of the problem and whose leaf nodes are labelled by premises or axioms. In the case of Horn clauses, problem-solving search can be performed by reasoning forwards from the premises or backwards from the problem. In the more general case of the clausal form of first-order logic, resolution is a single, axiom-free rule of inference, in which a problem is solved by proving a contradiction from premises that include the negation of the problem to be solved.\nInference in both Horn clause logic and first-order logic is undecidable, and therefore intractable. However, backward reasoning with Horn clauses, which underpins computation in the logic programming language Prolog, is Turing complete. Moreover, its efficiency is competitive with computation in other symbolic programming languages.\nFuzzy logic assigns a \"degree of truth\" between 0 and 1. It can therefore handle propositions that are vague and partially true.\nNon-monotonic logics, including logic programming with negation as failure, are designed to handle default reasoning. Other specialised versions of logic have been developed to describe many complex domains.\n\n\n=== Probabilistic methods for uncertain reasoning ===\n\nMany problems in AI (including reasoning, planning, learning, perception, and robotics) require the agent to operate with incomplete or uncertain information. AI researchers have devised a number of tools to solve these problems using methods from probability theory and economics. Precise mathematical tools have been developed that analyse how an agent can make choices and plan, using decision theory, decision analysis, and information value theory. These tools include models such as Markov decision processes, dynamic decision networks, game theory and mechanism design.\nBayesian networks are a tool that can be used for reasoning (using the Bayesian inference algorithm), learning (using the expectation–maximisation algorithm), planning (using decision networks) and perception (using dynamic Bayesian networks).\nProbabilistic algorithms can also be used for filtering, prediction, smoothing, and finding explanations for streams of data, thus helping perception systems analyse processes that occur over time (e.g., hidden Markov models or Kalman filters).\n\n\n=== Classifiers and statistical learning methods ===\nThe simplest AI applications can be divided into two types: classifiers (e.g., \"if shiny then diamond\"), on one hand, and controllers (e.g., \"if diamond then pick up\"), on the other hand. Classifiers are functions that use pattern matching to determine the closest match. They can be fine-tuned based on chosen examples using supervised learning. Each pattern (also called an \"observation\") is labeled with a certain predefined class. All the observations combined with their class labels are known as a data set. When a new observation is received, that observation is classified based on previous experience.\nThere are many kinds of classifiers in use. The decision tree is the simplest and most widely used symbolic machine learning algorithm. K-nearest neighbour algorithm was the most widely used analogical AI until the mid-1990s, and Kernel methods such as the support vector machine (SVM) displaced k-nearest neighbour in the 1990s.\nThe naive Bayes classifier is reportedly the \"most widely used learner\" at Google, due in part to its scalability.\nNeural networks are also used as classifiers.\n\n\n=== Artificial neural networks ===\n\nAn artificial neural network is based on a collection of nodes also known as artificial neurons, which loosely model the neurons in a biological brain. It is trained to recognise patterns; once trained, it can recognise those patterns in fresh data. There is an input, at least one hidden layer of nodes and an output. Each node applies a function and once the weight crosses its specified threshold, the data is transmitted to the next layer. A network is typically called a deep neural network if it has at least 2 hidden layers.\nLearning algorithms for neural networks use local search to choose the weights that will get the right output for each input during training. The most common training technique is the backpropagation algorithm. Neural networks learn to model complex relationships between inputs and outputs and find patterns in data. In theory, a neural network can learn any function.\nIn feedforward neural networks the signal passes in only one direction. The term perceptron typically refers to a single-layer neural network. In contrast, deep learning uses many layers. Recurrent neural networks (RNNs) feed the output signal back into the input, which allows short-term memories of previous input events. Long short-term memory networks (LSTMs) are recurrent neural networks that better preserve longterm dependencies and are less sensitive to the vanishing gradient problem. Convolutional neural networks (CNNs) use layers of kernels to more efficiently process local patterns. This local processing is especially important in image processing, where the early CNN layers typically identify simple local patterns such as edges and curves, with subsequent layers detecting more complex patterns like textures, and eventually whole objects.\n\n\n=== Deep learning ===\n\nDeep learning uses several layers of neurons between the network's inputs and outputs. The multiple layers can progressively extract higher-level features from the raw input. For example, in image processing, lower layers may identify edges, while higher layers may identify the concepts relevant to a human such as digits, letters, or faces.\nDeep learning has profoundly improved the performance of programs in many important subfields of artificial intelligence, including computer vision, speech recognition, natural language processing, image classification, and others. The sudden success of deep learning in 2012–2015 did not occur because of some new discovery or theoretical breakthrough (deep neural networks and backpropagation had been described by many people, as far back as the 1950s) but because of two factors: the increase in computer power (including the hundred-fold increase in speed by switching to GPUs) and the availability of vast amounts of training data, especially the giant curated datasets used for benchmark testing, such as ImageNet.\n\n\n=== GPT ===\nGenerative pre-trained transformers (GPTs) are large language models (LLMs) that generate text based on the semantic relationships between words in sentences. Text-based GPT models are pre-trained on a large corpus of text that can be from the Internet. The pretraining consists of predicting the next token (a token being usually a word, subword, or punctuation). Throughout this pretraining, GPT models accumulate knowledge about the world and can then generate human-like text by repeatedly predicting the next token. Typically, a subsequent training phase makes the model more truthful, useful, and harmless, usually with a technique called reinforcement learning from human feedback (RLHF). Current GPT models are prone to generating falsehoods called \"hallucinations\". These can be reduced with RLHF and quality data, but the problem has been getting worse for reasoning systems. Such systems are used in chatbots, which allow people to ask a question or request a task in simple text.\nCurrent models and services include ChatGPT, Claude, Gemini, Copilot, and Meta AI. Multimodal GPT models can process different types of data (modalities) such as images, videos, sound, and text.\n\n\n=== Hardware and software ===\n\nIn the late 2010s, graphics processing units (GPUs) that were increasingly designed with AI-specific enhancements and used with specialised TensorFlow software had replaced previously used central processing unit (CPUs) as the dominant means for large-scale (commercial and academic) machine learning models' training. Specialised programming languages such as Prolog were used in early AI research, but general-purpose programming languages like Python have become predominant.\nThe transistor density in integrated circuits has been observed to roughly double every 18 months—a trend known as Moore's law, named after the Intel co-founder Gordon Moore, who first identified it. Improvements in GPUs have been even faster, a trend sometimes called Huang's law, named after Nvidia co-founder and CEO Jensen Huang.\n\n\n== Applications ==\n\nAI and machine learning technology is used in most of the essential applications of the 2020s, including:\n\nsearch engines (such as Google Search)\ntargeting online advertisements\nrecommendation systems (offered by Netflix, YouTube or Amazon) driving internet traffic\ntargeted advertising (AdSense, Facebook)\nvirtual assistants (such as Siri or Alexa)\nautonomous vehicles (including drones, ADAS and self-driving cars)\nautomatic language translation (Microsoft Translator, Google Translate)\nfacial recognition (Apple's FaceID or Facebook's DeepFace and Google's FaceNet)\nimage labelling (used by Facebook, Apple's Photos and TikTok).\n\n\n=== AI-assisted software development ===\n\n\n=== Chatbots ===\n\n\n=== Physical AI ===\n\n\n=== Health and medicine ===\n\nAlphaFold 2 (2021) demonstrated the ability to approximate, in hours rather than months, the 3D structure of a protein. In 2023, it was reported that AI-guided drug discovery helped find a class of antibiotics capable of killing two different types of drug-resistant bacteria. In 2024, researchers used machine learning to accelerate the search for Parkinson's disease drug treatments. Their aim was to identify compounds that block the clumping, or aggregation, of alpha-synuclein (the protein that characterises Parkinson's disease). They were able to speed up the initial screening process ten-fold and reduce the cost by a thousand-fold.\nA 2026 Nature article titled \"Dozens of AI disease-prediction models were trained on dubious data\" highlighted the use of unreliable data being used to train AI medical prediction models for stroke and diabetes in 125 research articles. Evidence suggested some of the AI tools that were developed on unreliable data had been used on patients, although it was not clear if there were adverse outcomes.\n\n\n=== Gaming ===\n\nGame playing programs have been used since the 1950s to demonstrate and test AI's most advanced techniques. Deep Blue became the first computer chess-playing system to beat a reigning world chess champion, Garry Kasparov, on 11 May 1997. In 2011, in a Jeopardy! quiz show exhibition match, IBM's question answering system, Watson, defeated the two greatest Jeopardy! champions, Brad Rutter and Ken Jennings, by a significant margin. In March 2016, AlphaGo won 4 out of 5 games of Go in a match with Go champion Lee Sedol, becoming the first computer Go-playing system to beat a professional Go player without handicaps. Then, in 2017, it defeated Ke Jie, who was the best Go player in the world. Other programs handle imperfect-information games, such as the poker-playing program Pluribus. DeepMind developed increasingly generalistic reinforcement learning models, such as with MuZero, which could be trained to play chess, Go, or Atari games. In 2019, DeepMind's AlphaStar achieved grandmaster level in StarCraft II, a particularly challenging real-time strategy game that involves incomplete knowledge of what happens on the map. In 2021, an AI agent competed in a PlayStation Gran Turismo competition, winning against four of the world's best Gran Turismo drivers using deep reinforcement learning. In 2024, Google DeepMind introduced SIMA, a type of AI capable of autonomously playing nine previously unseen open-world video games by observing screen output, as well as executing short, specific tasks in response to natural language instructions.\n\n\n=== Mathematics ===\n\nIn January 2025, Microsoft proposed the technique rStar-Math that leverages Monte Carlo tree search and step-by-step reasoning, enabling a relatively small language model like Qwen-7B to solve 53% of the AIME 2024 and 90% of the MATH benchmark problems. Google DeepMind has developed models for solving mathematical problems: AlphaDev, AlphaEvolve, AlphaGeometry, AlphaProof, AlphaTensor, and FunSearch.\nWhen natural language is used to describe mathematical problems, converters can transform such prompts into a formal language such as Lean to define mathematical tasks. The experimental model Gemini Deep Think accepts natural language prompts directly and achieved gold medal results in the International Math Olympiad of 2025.\nIn June 2026, an international group of mathematicians published a statement calling on mathematicians, professional organizations, and policymakers to disclose the use of AI, uphold peer review, and protect the openness and values of the discipline.\nIn September 2026, OpenAI announced that one of their models had produced a counter-example to the Navier–Stokes existence and smoothness problem, a Millennium Prize Problem in mathematics. As of 2026, the claim has not been verified by any external mathematicians and is the subject of a priority dispute.\n\n\n=== Finance ===\nAccording to Nicolas Firzli, director of the World Pensions & Investments Forum, it may be too early to see the emergence of highly innovative AI-informed financial products and services. He argues that \"the deployment of AI tools will simply further automatise things: destroying tens of thousands of jobs in banking, financial planning, and pension advice in the process, but I'm not sure it will unleash a new wave of [e.g., sophisticated] pension innovation.\"\n\n\n=== Military ===\n\nMilitary AI usage attempts to enhance command and control, communications, sensors, integration and interoperability. Intelligence collection and analysis, logistics, cyber operations, information operations, and semiautonomous and autonomous vehicles are being researched. AI technologies are used in coordination of sensors and effectors, threat detection and identification, marking of enemy positions, target acquisition, coordination and deconfliction of distributed Joint Fires between networked combat vehicles, both human-operated and autonomous.\nAI has been used in military operations in Iraq, Syria, Israel and Ukraine.\n\n\n=== Generative AI ===\n\n\n=== Agents ===\n\nAI agents are software entities designed to perceive their environment, make decisions, and take actions autonomously to achieve specific goals. These agents can interact with users, their environment, or other agents. AI agents are used in various applications, including virtual assistants, chatbots, autonomous vehicles, game-playing systems, and industrial robotics. AI agents operate within the constraints of their programming, available computational resources, and hardware limitations. This means they are restricted to performing tasks within their defined scope and have finite memory and processing capabilities. In real-world applications, AI agents often face time constraints for decision-making and action execution. Many AI agents incorporate learning algorithms, enabling them to improve their performance over time through experience or training. Using machine learning, AI agents can adapt to new situations and optimise their behaviour for their designated tasks.\n\n\n=== Web search ===\nMicrosoft introduced Copilot Search in February 2023 under the name Bing Chat. Copilot Search provides AI-generated summaries.\nGenerative AI features added to Google Search include AI Overviews in May 2024 and AI Mode in May 2025.\n\n\n=== Sexuality ===\nApplications of AI in this domain include AI-enabled menstruation and fertility trackers that analyse user data to offer predictions, AI-integrated sex toys (e.g., teledildonics), AI-generated sexual education content, and AI agents that simulate sexual and romantic partners (e.g., Replika). AI is also used for the production of non-consensual deepfake pornography, raising significant ethical and legal concerns.\nAI technologies have also been used to attempt to identify online gender-based violence and online sexual grooming of minors.\n\n\n=== Other industry-specific tasks ===\nIn a 2017 survey, one in five companies reported having incorporated \"AI\" in some offerings or processes. \nIn the field of evacuation and disaster management, AI has been used to investigate patterns in large-scale and small-scale evacuations using historical data from GPS, videos or social media.\nDuring the 2024 Indian elections, US$50 million was spent on authorised AI-generated content, notably by creating deepfakes of allied (including sometimes deceased) politicians to better engage with voters, and by translating speeches to various local languages.\nThe use of generative AI by law firms for legal research resulted in the creation of the global \"AI Hallucination Cases\" database, in April 2025, established by HEC Paris and Sciences Po legal data analysis lecturer Damien Charlotin. By 2026, judges had issued sanctions and bar associations had issued warnings due to attorney submissions to the courts containing fabricated case law citations hallucinated by AI tools.\n\n\n== Ethics ==\nAI has potential benefits and potential risks. AI may be able to advance science and find solutions for serious problems: Demis Hassabis of DeepMind hopes to \"solve intelligence, and then use that to solve everything else\". However, as the use of AI has become widespread, several unintended consequences and risks have been identified. In-production systems can sometimes not factor ethics and bias into their AI training processes, especially when the AI algorithms are inherently unexplainable in deep learning.\n\n\n=== Risks and harm ===\n\n\n==== Privacy and copyright ====\n\nMachine learning algorithms require large amounts of data. The techniques used to acquire this data have raised concerns about privacy, surveillance and copyright.\nAI-powered devices and services, such as virtual assistants and IoT products, continuously collect personal information, raising concerns about intrusive data gathering and unauthorised access by third parties. The loss of privacy is further exacerbated by AI's ability to process and combine vast amounts of data, potentially leading to a surveillance society where individual activities are constantly monitored and analysed without adequate safeguards or transparency.\nSensitive user data collected may include online activity records, geolocation data, video, or audio. For example, in order to build speech recognition algorithms, Amazon has recorded millions of private conversations and allowed temporary workers to listen to and transcribe some of them. Opinions about this widespread surveillance range from those who see it as a necessary evil to those for whom it is clearly unethical and a violation of the right to privacy.\nAI developers argue that this is the only way to deliver valuable applications and have developed several techniques that attempt to preserve privacy while still obtaining the data, such as data aggregation, de-identification and differential privacy. Since 2016, some privacy experts, such as Cynthia Dwork, have begun to view privacy in terms of fairness. Brian Christian wrote that experts have pivoted \"from the question of 'what they know' to the question of 'what they're doing with it'.\"\nGenerative AI is often trained on unlicensed copyrighted works, including in domains such as images or computer code; the output is then used under the rationale of \"fair use\". Experts disagree about how well and under what circumstances this rationale will hold up in courts of law; relevant factors may include \"the purpose and character of the use of the copyrighted work\" and \"the effect upon the potential market for the copyrighted work\". Website owners can indicate that they do not want their content scraped via a \"robots.txt\" file. However, some companies will scrape content regardless because the robots.txt file has no real authority. In 2023, leading authors (including John Grisham and Jonathan Franzen) sued AI companies for using their work to train generative AI. Another discussed approach is to envision a separate sui generis system of protection for creations generated by AI to ensure fair attribution and compensation for human authors.\n\n\n==== Dominance by tech giants ====\nThe commercial AI scene is dominated by Big Tech companies such as Alphabet Inc., Amazon, Apple Inc., Meta Platforms, and Microsoft. Some of these companies already own the vast majority of existing cloud infrastructure and computing power from data centres, allowing them to entrench further in the marketplace. For most of 2025-2026, AI-related companies have also dominated stock market returns. According to a July 2024 report from S&P Global, investors had started questioning whether the enormous spending on AI infrastructure and data centers would generate sufficient profits, leading to increased volatility in technology and semiconductor stocks and to concerns about a potential AI bubble. In a 2025 Economic Policy article, economists Anton Korinek and Jai Vipra described some AI related market sectors as highly concentrated.\n\n\n==== Power needs and environmental impacts ====\n\nTechnology companies have built electricity and artificial intelligence infrastructure to facilitate the AI boom of the 2020s. A 2025 report from the consulting firm McKinsey & Company estimated that by 2030, $2.7 trillion would be invested into AI infrastructure and data centres in the US, surpassing World War II's Manhattan Project every month.\nIn January 2024, the International Energy Agency (IEA) released Electricity 2024, Analysis and Forecast to 2026. This is the first IEA report to make projections for data centres and power consumption by AI and cryptocurrency. The report states that power demand for these uses might double by 2026, with the additional power consumption equaling that of Japan.\nPower consumption by AI is responsible for an increase in fossil fuel use, and has delayed closings of obsolete, carbon-emitting coal energy facilities. A ChatGPT search involves 10 times as much electrical energy as a Google search.\nA 2024 Goldman Sachs research paper, AI Data Centers and the Coming US Power Demand Surge, found \"US power demand (is) likely to experience growth not seen in a generation....\" and forecasts that, by 2030, US data centres will consume 8% of US power, as opposed to 3% in 2022, presaging growth for the electrical power generation industry by a variety of means. Data centres' need for more and more electrical power is such that they might max out the electrical grid. The Big Tech companies counter that AI can be used to maximise the utilisation of the grid by all.\nIn 2024, The Wall Street Journal reported that big AI companies have begun negotiations with the US nuclear power providers to provide electricity to the data centres. In March 2024, Amazon purchased a Pennsylvania nuclear-powered data centre for US$650 million.\nIn September 2024, Microsoft announced an agreement with Constellation Energy to re-open the Three Mile Island nuclear power plant to provide Microsoft with 100% of all electric power produced by the plant for 20 years. Reopening the plant, which suffered a partial nuclear meltdown of its Unit 2 reactor in 1979, will require Constellation to get through strict regulatory processes which will include extensive safety scrutiny from the US Nuclear Regulatory Commission. If approved (this will be the first ever US re-commissioning of a nuclear plant), over 835 megawatts of power – enough for 800,000 homes – will be produced. The cost for re-opening and upgrading is estimated at US$1.6 billion and is dependent on tax breaks for nuclear power contained in the 2022 US Inflation Reduction Act. As of 2024, the US government and the state of Michigan have been investing almost US$2 billion to reopen the Palisades nuclear reactor on Lake Michigan. Closed since 2022, the plant was scheduled to be reopened in October 2025.\nAfter the last approval in September 2023, Taiwan suspended the approval of data centres north of Taoyuan with a capacity of more than 5 MW in 2024, due to power supply shortages. Taiwan aims to phase out nuclear power by 2025. \nSingapore imposed a ban on the opening of data centres in 2019 due to electric power, but lifted it in 2022.\nAlthough most nuclear plants in Japan have been shut down after the 2011 Fukushima nuclear accident, according to an October 2024 Bloomberg article in Japanese, cloud gaming services company Ubitus, in which Nvidia has a stake, is looking for land in Japan near a nuclear power plant for a new data centre for generative AI.\nOn 1 November 2024, the Federal Energy Regulatory Commission (FERC) rejected an application submitted by Talen Energy for approval to supply some electricity from the nuclear power station Susquehanna to Amazon's data centre.\nAccording to the Commission Chairman Willie L. Phillips, it is a burden on the electricity grid as well as a significant cost-shifting concern to households and other business sectors.\nIn 2025, a report prepared by the IEA estimated the greenhouse gas emissions from the energy consumption of AI at 180 million tons. By 2035, these emissions could rise to 300–500 million tonnes depending on what measures will be taken. This is below 1.5% of the energy sector emissions. The emissions reduction potential of AI was estimated at 5% of the energy sector emissions, but rebound effects (for example if people switch from public transport to autonomous cars) can reduce it.\nIn June 2026, the United Nations University Institute for Water, Environment and Health released a report about cumulative environmental impacts of AI, not only carbon. The report found that AI has significant negative impacts on land, water, and climate, as it uses large amounts of electricity, requires mining of minerals, and generates large amounts of electronic waste. In 2025, data centres consumed more electricity than Saudi Arabia, and ChatGPT alone consumed an amount of water equivalent to the needs of 500,000 people in Sub-Saharan Africa. The negative impacts are not distributed equally across the planet but are concentrated in some places. Data centres are often built in places already suffering from water deficit. For example, AI deployment has already led to an increase in water scarcity in Mexico and Uruguay. It created problems with energy supply in Ireland, which forbade construction of new data centres near Dublin until 2028. Measures can involve tradeoffs, for example, transitioning from coal to bioenergy reduces carbon footprint but increases the impact on land and water. Energy efficiency might not reduce environmental impact, because as AI becomes more efficient, it also becomes cheaper and more widely used.\n\n\n==== Misinformation ====\n\nYouTube, Facebook and others use recommender systems to guide users to more content. These AI programs were given the goal of maximising user engagement (that is, the only goal was to keep people watching). The AI learned that users tended to choose misinformation, conspiracy theories, and extreme partisan content, and, to keep them watching, the AI recommended more of it. Users also tended to watch more content on the same subject, so the AI led people into filter bubbles where they received multiple versions of the same misinformation. This convinced many users that the misinformation was true, and ultimately undermined trust in institutions, the media and the government. The AI program had correctly learned to maximise its goal, but the result was harmful to society. After the U.S. election in 2016, major technology companies took some steps to mitigate the problem.\nIn the early 2020s, generative AI began to create images, audio, and texts that are virtually indistinguishable from real photographs, recordings, or human writing, while realistic AI-generated videos became feasible in the mid-2020s. It is possible for bad actors to use this technology to create massive amounts of misinformation and computational propaganda through techniques such as deepfakes. AI pioneer and Nobel Prize-winning computer scientist Geoffrey Hinton expressed concern about AI enabling \"authoritarian leaders to manipulate their electorates\" on a large scale, among other risks. The ability to influence electorates has been proved in at least one study. This same study shows more inaccurate statements from the models when they advocate for candidates of the political right.\nAI researchers at Microsoft, OpenAI, universities and other organisations have suggested using \"personhood credentials\" as a way to overcome online deception enabled by AI models.\n\n\n==== Algorithmic bias and fairness ====\n\nMachine learning applications can be biased if they learn from biased data. The developers may not be aware that the bias exists. Discriminatory behaviour by some LLMs can be observed in their output. Bias can be introduced by the way training data is selected and by the way a model is deployed. If a biased algorithm is used to make decisions that can seriously harm people (as it can in medicine, finance, recruitment, housing or policing) then the algorithm may cause discrimination. The field of fairness studies how to prevent harms from algorithmic biases.\nOn 28 June 2015, Google Photos's new image labelling feature mistakenly identified Jacky Alcine and a friend as \"gorillas\" because they were black. The system was trained on a dataset that contained very few images of black people, a problem called \"sample size disparity\". Google \"fixed\" this problem by preventing the system from labelling anything as a \"gorilla\". Eight years later, in 2023, Google Photos still could not identify a gorilla, and neither could similar products from Apple, Facebook, Microsoft and Amazon.\nCOMPAS is a commercial program widely used by U.S. courts to assess the likelihood of a defendant becoming a recidivist. In 2016, Julia Angwin at ProPublica discovered that COMPAS exhibited racial bias, despite the fact that the program was not told the races of the defendants. Although the error rate for both whites and blacks was calibrated equal at exactly 61%, the errors for each race were different—the system consistently overestimated the chance that a black person would re-offend and would underestimate the chance that a white person would not re-offend. In 2017, several researchers showed that it was mathematically impossible for COMPAS to accommodate all possible measures of fairness when the base rates of re-offence were different for whites and blacks in the data.\nA program can make biased decisions even if the data does not explicitly mention a problematic feature (such as \"race\" or \"gender\"). The feature will correlate with other features (like \"address\", \"shopping history\" or \"first name\"), and the program will make the same decisions based on these features as it would on \"race\" or \"gender\". Moritz Hardt said \"the most robust fact in this research area is that fairness through blindness doesn't work.\"\nCriticism of COMPAS highlighted that machine learning models are designed to make \"predictions\" that are only valid if we assume that the future will resemble the past. If they are trained on data that includes the results of racist decisions in the past, machine learning models must predict that racist decisions will be made in the future. If an application then uses these predictions as recommendations, some of these \"recommendations\" will likely be racist. Thus, machine learning is not well suited to help make decisions in areas where there is hope that the future will be better than the past. It is descriptive rather than prescriptive.\nBias and unfairness may go undetected because the developers are overwhelmingly white and male: among AI engineers, about 4% are black and 20% are women.\nThere are various conflicting definitions and mathematical models of fairness. These notions depend on ethical assumptions, and are influenced by beliefs about society. One broad category is distributive fairness, which focuses on the outcomes, often identifying groups and seeking to compensate for statistical disparities. Representational fairness tries to ensure that AI systems do not reinforce negative stereotypes or render certain groups invisible. Procedural fairness focuses on the decision process rather than the outcome. The most relevant notions of fairness may depend on the context, notably the type of AI application and the stakeholders. The subjectivity in the notions of bias and fairness makes it difficult for companies to operationalise them. Having access to sensitive attributes such as race or gender is also considered by many AI ethicists to be necessary in order to compensate for biases, but it may conflict with anti-discrimination laws.\nAt the 2022 ACM Conference on Fairness, Accountability, and Transparency a paper reported that a CLIP‑based (Contrastive Language-Image Pre-training) robotic system reproduced harmful gender‑ and race‑linked stereotypes in a simulated manipulation task. The authors recommended robot‑learning methods which physically manifest such harms be \"paused, reworked, or even wound down when appropriate, until outcomes can be proven safe, effective, and just.\"\n\n\n==== Lack of transparency ====\n\nMany AI systems are so complex that their designers cannot explain how they reach their decisions. Particularly with deep neural networks, in which there are many non-linear relationships between inputs and outputs. But some popular explainability techniques exist.\nIt is impossible to be certain that a program is operating correctly if no one knows how exactly it works. There have been many cases where a machine learning program passed rigorous tests, but nevertheless learned something different from what the programmers intended. For example, a system that could identify skin diseases better than medical professionals was found to actually have a strong tendency to classify images with a ruler as \"cancerous\", because pictures of malignancies typically include a ruler to show the scale. Another machine learning system designed to help effectively allocate medical resources was found to classify patients with asthma as being at \"low risk\" of dying from pneumonia. Having asthma is actually a severe risk factor, but since the patients having asthma would usually get much more medical care, they were relatively unlikely to die according to the training data. The correlation between asthma and low risk of dying from pneumonia was real, but misleading.\nPeople who have been harmed by an algorithm's decision have a right to an explanation. Doctors, for example, are expected to clearly and completely explain to their colleagues the reasoning behind any decision they make. Early drafts of the European Union's General Data Protection Regulation in 2016 included an explicit statement that this right exists. Industry experts noted that this is an unsolved problem with no solution in sight. Regulators argued that nevertheless the harm is real: if the problem has no solution, the tools should not be used.\nDARPA established the XAI (\"Explainable Artificial Intelligence\") program in 2014 to try to solve these problems.\nSeveral approaches aim to address the transparency problem. SHAP enables to visualise the contribution of each feature to the output. LIME can locally approximate a model's outputs with a simpler, interpretable model. Multitask learning provides a large number of outputs in addition to the target classification. These other outputs can help developers deduce what the network has learned. Deconvolution, DeepDream and other generative methods can allow developers to see what different layers of a deep network for computer vision have learned, and produce output that can suggest what the network is learning. For generative pre-trained transformers, Anthropic developed a technique based on dictionary learning that associates patterns of neuron activations with human-understandable concepts.\n\n\n==== Bad actors ====\n\nArtificial intelligence provides a number of tools that are useful to bad actors, such as authoritarian governments, terrorists, criminals or rogue states. There are many ways in which AI is expected to help bad actors, some of which can not be foreseen. For example, machine-learning AI is able to design tens of thousands of toxic molecules in a matter of hours.\n\n\n===== Authoritarian governance =====\nAI tools make it easier for governments to efficiently control their citizens and can increase digital authoritarianism and political repression. Face and voice recognition allow widespread surveillance. Machine learning, operating this data, can classify potential enemies of the state and prevent them from hiding. Recommendation systems can precisely target propaganda and misinformation for maximum effect. Deepfakes and generative AI aid in producing misinformation. Advanced AI can make authoritarian centralised decision-making more competitive than liberal and decentralised systems such as markets. It lowers the cost and difficulty of digital warfare and advanced spyware. AI facial recognition systems are already being used for mass surveillance in China.\n\n\n==== Technological unemployment ====\n\nEconomists have frequently highlighted the risks of redundancies from AI, and speculated about unemployment if there is no adequate social policy for full employment.\nThe Economist said in 2015 that \"the worry that AI could do to white-collar jobs what steam power did to blue-collar ones during the Industrial Revolution\" is \"worth taking seriously\". Jobs at extreme risk range from paralegals to fast food cooks, while job demand is likely to increase for care-related professions ranging from personal healthcare to the clergy. In July 2025, Ford CEO Jim Farley predicted that \"artificial intelligence is going to replace literally half of all white-collar workers in the U.S.\"\nAccording to the Stanford Digital Economy Lab in 2026, early-career workers showed decreasing employment rates in some AI-exposed occupations.\nIn April 2023, Rest of World reported that 70% of the jobs for Chinese video game illustrators had been eliminated by generative artificial intelligence.\n\n\n==== Substitution for human–human interaction ====\n\nWith the increase of loneliness in the early 21st century, AI is sometimes identified as a potential source of relief to this problem. It would be possible, via human-like qualities built into AI products, for individuals to assume that this need can be met by artificial means. In some cases, people approach artificial intelligence for companionship when they believe that they would not find acceptance due to feeling outcast. Examples of harm coming to humans from advanced chatbots have been reported in courts in the United States, with AI companies accused of creating products that endanger humans through emotional confusion or deception.\n\n\n==== Weaponised AI ====\n\nA lethal autonomous weapon is a machine that locates, selects and engages human targets without human supervision. Widely available AI tools can be used by bad actors to develop inexpensive autonomous weapons and, if produced at scale, they are potentially weapons of mass destruction. Even when used in conventional warfare, they currently cannot reliably choose targets and could potentially kill an innocent person. In 2014, 30 nations (including China) supported a ban on autonomous weapons under the United Nations' Convention on Certain Conventional Weapons, however the United States and others disagreed. By 2015, over fifty countries were reported to be researching battlefield robots.\n\n\n==== Existential risk ====\n\nRecent public debates in artificial intelligence have increasingly focused on its broader societal and ethical implications. It has been argued AI will become so powerful that humanity may irreversibly lose control of it. This could, as physicist Stephen Hawking stated, \"spell the end of the human race\". This scenario has been common in science fiction, when a computer or robot suddenly develops a human-like \"self-awareness\" (or \"sentience\" or \"consciousness\") and becomes a malevolent character. These sci-fi scenarios are misleading in several ways.\nFirst, AI does not require human-like sentience to be an existential risk. Modern AI programs are given specific goals and use learning and intelligence to achieve them. Philosopher Nick Bostrom argued that if one gives almost any goal to a sufficiently powerful AI, it may choose to destroy humanity to achieve it (he used the example of an automated paperclip factory that destroys the world to get more iron for paperclips). Stuart Russell gives the example of household robot that tries to find a way to kill its owner to prevent it from being unplugged, reasoning that \"you can't fetch the coffee if you're dead.\" (These are examples of \"instrumental convergence\".) In order to be safe for humanity, a superintelligence would have to be genuinely aligned with humanity's morality and values so that it is \"fundamentally on our side\".\nSecond, Yuval Noah Harari argues that AI does not require a robot body or physical control to pose an existential risk. The essential parts of civilisation are not physical. Things like ideologies, law, government, money and the economy are built on language; they exist because there are stories that billions of people believe. The current prevalence of misinformation suggests that an AI could use language to convince people to believe anything, even to take actions that are destructive. Geoffrey Hinton said in 2025 that modern AI is particularly \"good at persuasion\" and getting better all the time. He asks, \"Suppose you wanted to invade the capital of the US. Do you have to go there and do it yourself? No. You just have to be good at persuasion.\"\nThe opinions amongst experts and industry insiders are mixed, with sizeable fractions both concerned and unconcerned by risk from eventual superintelligent AI. Personalities such as Stephen Hawking, Bill Gates, and Elon Musk, as well as AI pioneers such as Geoffrey Hinton, Yoshua Bengio, Stuart Russell, Demis Hassabis, and Sam Altman, have expressed concerns about existential risk from AI.\nIn May 2023, Geoffrey Hinton announced his resignation from Google in order to be able to \"freely speak out about the risks of AI\" without \"considering how this impacts Google\". He notably mentioned risks of an AI takeover, and stressed that in order to avoid the worst outcomes, establishing safety guidelines will require cooperation among those competing in use of AI.\nIn 2023, many leading AI experts endorsed the joint statement that \"Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war\".\nSome other researchers were more optimistic. AI pioneer Jürgen Schmidhuber did not sign the joint statement, emphasising that in 95% of all cases, AI research is about making \"human lives longer and healthier and easier.\" While the tools that are now being used to improve lives can also be used by bad actors, \"they can also be used against the bad actors.\" Andrew Ng also argued that \"it's a mistake to fall for the doomsday hype on AI—and that regulators who do will only benefit vested interests.\" Yann LeCun, a Turing Award winner, disagreed with the idea that AI will subordinate humans \"simply because they are smarter, let alone destroy [us]\", \"scoff[ing] at his peers' dystopian scenarios of supercharged misinformation and even, eventually, human extinction.\" In contrast, he claimed that \"intelligent machines will usher in a new renaissance for humanity, a new era of enlightenment.\" In the early 2010s, experts argued that the risks are too distant in the future to warrant research or that humans will be valuable from the perspective of a superintelligent machine. However, after 2016, the study of current and future risks and possible solutions became a serious area of research. Some have argued that concerns about existential risks of AI could draw attention away from other AI risks.\n\n\n=== Ethical machines and alignment ===\n\nFriendly AI are machines that have been designed from the beginning to minimise risks and to make choices that benefit humans. Eliezer Yudkowsky, who coined the term, argues that developing friendly AI should be a higher research priority: it may require a large investment and it must be completed before AI becomes an existential risk.\nMachines with intelligence have the potential to use their intelligence to make ethical decisions. The field of machine ethics provides machines with ethical principles and procedures for resolving ethical dilemmas.\nThe field of machine ethics is also called computational morality,\nand was founded at an AAAI symposium in 2005.\nGeoffrey Hinton has suggested researchers explore instilling human-aligned values, such as compassion, into AI systems as a safeguard, speculating about the possibility of building \"maternal instincts\" into AI models, so “they really care about people”.\nOther approaches include Wendell Wallach's \"artificial moral agents\" and Stuart J. Russell's three principles for developing provably beneficial machines.\n\n\n=== Open source ===\n\nActive organisations in the AI open-source community include Hugging Face, Google, EleutherAI and Meta. Various AI models, such as Llama 2, Mistral or Stable Diffusion, have been made open-weight, meaning that their architecture and trained parameters (the \"weights\") are publicly available. Open-weight models can be freely fine-tuned, which allows companies to specialise them with their own data and for their own use-case. Open-weight models are useful for research and innovation but can also be misused. Since they can be fine-tuned, any built-in security measure, such as objecting to harmful requests, can be trained away until it becomes ineffective. Some researchers warn that future AI models may develop dangerous capabilities (such as the potential to drastically facilitate bioterrorism) and that once released on the Internet, they cannot be deleted everywhere if needed. They recommend pre-release audits and cost-benefit analyses.\n\n\n=== Frameworks ===\nArtificial intelligence projects can be guided by ethical considerations during the design, development, and implementation of an AI system. An AI framework such as the Care and Act Framework, developed by the Alan Turing Institute and based on the SUM values, outlines four main ethical dimensions, defined as follows:\n\nRespect the dignity of individual people\nConnect with other people sincerely, openly, and inclusively\nCare for the wellbeing of everyone\nProtect social values, justice, and the public interest\nOther developments in ethical frameworks include those decided upon during the Asilomar Conference, the Montreal Declaration for Responsible AI, and the IEEE's Ethics of Autonomous Systems initiative, among others; however, these principles are not without criticism, especially regarding the people chosen to contribute to these frameworks.\nPromotion of the wellbeing of the people and communities that these technologies affect requires consideration of the social and ethical implications at all stages of AI system design, development and implementation, and collaboration between job roles such as data scientists, product managers, data engineers, domain experts, and delivery managers.\nThe UK AI Safety Institute released in 2024 a testing toolset called 'Inspect' for AI safety evaluations available under an MIT open-source licence which is freely available on GitHub and can be improved with third-party packages. It can be used to evaluate AI models in a range of areas including core knowledge, ability to reason, and autonomous capabilities.\n\n\n=== Regulation ===\n\nThe regulation of artificial intelligence is the development of public sector policies and laws for promoting and regulating AI; it is therefore related to the broader regulation of algorithms. The regulatory and policy landscape for AI is an emerging issue in jurisdictions globally. According to AI Index at Stanford, the annual number of AI-related laws passed in the 127 survey countries jumped from one passed in 2016 to 37 passed in 2022 alone. Between 2016 and 2020, more than 30 countries adopted dedicated strategies for AI. Most EU member states had released national AI strategies, as had Canada, China, India, Japan, Mauritius, the Russian Federation, Saudi Arabia, United Arab Emirates, U.S., and Vietnam. Others were in the process of elaborating their own AI strategy, including Bangladesh, Malaysia and Tunisia. The Global Partnership on Artificial Intelligence was launched in June 2020, stating a need for AI to be developed in accordance with human rights and democratic values, to ensure public confidence and trust in the technology. Henry Kissinger, Eric Schmidt, and Daniel Huttenlocher published a joint statement in November 2021 calling for a government commission to regulate AI. In 2023, OpenAI leaders published recommendations for the governance of superintelligence, which they believe may happen in less than 10 years. In 2023, the United Nations also launched an advisory body to provide recommendations on AI governance; the body comprises technology company executives, government officials and academics. On 1 August 2024, the EU Artificial Intelligence Act entered into force, establishing the first comprehensive EU-wide AI regulation. In 2024, the Council of Europe created the first international legally binding treaty on AI, called the \"Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law\". It was adopted by the European Union, the United States, the United Kingdom, and other signatories.\nIn a 2022 Ipsos survey, attitudes towards AI varied greatly by country; 78% of Chinese citizens, but only 35% of Americans, agreed that \"products and services using AI have more benefits than drawbacks\". A 2023 Reuters/Ipsos poll found that 61% of Americans agree, and 22% disagree, that AI poses risks to humanity. In a 2023 Fox News poll, 35% of Americans thought it \"very important\", and an additional 41% thought it \"somewhat important\", for the federal government to regulate AI, versus 13% responding \"not very important\" and 8% responding \"not at all important\". A 2026 NBC News Decision Desk poll powered by SurveyMonkey found that 52% of American adults used AI tools very often or sometimes, but only 18% said they could trust AI-generated information most or almost all of the time, and 70% were more worried than excited about AI. Polls in the United States found the main public concerns were job losses, impersonation, misinformation and information privacy.\nIn November 2023, the first global AI Safety Summit was held in Bletchley Park in the UK to discuss the near and far term risks of AI and the possibility of mandatory and voluntary regulatory frameworks. 28 countries including the United States, China, and the European Union issued a declaration at the start of the summit, calling for international co-operation to manage the challenges and risks of artificial intelligence. In May 2024 at the AI Seoul Summit, 16 global AI tech companies agreed to safety commitments on the development of AI.\nIn March 2026, the United Nations convened the inaugural meeting of the Independent International Scientific Panel on AI, a 40-member expert body established under the Global Digital Compact to produce annual evidence-based reports on AI's societal impacts.\n\n\n== History ==\n\nThe study of mechanical or \"formal\" reasoning began with philosophers and mathematicians in antiquity. The study of logic led directly to Alan Turing's theory of computation, which suggested that a machine, by shuffling symbols as simple as \"0\" and \"1\", could simulate any conceivable form of mathematical reasoning. This, along with concurrent discoveries in cybernetics, information theory and neurobiology, led researchers to consider the possibility of building an \"electronic brain\". They developed several areas of research that would become part of AI, such as McCulloch and Pitts design for \"artificial neurons\" in 1943, and Turing's influential 1950 paper 'Computing Machinery and Intelligence', which introduced the Turing test and showed that \"machine intelligence\" was plausible.\nThe field of AI research was founded at a workshop at Dartmouth College in 1956. The first AI program, Logic Theorist, was presented at the workshop, created by future Turing Award winner Allen Newell and future Nobel Laureate Herbert A. Simon, in collaboration with J. C. Shaw. Many of the workshop attendees became the leaders of AI research in the 1960s. They and their students produced programs that the press described as \"astonishing\": computers were learning checkers strategies, solving word problems in algebra, proving logical theorems and speaking English. Artificial intelligence laboratories were set up at a number of British and U.S. universities in the latter 1950s and early 1960s.\nResearchers in the 1960s and the 1970s were convinced that their methods would eventually succeed in creating a machine with general intelligence and considered this the goal of their field. In 1965 Herbert Simon predicted, \"machines will be capable, within twenty years, of doing any work a man can do\".In 1967 Marvin Minsky agreed, writing that \"within a generation ... the problem of creating 'artificial intelligence' will substantially be solved\". They had, however, underestimated the difficulty of the problem. In 1974, both the U.S. and British governments cut off exploratory research in response to the criticism of Sir James Lighthill and ongoing pressure from the U.S. Congress to fund more productive projects. Minsky and Papert's book Perceptrons was understood as proving that artificial neural networks would never be useful for solving real-world tasks, thus discrediting the approach altogether. The \"AI winter\", a period when obtaining funding for AI projects was difficult, followed.\nIn the early 1980s, AI research was revived by the commercial success of expert systems, a form of AI program that simulated the knowledge and analytical skills of human experts. By 1985, the market for AI had reached over a billion dollars. At the same time, Japan's fifth generation computer project inspired the U.S. and British governments to restore funding for academic research. However, beginning with the collapse of the Lisp Machine market in 1987, AI once again fell into disrepute, and a second, longer-lasting winter began.\nUp to this point, most of AI's funding had gone to projects that used high-level symbols to represent mental objects like plans, goals, beliefs, and known facts. In the 1980s, some researchers began to doubt that this approach would be able to imitate all the processes of human cognition, especially perception, robotics, learning and pattern recognition, and began to look into \"sub-symbolic\" approaches. Rodney Brooks rejected \"representation\" in general and focussed directly on engineering machines that move and survive. Judea Pearl, Lotfi Zadeh, and others developed methods that handled incomplete and uncertain information by making reasonable guesses rather than precise logic. But the most important development was the revival of \"connectionism\", including neural network research, by Geoffrey Hinton and others. In 1990, Yann LeCun successfully showed that convolutional neural networks can recognise handwritten digits, the first of many successful applications of neural networks.\nAI gradually restored its reputation in the late 1990s and early 21st century by exploiting formal mathematical methods and by finding specific solutions to specific problems. This \"narrow\" and \"formal\" focus allowed researchers to produce verifiable results and collaborate with other fields (such as statistics, economics and mathematics). By 2000, solutions developed by AI researchers were being widely used, although in the 1990s they were rarely described as \"artificial intelligence\" (a tendency known as the AI effect).\nHowever, several academic researchers became concerned that AI was no longer pursuing its original goal of creating versatile, fully intelligent machines. Beginning around 2002, they founded the subfield of artificial general intelligence (or \"AGI\"), which had several well-funded institutions by the 2010s.\nDeep learning began to dominate industry benchmarks in 2012 and was adopted throughout the field.\nFor many specific tasks, other methods were abandoned.\nDeep learning's success was based on both hardware improvements (faster computers, graphics processing units, cloud computing) and access to large amounts of data (including curated datasets, such as ImageNet). Neural-network approaches also advanced natural language processing. In 2013, word2vec introduced efficient methods for learning distributed word representations. In 2014–2015, sequence-to-sequence and attention mechanisms advanced neural machine translation.\nDeep learning's success led to an enormous increase in interest and funding in AI. The amount of machine learning research (measured by total publications) increased by 50% in the years 2015–2019.\n\nIn 2016, issues of fairness and the misuse of technology were catapulted into centre stage at machine learning conferences, publications vastly increased, funding became available, and many researchers re-focussed their careers on these issues. The alignment problem became a serious field of academic study.\nIn 2015, AlphaGo, developed by DeepMind, beat the world champion Go player. The program taught only the game's rules and developed a strategy by itself. GPT-3 is a large language model that was released in 2020 by OpenAI and is capable of generating high-quality human-like text. ChatGPT, launched on 30 November 2022, became the fastest-growing consumer software application in history, gaining over 100 million users in two months. It marked what is widely regarded as AI's breakout year, bringing it into the public consciousness. These programs, and others, inspired an aggressive AI boom, where large companies began investing billions of dollars in AI research. According to AI Impacts, about US$50 billion annually was invested in \"AI\" around 2022 in the U.S. alone and about 20% of the new U.S. computer science PhD graduates have specialised in \"AI\". About 800,000 \"AI\"-related U.S. job openings existed in 2022. According to PitchBook research, 22% of newly funded startups in 2024 claimed to be AI companies.\n\n\n== Philosophy ==\n\nPhilosophical debates have historically sought to determine the nature of intelligence and how to make intelligent machines. Another major focus has been whether machines can be conscious, and the associated ethical implications. Many other topics in philosophy are relevant to AI, such as epistemology and free will. Rapid advancements have intensified public discussions on the philosophy and ethics of AI.\n\n\n=== Defining artificial intelligence ===\n\nAlan Turing investigated whether machines can show intelligent behaviour and think. In 1950, he proposed the Turing test, which measures the ability of a machine to simulate human conversation. Since we can only observe the behaviour of the machine, it does not matter if it is \"actually\" thinking or literally has a \"mind\". Turing notes that we can not determine these things about other people but \"it is usual to have a polite convention that everyone thinks.\"\n\nRussell and Norvig agree with Turing that intelligence must be defined in terms of external behaviour, not internal structure. However, they are critical that the test requires the machine to imitate humans. \"Aeronautical engineering texts\", they wrote, \"do not define the goal of their field as making 'machines that fly so exactly like pigeons that they can fool other pigeons.'\" AI founder John McCarthy agreed, writing that \"Artificial intelligence is not, by definition, simulation of human intelligence\".\nMcCarthy defines intelligence as \"the computational part of the ability to achieve goals in the world\". Another AI founder, Marvin Minsky, similarly describes it as \"the ability to solve hard problems\". Artificial Intelligence: A Modern Approach defines it as the study of agents that perceive their environment and take actions that maximise their chances of achieving defined goals. \nThe many differing definitions of AI have been critically analysed. During the 2020s AI boom, the term has been used as a marketing buzzword to promote products and services which do not use AI.\n\n\n==== Legal definitions ====\nThe International Organization for Standardization describes an AI system as a \"an engineered system that generates outputs such as content, forecasts, recommendations, or decisions for a given set of human‑defined objectives, and can operate with varying levels of automation\". The EU AI Act defines an AI system as \"a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments\". In the United States, influential but non‑binding guidance such as the National Institute of Standards and Technology's AI Risk Management Framework describes an AI system as \"an engineered or machine-based system that can, for a given set of objectives, generate outputs such as predictions, recommendations, or decisions influencing real or virtual environments. AI systems are designed to operate with varying levels of autonomy\".\n\n\n=== Evaluating approaches to AI ===\nNo established unifying theory or paradigm has guided AI research for most of its history. The unprecedented success of statistical machine learning in the 2010s eclipsed all other approaches (so much so that some sources, especially in the business world, use the term \"artificial intelligence\" to mean \"machine learning with neural networks\"). This approach is mostly sub-symbolic, soft and narrow.\n\n\n==== Symbolic AI and its limits ====\nSymbolic AI (or \"GOFAI\") simulated the high-level conscious reasoning that people use when they solve puzzles, express legal reasoning and do mathematics. It was highly successful at some \"intelligent\" tasks such as algebra or IQ tests. In 1976, Newell and Simon proposed the physical symbol systems hypothesis: \"A physical symbol system has the necessary and sufficient means of general intelligent action.\"\nHowever, the symbolic approach failed on many tasks that humans solve easily, such as learning, recognising an object or commonsense reasoning. Moravec's paradox is the discovery that high-level \"intelligent\" tasks were easy for AI, but low level \"instinctive\" tasks were extremely difficult. Philosopher Hubert Dreyfus had argued since the 1960s that human expertise depends on embodied, situational know-how that is not representable as explicit symbolic rules rather than conscious symbol manipulation, and on having a \"feel\" for the situation, rather than explicit symbolic knowledge. Although his arguments had been ridiculed and ignored when they were first presented, eventually, AI research came to agree with him.\nThe issue is not resolved: sub-symbolic reasoning can make many of the same inscrutable mistakes that human intuition does, such as algorithmic bias. Critics such as Noam Chomsky argue continuing research into symbolic AI will still be necessary to attain general intelligence, in part because sub-symbolic AI is a move away from explainable AI: it can be difficult or impossible to understand why a modern statistical AI program made a particular decision. The emerging field of neuro-symbolic artificial intelligence attempts to bridge the two approaches.\n\n\n==== Neat vs. scruffy ====\n\n\"Neats\" hope that intelligent behaviour is described using simple, elegant principles (such as logic or optimisation). \"Scruffies\" expect that it necessarily requires solving a large number of unrelated problems. Neats defend their programs with theoretical rigour, scruffies rely mainly on incremental testing to see if they work. This issue was actively discussed in the 1970s and 1980s. The rise of deep learning may represent a shift toward the scruffies. \n\n\n==== Soft vs. hard computing ====\n\nFinding a provably correct or optimal solution is intractable for many important problems. Soft computing is a set of techniques, including genetic algorithms, fuzzy logic and neural networks, that are tolerant of imprecision, uncertainty, partial truth and approximation. Soft computing was introduced in the late 1980s and most successful AI programs in the 21st century are examples of soft computing with neural networks.\n\n\n==== Narrow vs. general AI ====\n\nAI researchers are divided as to whether to pursue the goals of artificial general intelligence and superintelligence directly or to solve as many specific problems as possible (narrow AI) in hopes these solutions will lead indirectly to the field's long-term goals. General intelligence is difficult to define and difficult to measure, and modern AI has had more verifiable successes by focusing on specific problems with specific solutions. The sub-field of artificial general intelligence studies this area exclusively.\n\n\n=== Machine consciousness, sentience, and mind ===\n\nThere is no settled consensus in philosophy of mind on whether a machine can have a mind, consciousness or subjective experiences in the same sense that human beings do. This issue considers the internal experiences of the machine, rather than its external behaviour. Mainstream AI research considers this issue irrelevant because it does not affect the goals of the field: to build machines that can solve problems using intelligence. Russell and Norvig add that \"[t]he additional project of making a machine conscious in exactly the way humans are is not one that we are equipped to take on.\" However, the question has become central to the philosophy of mind. It is also typically the central question at issue in artificial intelligence in fiction.\n\n\n==== Consciousness ====\n\nDavid Chalmers identified two problems in understanding the mind, which he named the \"hard\" and \"easy\" problems of consciousness. The easy problem is understanding how the brain processes signals, makes plans and controls behaviour. The hard problem is explaining how this feels or why it should feel like anything at all, assuming we are right in thinking that it truly does feel like something (Dennett's consciousness illusionism says this is an illusion). While human information processing is easy to explain, human subjective experience is difficult to explain. For example, it is easy to imagine a colour-blind person who has learned to identify which objects in their field of view are red, but it is not clear what would be required for the person to know what red looks like.\n\n\n==== Computationalism and functionalism ====\n\nComputationalism is the position in the philosophy of mind that the human mind is an information processing system and that thinking is a form of computing. Computationalism argues that the relationship between mind and body is similar or identical to the relationship between software and hardware and thus may be a solution to the mind–body problem. This philosophical position was inspired by the work of AI researchers and cognitive scientists in the 1960s and was originally proposed by philosophers Jerry Fodor and Hilary Putnam.\nPhilosopher John Searle characterised this position as \"strong AI\": \"The appropriately programmed computer with the right inputs and outputs would thereby have a mind in exactly the same sense human beings have minds.\" Searle challenges this claim with his Chinese room argument, which attempts to show that even a computer capable of perfectly simulating human behaviour would not have a mind.\n\n\n==== AI welfare and rights ====\n\nIt is difficult or impossible to reliably evaluate whether an advanced AI is sentient (has the ability to feel), and if so, to what degree. But if there is a significant chance that a given machine can feel and suffer, then it may be entitled to certain rights or welfare protection measures, similarly to animals. Sapience (a set of capacities related to high intelligence, such as discernment or self-awareness) may provide another moral basis for AI rights. Robot rights are also sometimes proposed as a practical way to integrate autonomous agents into society.\nIn 2017, the European Union considered granting \"electronic personhood\" to some of the most capable AI systems. Similarly to the legal status of companies, it would have conferred rights but also responsibilities. Critics argued in 2018 that granting rights to AI systems would downplay the importance of human rights, and that legislation should focus on user needs rather than speculative futuristic scenarios. They also noted that robots lacked the autonomy to take part in society on their own.\nProgress in AI increased interest in the topic. Proponents of AI welfare and rights often argue that AI sentience, if it emerges, would be particularly easy to deny. They warn that this may be a moral blind spot analogous to slavery or factory farming, which could lead to large-scale suffering if sentient AI is created and carelessly exploited.\n\n\n== Future ==\n\n\n=== Superintelligence and the singularity ===\nA superintelligence is a hypothetical agent that would possess intelligence far surpassing that of the brightest and most gifted human mind. If research into artificial general intelligence produced sufficiently intelligent software, it might be able to reprogram and improve itself. The improved software would be even better at improving itself, leading to what I. J. Good called an \"intelligence explosion\" and Vernor Vinge called a \"singularity\".\nIn Artificial Intelligence: A Modern Approach, Russel and Norvig cast doubt on the possibility of a singularity, arguing that \"so far, every technology has followed an S-shaped curve\", and that \"sometimes it is not possible to keep the growth going.\"\n\n\n=== Transhumanism ===\n\nRobot designer Hans Moravec, cyberneticist Kevin Warwick and inventor Ray Kurzweil have predicted that humans and machines may merge in the future into cyborgs that are more capable and powerful than either. This idea, called transhumanism, has roots in the writings of Aldous Huxley and Robert Ettinger.\nEdward Fredkin argues that \"artificial intelligence is the next step in evolution\", an idea first proposed by Samuel Butler's \"Darwin among the Machines\" as far back as 1863, and expanded upon by George Dyson in his 1998 book Darwin Among the Machines: The Evolution of Global Intelligence.\n\n\n== In fiction ==\n\nThought-capable artificial beings have appeared as storytelling devices since antiquity, and have been a persistent theme in science fiction.\nA common trope in these works began with Mary Shelley's Frankenstein, where a human creation becomes a threat to its masters. This includes such works as Arthur C. Clarke's and Stanley Kubrick's 2001: A Space Odyssey (both 1968), with HAL 9000, the murderous computer in charge of the Discovery One spaceship, as well as Blade Runner (1982), The Terminator (1984) and The Matrix (1999). In contrast, the rare loyal robots such as Gort from The Day the Earth Stood Still (1951) and Bishop from Aliens (1986) are less prominent in popular culture.\nIsaac Asimov introduced the Three Laws of Robotics in many stories, most notably with the \"Multivac\" super-intelligent computer. Asimov's laws are often brought up during lay discussions of machine ethics; while almost all artificial intelligence researchers are familiar with Asimov's laws through popular culture, they generally consider the laws useless for many reasons, one of which is their ambiguity.\nSeveral works use AI to force us to confront the fundamental question of what makes us human, showing us artificial beings that have the ability to feel, and thus to suffer. This appears in Karel Čapek's R.U.R., the films A.I. Artificial Intelligence and Ex Machina, as well as the novel Do Androids Dream of Electric Sheep?, by Philip K. Dick. Dick considers the idea that our understanding of human subjectivity is altered by technology created with artificial intelligence.\nIn the manga and anime franchise Doraemon, there are many gadgets associated with Artificial Inteligence and Machine Learning technologies and it's consequences if used in the wrong way were depicted by Fujiko F. Fujio in manga.\nDoraemon: Nobita and the Tin Labyrinth movie depicted consequences of making a AI robot, which can think on it's own and innovate technologies but instead it puts the humans in jail. \n\n\n== See also ==\n\n\n== Explanatory notes ==\n\n\n== Notes ==\n\n\n== References ==\n\n\n=== Textbooks ===\n\n\n=== History of AI ===\n\n\n=== Other sources ===\n\n\n== External links ==\n\nHauser, Larry. \"Artificial Intelligence\". In Fieser, James; Dowden, Bradley (eds.). Internet Encyclopedia of Philosophy. ISSN 2161-0002. OCLC 37741658.", "extra": {}}
{"source": "wikipedia", "title": "List of artificial intelligence journals", "url": "https://en.wikipedia.org/wiki/List_of_artificial_intelligence_journals", "summary": "This is a list of notable peer-reviewed academic journals that publish research in the field of artificial intelligence (AI), including areas such as machine learning, computer vision, natural language processing, robotics, and intelligent systems.", "content": "This is a list of notable peer-reviewed academic journals that publish research in the field of artificial intelligence (AI), including areas such as machine learning, computer vision, natural language processing, robotics, and intelligent systems.\n\n\n== General artificial intelligence ==\nArtificial Intelligence (journal) – Elsevier\nJournal of Artificial Intelligence Research (JAIR) – AI Access Foundation\nKnowledge-Based Systems – Elsevier\n\n\n== Machine learning ==\nData Mining and Knowledge Discovery – Springer\nMachine Learning (journal) – Springer\nJournal of Machine Learning Research – Microtome\nPattern Recognition (journal) – Elsevier\nNeural Networks (journal) – Elsevier\nNeural Computation (journal) – MIT Press\nNeurocomputing (journal) - Elsevier\n\n\n== Deep learning and neural computation ==\nIEEE Transactions on Evolutionary Computation – IEEE\nIEEE Transactions on Neural Networks and Learning Systems – IEEE\nNature Machine Intelligence – Springer Nature\n\n\n== Computer vision ==\nInternational Journal of Computer Vision – Springer\nIEEE Transactions on Pattern Analysis and Machine Intelligence – IEEE\nMachine Vision and Applications – Springer\n\n\n== Natural language processing ==\nComputational Linguistics (journal) – MIT Press\nNatural Language Processing (journal) - Elsevier\nTransactions of the Association for Computational Linguistics – ACL\nNatural Language Processing (journal) - Cambridge\n\n\n== Robotics and intelligent systems ==\n\nIEEE Transactions on Robotics – IEEE\nAutonomous Robots – Springer\nJournal of Intelligent & Robotic Systems – Springer\n\n\n== Interdisciplinary and ethics in AI ==\nAI & Society – Springer\nArtificial Life – MIT Press\nPhilosophy & Technology – Springer\nMinds and Machines – Springer\n\n\n== See also ==\nAssociation for the Advancement of Artificial Intelligence\nLists of academic journals\nList of software programming journals\nList of computer science journals\nList of information systems journals\nList of large language models\nList of artificial intelligence books\nList of artificial intelligence algorithms\nList of artificial intelligence projects\nLists of open-source artificial intelligence software\n\n\n=== Artificial intelligence conferences ===\n\nAAAI Conference on Artificial Intelligence\nAI Action Summit\nInternational Joint Conference on Artificial Intelligence\nConference on Neural Information Processing Systems\nInternational Conference on Learning Representations\nEuropean Conference on Artificial Intelligence\nNvidia GTC\nInternational Conference on Autonomous Agents and Multiagent Systems\nInternational Conference on Machine Learning\n\n\n== External links ==\nDBLP – Computer Science Bibliography", "extra": {}}
{"source": "wikipedia", "title": "2025–present global memory supply shortage", "url": "https://en.wikipedia.org/wiki/2025–present_global_memory_supply_shortage", "summary": "A global computer memory supply shortage started in 2025 due to supply constraints and rapid price escalation in the semiconductor memory market, particularly affecting DRAM and NAND flash integrated circuits. This shortage has been referred to in media outlets as \"RAMmageddon\" or the \"RAMpocalypse\". Unlike the 2020–2023 global chip shortage, which stemmed primarily from pandemic-related supply chain disruptions from COVID-19, this shortage is driven by the intentional reallocation of manufacturing capacity toward highly profitable products intended for AI data center development, creating a scarcity of integrated circuits intended for the consumer and enterprise markets. According to a 2026 Kearney's PERLab analysis, the shortage is expected to last at least until 2030. In June 2026, Micron Technology CEO Sanjay Mehrotra expected the shortage to last through 2027, and supply to gradually improve by 2028. In August 2026, Jaejune Kim of Samsung Electronics expects the shortage to continue through 2028, while Kwak Noh-Jung of SK Hynix expects it to persist until 2030.", "content": "A global computer memory supply shortage started in 2025 due to supply constraints and rapid price escalation in the semiconductor memory market, particularly affecting DRAM and NAND flash integrated circuits. This shortage has been referred to in media outlets as \"RAMmageddon\" or the \"RAMpocalypse\". Unlike the 2020–2023 global chip shortage, which stemmed primarily from pandemic-related supply chain disruptions from COVID-19, this shortage is driven by the intentional reallocation of manufacturing capacity toward highly profitable products intended for AI data center development, creating a scarcity of integrated circuits intended for the consumer and enterprise markets. According to a 2026 Kearney's PERLab analysis, the shortage is expected to last at least until 2030. In June 2026, Micron Technology CEO Sanjay Mehrotra expected the shortage to last through 2027, and supply to gradually improve by 2028. In August 2026, Jaejune Kim of Samsung Electronics expects the shortage to continue through 2028, while Kwak Noh-Jung of SK Hynix expects it to persist until 2030.\n\n\n== Background ==\nFollowing a severe market downturn during 2022–2023, major memory manufacturers—Samsung Electronics, SK Hynix, and Micron Technology—implemented strategic production cuts to stabilize pricing. By mid-2024, the rapid expansion of generative AI services triggered unprecedented demand for specialized memory products, particularly high bandwidth memory (HBM) used in AI accelerators and data center GPUs. Specialized components of semiconductor technology are also experiencing supply constraints due to high demand in AI applications. For example, glass cloth, a high-performance glass fiber substrate used for power-efficient high speed data transfer and a crucial component of semiconductor manufacturing, is experiencing a supply crisis. Nitto Boseki, a Japanese firm having overwhelming monopoly in its production, is not able to meet increased demands, making chip-makers such as Qualcomm, Apple, Nvidia, and AMD compete for securing supply. There are also reports of smaller electronics companies struggling to find suppliers for components such as NAND flash. Memory suppliers are adapting to increased demands and market unpredictability by requiring prepayment or shorter time-frame of payment, which makes it more difficult for smaller firms to acquire capital to survive. By 2026, due to steadily increased demand on resources, CPUs are also experiencing shortage issues due to low fabrication capacity, prioritization of server CPUs, and increased demand, with CPU prices also being forecast to increase by as much as 15%.\nThe demand on memory has also increased strain on other electronic components such as hard disk devices, with reports such as Western Digital's hard disk supply for 2026 being booked for enterprise applications before February 2026.\nA 2024 McKinsey analysis projected that global demand for AI-ready data center capacity would grow at approximately 33% annually through 2030, with AI workloads consuming roughly 70% of total data center capacity by the decade's end. In addition, according to Kearney's State of Semiconductor 2025 Report, executives were already expecting a shortage in the sub–8 nm wafer size with memory chips being mentioned as an acute source of concern. Multiple companies mentioned being prepared for it through long-term agreements with RAM suppliers or amassing additional inventory.\nOn March 24, 2026, Google announced TurboQuant, a memory compression technology focused on large language models (LLM) and vector search engines, which it claimed achieves 6 times lower memory consumption in tested local LLMs, and 8-fold performance enhancement in tests running on H100 accelerators. The technology is also a drop-in enhancement for existing inference pipeline. Amid speculation about memory demand trends, memory manufacturers, Sandisk, Micron, Western Digital, and Seagate, among other companies involved in memory manufacture experienced stock price declines. Prices of memory kits also reduced in the following months, although still at inflated prices.\nIn August 2026, Intel hinted it was investigating a new memory architecture.\n\n\n== Development ==\nOn July 8, 2026, Micron announced it had broken ground on a new $9.3 billion fab for memory production.  It is expected to start production in Q3 2028.\n\n\n== Causes ==\n\n\n=== HBM production displacement ===\nHBM manufacturing requires significantly more wafer capacity per bit than standard DRAM modules. Industry sources reported that as manufacturers allocated increasing wafer capacity to HBM production to meet contracts with AI infrastructure providers, the supply of conventional DDR4 and DDR5 modules for consumer PCs and smartphones contracted sharply.\nBy September 2025, Samsung Electronics had reportedly expanded its 1c DRAM capacity to target 60,000 wafers per month specifically for HBM4 production, further diverting resources from consumer memory lines.\n\n\n=== Geopolitical and trade barriers ===\nThe supply chain was further constrained by escalating trade tensions between the United States and China. Throughout 2025, fears of U.S. regulatory backlash and new tariff structures led major manufacturers like Samsung and SK Hynix to halt sales of older semiconductor manufacturing equipment to Chinese entities, effectively capping production capacity in the region. Additionally, proposed tariff policies by the U.S. administration in late 2025 prompted supply chain realignments, with Apple reportedly accelerating plans to source all U.S.-bound iPhones from India to avoid potential levies.\n\n\n=== NAND flash capacity constraints ===\nIn the NAND flash segment, manufacturers prioritized higher-margin enterprise SSDs for data center applications while phasing out older process nodes more rapidly than anticipated. In November 2025, contract prices for NAND wafers increased by more than 60% month-over-month for certain product categories, with 512 GB TLC experiencing the steepest rise as legacy manufacturing capacity was retired.\n\n\n== Impact on other industries ==\n\n\n=== Memory manufacturer responses ===\nSamsung is concerned about the possibility of a demand shortfall by 2028. Chey Tae-won, Chairman of SK Group, estimated the memory shortage will last until 2030. In 2026, Kye-hyun Kyung, advisor to Samsung and former head of Samsung's chip group, suggested memory prices could drop within a year.\n\n\n=== PC manufacturers ===\nMajor PC manufacturers responded to component cost increases with significant price raises and supply chain strategies. Dell Technologies Chief Operating Officer Jeff Clarke stated during a November 2025 analyst call that the company had \"never witnessed costs escalating at the current pace\", describing tighter availability across DRAM, hard drives, and NAND flash memory.\nAnalysts at Morgan Stanley downgraded Dell Technologies stock from \"Overweight\" to \"Underweight\" in late 2025, citing the company's heavy exposure to rising server memory costs. The firm warned that skyrocketing memory prices could erode margins for server and PC OEMs. Conversely, Apple Inc. was reportedly less affected than its competitors, having secured long-term supply agreements for DRAM through the first quarter of 2026. However, on June 25, 2026, Apple announced price hikes effective immediately on all iPads, Macs, HomePods, Apple Vision Pro, and Apple TV, causing Apple shares to decline by more than 6%, the worst day for the stock since the April 2025 stock market crash. \nLenovo chief financial officer Winston Cheng described the cost surge as \"unprecedented\" and disclosed that the company's memory inventories were approximately 50% above normal levels in anticipation of further price increases.\n\n\n=== Consumer electronics sector ===\nThe shortage particularly affected smartphone manufacturers and other consumer electronics producers. DRAM prices reportedly rose by 172% throughout 2025, leading manufacturers like Samsung to halt new orders for DDR5 modules to reassess pricing structures and Micron to exit its 'Crucial' brand of consumer products.\nIn Tokyo's Akihabara electronics district, retailers began limiting purchases of memory products to prevent hoarding, with prices for popular DDR5 memory modules more than doubling in some cases. Despite the broad trend of rising hardware costs, some companies engaged in aggressive pricing strategies to maintain market share; for example, Sony reduced the price of the PlayStation 5 by US$100 for Black Friday 2025, potentially absorbing increased component costs to stimulate software ecosystem growth.\nDue to memory prices more than doubling in a single quarter, HP revealed in its Q1 2026 earnings call that memory costs accounted for 35% of PC build materials up from 15 to 18% previous quarter. Despite showing strong Q1 2026 earnings driven by Windows 11 upgrade cycle and AI PC adoption, HP warned investors of low operating margins and up to double digit percentage decline for the coming quarter. Trendforce, an IT analytics company, updated its forecast from 1.7% year-over-year growth in PC market to 2.6% year-over-year decline for 2026, amid backdrop of steadily increasing prices and supply crisis. Research and analytics firms, Gartner and IDC, expected the worldwide PC market to decline 10–11%, and the smartphone market to decline 8–9% in 2026. Gartner also projected that rising memory prices would make low-margin entry level laptops under US$500 financially unviable in two years.\nValve's second Steam Machine, originally set to launch in early 2026, was delayed because of the shortage. When it released in June, the company said \"our original goal for the price of Steam Machine is no longer viable\" and retailed it at US$1,049. Analyst Piers Harding-Rolls noted the price was 75% more than the PlayStation 5 at the time, calling it a \"niche offering\".\n\n\n=== AI infrastructure competition ===\nTechnology companies including Google, Amazon, Microsoft, and Meta Platforms placed open-ended orders with memory suppliers, indicating they would accept as much supply as available regardless of cost, according to Reuters sources. \nThe limited supply of AI chips has been cited as a reason for the slow down in compute growth.\nIn October 2025, OpenAI formally announced a strategic partnership using letters of intent with Samsung Electronics and SK Hynix to secure supply for its Stargate Project AI infrastructure. Reports indicated that the Stargate initiative alone would consume up to 40% of global DRAM output, requiring approximately 900,000 wafers per month. The deal involved the supply of undiced wafers rather than packaged chips to streamline logistics for the massive data center build-out.\nNvidia, whose AI processors require substantial amounts of high-bandwidth memory, acknowledged significant price increases but stated it had secured adequate supply. CNBC reported that Nvidia's increasing adoption of LPDDR memory for its AI products added additional pressure to supply chains, as this memory type is also used in premium consumer electronics.\n\n\n== Legal issues ==\nSamsung, SK Hynix, and Micron—the three largest manufacturers of DRAM—were sued in an antitrust class-action lawsuit over potentially taking advantage of the supply shortage by price-fixing. The complaint alleges they conspired to manipulate the commodity DRAM market through price-fixing and artificial supply constraints. Plaintiffs assert that the manufacturers used the transition to HBM as a pretext to simultaneously cut production of conventional DRAM and exit legacy product lines like DDR3.\n\n\n== See also ==\nHigh Bandwidth Memory\nSemiconductor industry\nAI boom\nAI bubble\nList of data compression algorithms\nList of DRAM manufacturers\nTurboQuant\nvLLM – open-source LLM inference engine that uses the PagedAttention algorithm to improve memory efficiency of AI models\n\n\n== References ==", "extra": {}}
{"source": "wikipedia", "title": "The 2028 Global Intelligence Crisis", "url": "https://en.wikipedia.org/wiki/The_2028_Global_Intelligence_Crisis", "summary": "The 2028 Global Intelligence Crisis is a report authored by James van Geelen and Alap Shah and published by Citrini Research in February 2026, on the impact of artificial intelligence on humanity's future. Written in the form of a scenario analysis, it was viewed millions of times online and reportedly caused a fall in the stock market prices of major tech and financial firms. It also received criticism among others, for its allegedly flawed economic logic.", "content": "The 2028 Global Intelligence Crisis is a report authored by James van Geelen and Alap Shah and published by Citrini Research in February 2026, on the impact of artificial intelligence on humanity's future. Written in the form of a scenario analysis, it was viewed millions of times online and reportedly caused a fall in the stock market prices of major tech and financial firms. It also received criticism among others, for its allegedly flawed economic logic.\nThe 'thought exercise', as the authors called it, painted a gloomy picture for the near future, where outputs keep growing while consumer's ability to spend collapses. \"...driven by ai agents that don’t sleep, take sick days or require health insurance”, \"outputs that are shown in national accounts increases, \"but never circulates through the real economy\"(which the report calls 'Ghost GDP'), the authors argued. In other words, the authors predict a scenario where the owners of the AI firms will accumulate a vast fortune but there will be scant demand from consumers as AI would cause massive unemployment.  \nThe authors caution the reader that what they make is a scenario and not a prediction. In the scenario they visualise, any service whose value proposition is “I will navigate complexity that you find tedious” is getting disrupted. The reports argues that the unique ability of human beings to analyse, decide, create, persuade, and coordinate was “the thing that could not be replicated at scale,” and call the historical scarcity of this precious entity 'friction'. When this friction becomes zero, a gamut of changes occur which then triggers a cascading of changes across the economy. \n”Travel booking platforms are an early casualty; Financial advice. tax prep., and routine legal work follow suit. National unemployment rate go as high 10.2% and the S&P 500 goes for a massive 38% peak-to-trough crash. In contrast to the previous technological revolutions the high-earning professionals suffers more and get forced to take up roles in the gig economy. Labour supply becomes abundant and this cuts wages all across the economy. The dent in income for the employees then affects other sectors of the economy such as the residential mortgage market. The losses for the software companies triggers loan defaults and heralds peril for the private credit sector.\n\n\n== References ==", "extra": {}}
{"source": "wikipedia", "title": "Actor-critic algorithm", "url": "https://en.wikipedia.org/wiki/Actor-critic_algorithm", "summary": "The actor-critic algorithm (AC) is a family of reinforcement learning (RL) algorithms that combine policy-based RL algorithms such as policy gradient methods, and value-based RL algorithms such as value iteration, Q-learning, SARSA, and TD learning.", "content": "The actor-critic algorithm (AC) is a family of reinforcement learning (RL) algorithms that combine policy-based RL algorithms such as policy gradient methods, and value-based RL algorithms such as value iteration, Q-learning, SARSA, and TD learning.\nAn AC algorithm consists of two main components: an \"actor\" that determines which actions to take according to a policy function, and a \"critic\" that evaluates those actions according to a value function. Some AC algorithms are on-policy, some are off-policy. Some apply to either continuous or discrete action spaces. Some work in both cases.\n\n\n== Overview ==\nThe actor-critic methods can be understood as an improvement over pure policy gradient methods like REINFORCE via introducing a baseline.\n\n\n=== Actor ===\nThe actor uses a policy function \n  \n    \n      \n        π\n        (\n        a\n        \n          |\n        \n        s\n        )\n      \n    \n    {\\displaystyle \\pi (a|s)}\n  \n, while the critic estimates either the value function \n  \n    \n      \n        V\n        (\n        s\n        )\n      \n    \n    {\\displaystyle V(s)}\n  \n, the action-value Q-function \n  \n    \n      \n        Q\n        (\n        s\n        ,\n        a\n        )\n        ,\n      \n    \n    {\\displaystyle Q(s,a),}\n  \n the advantage function \n  \n    \n      \n        A\n        (\n        s\n        ,\n        a\n        )\n      \n    \n    {\\displaystyle A(s,a)}\n  \n, or any combination thereof.\nThe actor is a parameterized function \n  \n    \n      \n        \n          π\n          \n            θ\n          \n        \n      \n    \n    {\\displaystyle \\pi _{\\theta }}\n  \n, where \n  \n    \n      \n        θ\n      \n    \n    {\\displaystyle \\theta }\n  \n are the parameters of the actor. The actor takes as argument the state of the environment \n  \n    \n      \n        s\n      \n    \n    {\\displaystyle s}\n  \n and produces a probability distribution \n  \n    \n      \n        \n          π\n          \n            θ\n          \n        \n        (\n        ⋅\n        \n          |\n        \n        s\n        )\n      \n    \n    {\\displaystyle \\pi _{\\theta }(\\cdot |s)}\n  \n.\nIf the action space is discrete, then \n  \n    \n      \n        \n          ∑\n          \n            a\n          \n        \n        \n          π\n          \n            θ\n          \n        \n        (\n        a\n        \n          |\n        \n        s\n        )\n        =\n        1\n      \n    \n    {\\displaystyle \\sum _{a}\\pi _{\\theta }(a|s)=1}\n  \n. If the action space is continuous, then \n  \n    \n      \n        \n          ∫\n          \n            a\n          \n        \n        \n          π\n          \n            θ\n          \n        \n        (\n        a\n        \n          |\n        \n        s\n        )\n        d\n        a\n        =\n        1\n      \n    \n    {\\displaystyle \\int _{a}\\pi _{\\theta }(a|s)da=1}\n  \n.\nThe goal of policy optimization is to improve the actor. That is, to find some \n  \n    \n      \n        θ\n      \n    \n    {\\displaystyle \\theta }\n  \n that maximizes the expected episodic reward \n  \n    \n      \n        J\n        (\n        θ\n        )\n      \n    \n    {\\displaystyle J(\\theta )}\n  \n:\n  \n    \n      \n        J\n        (\n        θ\n        )\n        =\n        \n          \n            E\n          \n          \n            \n              π\n              \n                θ\n              \n            \n          \n        \n        \n          [\n          \n            \n              ∑\n              \n                t\n                =\n                0\n              \n              \n                T\n              \n            \n            \n              γ\n              \n                t\n              \n            \n            \n              r\n              \n                t\n              \n            \n          \n          ]\n        \n      \n    \n    {\\displaystyle J(\\theta )=\\mathbb {E} _{\\pi _{\\theta }}\\left[\\sum _{t=0}^{T}\\gamma ^{t}r_{t}\\right]}\n  \nwhere \n  \n    \n      \n        γ\n      \n    \n    {\\displaystyle \\gamma }\n  \n is the discount factor, \n  \n    \n      \n        \n          r\n          \n            t\n          \n        \n      \n    \n    {\\displaystyle r_{t}}\n  \n is the reward at step \n  \n    \n      \n        t\n      \n    \n    {\\displaystyle t}\n  \n, and \n  \n    \n      \n        T\n      \n    \n    {\\displaystyle T}\n  \n is the time-horizon (which can be infinite).\nThe goal of policy gradient method is to optimize \n  \n    \n      \n        J\n        (\n        θ\n        )\n      \n    \n    {\\displaystyle J(\\theta )}\n  \n by gradient ascent on the policy gradient \n  \n    \n      \n        ∇\n        J\n        (\n        θ\n        )\n      \n    \n    {\\displaystyle \\nabla J(\\theta )}\n  \n.\nAs detailed on the policy gradient method page, there are many unbiased estimators of the policy gradient:\n  \n    \n      \n        \n          ∇\n          \n            θ\n          \n        \n        J\n        (\n        θ\n        )\n        =\n        \n          \n            E\n          \n          \n            \n              π\n              \n                θ\n              \n            \n          \n        \n        \n          [\n          \n            \n              ∑\n              \n                0\n                ≤\n                j\n                ≤\n                T\n              \n            \n            \n              ∇\n              \n                θ\n              \n            \n            ln\n            ⁡\n            \n              π\n              \n                θ\n              \n            \n            (\n            \n              A\n              \n                j\n              \n            \n            \n              |\n            \n            \n              S\n              \n                j\n              \n            \n            )\n            ⋅\n            \n              Ψ\n              \n                j\n              \n            \n            \n              \n                |\n              \n            \n            \n              S\n              \n                0\n              \n            \n            =\n            \n              s\n              \n                0\n              \n            \n          \n          ]\n        \n      \n    \n    {\\displaystyle \\nabla _{\\theta }J(\\theta )=\\mathbb {E} _{\\pi _{\\theta }}\\left[\\sum _{0\\leq j\\leq T}\\nabla _{\\theta }\\ln \\pi _{\\theta }(A_{j}|S_{j})\\cdot \\Psi _{j}{\\Big |}S_{0}=s_{0}\\right]}\n  \nwhere \n  \n    \n      \n        \n          Ψ\n          \n            j\n          \n        \n      \n    \n    {\\textstyle \\Psi _{j}}\n  \n is a linear sum of the following:\n\n  \n    \n      \n        \n          ∑\n          \n            0\n            ≤\n            i\n            ≤\n            T\n          \n        \n        (\n        \n          γ\n          \n            i\n          \n        \n        \n          R\n          \n            i\n          \n        \n        )\n      \n    \n    {\\textstyle \\sum _{0\\leq i\\leq T}(\\gamma ^{i}R_{i})}\n  \n.\n\n  \n    \n      \n        \n          γ\n          \n            j\n          \n        \n        \n          ∑\n          \n            j\n            ≤\n            i\n            ≤\n            T\n          \n        \n        (\n        \n          γ\n          \n            i\n            −\n            j\n          \n        \n        \n          R\n          \n            i\n          \n        \n        )\n      \n    \n    {\\textstyle \\gamma ^{j}\\sum _{j\\leq i\\leq T}(\\gamma ^{i-j}R_{i})}\n  \n: the REINFORCE algorithm.\n\n  \n    \n      \n        \n          γ\n          \n            j\n          \n        \n        \n          ∑\n          \n            j\n            ≤\n            i\n            ≤\n            T\n          \n        \n        (\n        \n          γ\n          \n            i\n            −\n            j\n          \n        \n        \n          R\n          \n            i\n          \n        \n        )\n        −\n        b\n        (\n        \n          S\n          \n            j\n          \n        \n        )\n      \n    \n    {\\textstyle \\gamma ^{j}\\sum _{j\\leq i\\leq T}(\\gamma ^{i-j}R_{i})-b(S_{j})}\n  \n: the REINFORCE with baseline algorithm. Here \n  \n    \n      \n        b\n      \n    \n    {\\displaystyle b}\n  \n is an arbitrary function.\n\n  \n    \n      \n        \n          γ\n          \n            j\n          \n        \n        \n          (\n          \n            \n              R\n              \n                j\n              \n            \n            +\n            γ\n            \n              V\n              \n                \n                  π\n                  \n                    θ\n                  \n                \n              \n            \n            (\n            \n              S\n              \n                j\n                +\n                1\n              \n            \n            )\n            −\n            \n              V\n              \n                \n                  π\n                  \n                    θ\n                  \n                \n              \n            \n            (\n            \n              S\n              \n                j\n              \n            \n            )\n          \n          )\n        \n      \n    \n    {\\textstyle \\gamma ^{j}\\left(R_{j}+\\gamma V^{\\pi _{\\theta }}(S_{j+1})-V^{\\pi _{\\theta }}(S_{j})\\right)}\n  \n: TD(1) learning.\n\n  \n    \n      \n        \n          γ\n          \n            j\n          \n        \n        \n          Q\n          \n            \n              π\n              \n                θ\n              \n            \n          \n        \n        (\n        \n          S\n          \n            j\n          \n        \n        ,\n        \n          A\n          \n            j\n          \n        \n        )\n      \n    \n    {\\textstyle \\gamma ^{j}Q^{\\pi _{\\theta }}(S_{j},A_{j})}\n  \n.\n\n  \n    \n      \n        \n          γ\n          \n            j\n          \n        \n        \n          A\n          \n            \n              π\n              \n                θ\n              \n            \n          \n        \n        (\n        \n          S\n          \n            j\n          \n        \n        ,\n        \n          A\n          \n            j\n          \n        \n        )\n      \n    \n    {\\textstyle \\gamma ^{j}A^{\\pi _{\\theta }}(S_{j},A_{j})}\n  \n: Advantage Actor-Critic (A2C).\n\n  \n    \n      \n        \n          γ\n          \n            j\n          \n        \n        \n          (\n          \n            \n              R\n              \n                j\n              \n            \n            +\n            γ\n            \n              R\n              \n                j\n                +\n                1\n              \n            \n            +\n            \n              γ\n              \n                2\n              \n            \n            \n              V\n              \n                \n                  π\n                  \n                    θ\n                  \n                \n              \n            \n            (\n            \n              S\n              \n                j\n                +\n                2\n              \n            \n            )\n            −\n            \n              V\n              \n                \n                  π\n                  \n                    θ\n                  \n                \n              \n            \n            (\n            \n              S\n              \n                j\n              \n            \n            )\n          \n          )\n        \n      \n    \n    {\\textstyle \\gamma ^{j}\\left(R_{j}+\\gamma R_{j+1}+\\gamma ^{2}V^{\\pi _{\\theta }}(S_{j+2})-V^{\\pi _{\\theta }}(S_{j})\\right)}\n  \n: TD(2) learning.\n\n  \n    \n      \n        \n          γ\n          \n            j\n          \n        \n        \n          (\n          \n            \n              ∑\n              \n                k\n                =\n                0\n              \n              \n                n\n                −\n                1\n              \n            \n            \n              γ\n              \n                k\n              \n            \n            \n              R\n              \n                j\n                +\n                k\n              \n            \n            +\n            \n              γ\n              \n                n\n              \n            \n            \n              V\n              \n                \n                  π\n                  \n                    θ\n                  \n                \n              \n            \n            (\n            \n              S\n              \n                j\n                +\n                n\n              \n            \n            )\n            −\n            \n              V\n              \n                \n                  π\n                  \n                    θ\n                  \n                \n              \n            \n            (\n            \n              S\n              \n                j\n              \n            \n            )\n          \n          )\n        \n      \n    \n    {\\textstyle \\gamma ^{j}\\left(\\sum _{k=0}^{n-1}\\gamma ^{k}R_{j+k}+\\gamma ^{n}V^{\\pi _{\\theta }}(S_{j+n})-V^{\\pi _{\\theta }}(S_{j})\\right)}\n  \n: TD(n) learning.\n\n  \n    \n      \n        \n          γ\n          \n            j\n          \n        \n        \n          ∑\n          \n            n\n            =\n            1\n          \n          \n            ∞\n          \n        \n        \n          \n            \n              λ\n              \n                n\n                −\n                1\n              \n            \n            \n              1\n              −\n              λ\n            \n          \n        \n        ⋅\n        \n          (\n          \n            \n              ∑\n              \n                k\n                =\n                0\n              \n              \n                n\n                −\n                1\n              \n            \n            \n              γ\n              \n                k\n              \n            \n            \n              R\n              \n                j\n                +\n                k\n              \n            \n            +\n            \n              γ\n              \n                n\n              \n            \n            \n              V\n              \n                \n                  π\n                  \n                    θ\n                  \n                \n              \n            \n            (\n            \n              S\n              \n                j\n                +\n                n\n              \n            \n            )\n            −\n            \n              V\n              \n                \n                  π\n                  \n                    θ\n                  \n                \n              \n            \n            (\n            \n              S\n              \n                j\n              \n            \n            )\n          \n          )\n        \n      \n    \n    {\\textstyle \\gamma ^{j}\\sum _{n=1}^{\\infty }{\\frac {\\lambda ^{n-1}}{1-\\lambda }}\\cdot \\left(\\sum _{k=0}^{n-1}\\gamma ^{k}R_{j+k}+\\gamma ^{n}V^{\\pi _{\\theta }}(S_{j+n})-V^{\\pi _{\\theta }}(S_{j})\\right)}\n  \n: TD(λ) learning, also known as GAE (generalized advantage estimate). This is obtained by an exponentially decaying sum of the TD(n) learning terms.\n\n\n=== Critic ===\nIn the unbiased estimators given above, certain functions such as \n  \n    \n      \n        \n          V\n          \n            \n              π\n              \n                θ\n              \n            \n          \n        \n        ,\n        \n          Q\n          \n            \n              π\n              \n                θ\n              \n            \n          \n        \n        ,\n        \n          A\n          \n            \n              π\n              \n                θ\n              \n            \n          \n        \n      \n    \n    {\\displaystyle V^{\\pi _{\\theta }},Q^{\\pi _{\\theta }},A^{\\pi _{\\theta }}}\n  \n appear. These are approximated by the critic. Since these functions all depend on the actor, the critic must learn alongside the actor. The critic is learned by value-based RL algorithms.\nFor example, if the critic is estimating the state-value function \n  \n    \n      \n        \n          V\n          \n            \n              π\n              \n                θ\n              \n            \n          \n        \n        (\n        s\n        )\n      \n    \n    {\\displaystyle V^{\\pi _{\\theta }}(s)}\n  \n, then it can be learned by any value function approximation method. Let the critic be a function approximator \n  \n    \n      \n        \n          V\n          \n            ϕ\n          \n        \n        (\n        s\n        )\n      \n    \n    {\\displaystyle V_{\\phi }(s)}\n  \n with parameters \n  \n    \n      \n        ϕ\n      \n    \n    {\\displaystyle \\phi }\n  \n.\nThe simplest example is TD(1) learning, which trains the critic to minimize the TD(1) error:\n  \n    \n      \n        \n          δ\n          \n            i\n          \n        \n        =\n        \n          R\n          \n            i\n          \n        \n        +\n        γ\n        \n          V\n          \n            ϕ\n          \n        \n        (\n        \n          S\n          \n            i\n            +\n            1\n          \n        \n        )\n        −\n        \n          V\n          \n            ϕ\n          \n        \n        (\n        \n          S\n          \n            i\n          \n        \n        )\n      \n    \n    {\\displaystyle \\delta _{i}=R_{i}+\\gamma V_{\\phi }(S_{i+1})-V_{\\phi }(S_{i})}\n  \nThe critic parameters are updated by gradient descent on the squared TD error:\n  \n    \n      \n        ϕ\n        ←\n        ϕ\n        −\n        α\n        \n          ∇\n          \n            ϕ\n          \n        \n        (\n        \n          δ\n          \n            i\n          \n        \n        \n          )\n          \n            2\n          \n        \n        =\n        ϕ\n        +\n        α\n        \n          δ\n          \n            i\n          \n        \n        \n          ∇\n          \n            ϕ\n          \n        \n        \n          V\n          \n            ϕ\n          \n        \n        (\n        \n          S\n          \n            i\n          \n        \n        )\n      \n    \n    {\\displaystyle \\phi \\leftarrow \\phi -\\alpha \\nabla _{\\phi }(\\delta _{i})^{2}=\\phi +\\alpha \\delta _{i}\\nabla _{\\phi }V_{\\phi }(S_{i})}\n  \nwhere \n  \n    \n      \n        α\n      \n    \n    {\\displaystyle \\alpha }\n  \n is the learning rate. Note that the gradient is taken with respect to the \n  \n    \n      \n        ϕ\n      \n    \n    {\\displaystyle \\phi }\n  \n in \n  \n    \n      \n        \n          V\n          \n            ϕ\n          \n        \n        (\n        \n          S\n          \n            i\n          \n        \n        )\n      \n    \n    {\\displaystyle V_{\\phi }(S_{i})}\n  \n only, since the \n  \n    \n      \n        ϕ\n      \n    \n    {\\displaystyle \\phi }\n  \n in \n  \n    \n      \n        γ\n        \n          V\n          \n            ϕ\n          \n        \n        (\n        \n          S\n          \n            i\n            +\n            1\n          \n        \n        )\n      \n    \n    {\\displaystyle \\gamma V_{\\phi }(S_{i+1})}\n  \n constitutes a moving target, and the gradient is not taken with respect to that. This is a common source of error in implementations that use automatic differentiation, and requires \"stopping the gradient\" at that point.\nSimilarly, if the critic is estimating the action-value function \n  \n    \n      \n        \n          Q\n          \n            \n              π\n              \n                θ\n              \n            \n          \n        \n      \n    \n    {\\displaystyle Q^{\\pi _{\\theta }}}\n  \n, then it can be learned by Q-learning or SARSA. In SARSA, the critic maintains an estimate of the Q-function, parameterized by \n  \n    \n      \n        ϕ\n      \n    \n    {\\displaystyle \\phi }\n  \n, denoted as \n  \n    \n      \n        \n          Q\n          \n            ϕ\n          \n        \n        (\n        s\n        ,\n        a\n        )\n      \n    \n    {\\displaystyle Q_{\\phi }(s,a)}\n  \n. The temporal difference error is then calculated as \n  \n    \n      \n        \n          δ\n          \n            i\n          \n        \n        =\n        \n          R\n          \n            i\n          \n        \n        +\n        γ\n        \n          Q\n          \n            θ\n          \n        \n        (\n        \n          S\n          \n            i\n            +\n            1\n          \n        \n        ,\n        \n          A\n          \n            i\n            +\n            1\n          \n        \n        )\n        −\n        \n          Q\n          \n            θ\n          \n        \n        (\n        \n          S\n          \n            i\n          \n        \n        ,\n        \n          A\n          \n            i\n          \n        \n        )\n      \n    \n    {\\displaystyle \\delta _{i}=R_{i}+\\gamma Q_{\\theta }(S_{i+1},A_{i+1})-Q_{\\theta }(S_{i},A_{i})}\n  \n. The critic is then updated by\n  \n    \n      \n        θ\n        ←\n        θ\n        +\n        α\n        \n          δ\n          \n            i\n          \n        \n        \n          ∇\n          \n            θ\n          \n        \n        \n          Q\n          \n            θ\n          \n        \n        (\n        \n          S\n          \n            i\n          \n        \n        ,\n        \n          A\n          \n            i\n          \n        \n        )\n      \n    \n    {\\displaystyle \\theta \\leftarrow \\theta +\\alpha \\delta _{i}\\nabla _{\\theta }Q_{\\theta }(S_{i},A_{i})}\n  \nThe advantage critic can be trained by training both a Q-function \n  \n    \n      \n        \n          Q\n          \n            ϕ\n          \n        \n        (\n        s\n        ,\n        a\n        )\n      \n    \n    {\\displaystyle Q_{\\phi }(s,a)}\n  \n and a state-value function \n  \n    \n      \n        \n          V\n          \n            ϕ\n          \n        \n        (\n        s\n        )\n      \n    \n    {\\displaystyle V_{\\phi }(s)}\n  \n, then let \n  \n    \n      \n        \n          A\n          \n            ϕ\n          \n        \n        (\n        s\n        ,\n        a\n        )\n        =\n        \n          Q\n          \n            ϕ\n          \n        \n        (\n        s\n        ,\n        a\n        )\n        −\n        \n          V\n          \n            ϕ\n          \n        \n        (\n        s\n        )\n      \n    \n    {\\displaystyle A_{\\phi }(s,a)=Q_{\\phi }(s,a)-V_{\\phi }(s)}\n  \n. Although, it is more common to train just a state-value function \n  \n    \n      \n        \n          V\n          \n            ϕ\n          \n        \n        (\n        s\n        )\n      \n    \n    {\\displaystyle V_{\\phi }(s)}\n  \n, then estimate the advantage by\n  \n    \n      \n        \n          A\n          \n            ϕ\n          \n        \n        (\n        \n          S\n          \n            i\n          \n        \n        ,\n        \n          A\n          \n            i\n          \n        \n        )\n        ≈\n        \n          ∑\n          \n            j\n            ∈\n            0\n            :\n            n\n            −\n            1\n          \n        \n        \n          γ\n          \n            j\n          \n        \n        \n          R\n          \n            i\n            +\n            j\n          \n        \n        +\n        \n          γ\n          \n            n\n          \n        \n        \n          V\n          \n            ϕ\n          \n        \n        (\n        \n          S\n          \n            i\n            +\n            n\n          \n        \n        )\n        −\n        \n          V\n          \n            ϕ\n          \n        \n        (\n        \n          S\n          \n            i\n          \n        \n        )\n      \n    \n    {\\displaystyle A_{\\phi }(S_{i},A_{i})\\approx \\sum _{j\\in 0:n-1}\\gamma ^{j}R_{i+j}+\\gamma ^{n}V_{\\phi }(S_{i+n})-V_{\\phi }(S_{i})}\n  \nHere, \n  \n    \n      \n        n\n      \n    \n    {\\displaystyle n}\n  \n is a positive integer. The higher \n  \n    \n      \n        n\n      \n    \n    {\\displaystyle n}\n  \n is, the more lower is the bias in the advantage estimation, but at the price of higher variance.\nThe Generalized Advantage Estimation (GAE) introduces a hyperparameter \n  \n    \n      \n        λ\n      \n    \n    {\\displaystyle \\lambda }\n  \n that smoothly interpolates between Monte Carlo returns (\n  \n    \n      \n        λ\n        =\n        1\n      \n    \n    {\\displaystyle \\lambda =1}\n  \n, high variance, no bias) and 1-step TD learning (\n  \n    \n      \n        λ\n        =\n        0\n      \n    \n    {\\displaystyle \\lambda =0}\n  \n, low variance, high bias). This hyperparameter can be adjusted to pick the optimal bias-variance trade-off in advantage estimation. It uses an exponentially decaying average of n-step returns with \n  \n    \n      \n        λ\n      \n    \n    {\\displaystyle \\lambda }\n  \n being the decay strength.\n\n\n== Variants ==\nAsynchronous Advantage Actor-Critic (A3C): Parallel and asynchronous version of A2C.\nSoft Actor-Critic (SAC): Incorporates entropy maximization for improved exploration.\nDeep Deterministic Policy Gradient (DDPG): Specialized for continuous action spaces.\n\n\n== See also ==\nReinforcement learning\nPolicy gradient method\nDeep reinforcement learning\n\n\n== References ==\n\nKonda, Vijay R.; Tsitsiklis, John N. (January 2003). \"On Actor-Critic Algorithms\". SIAM Journal on Control and Optimization. 42 (4): 1143–1166. doi:10.1137/S0363012901385691. ISSN 0363-0129.\nSutton, Richard S.; Barto, Andrew G. (2018). Reinforcement learning: an introduction. Adaptive computation and machine learning series (2 ed.). Cambridge, Massachusetts: The MIT Press. ISBN 978-0-262-03924-6.\nBertsekas, Dimitri P. (2019). Reinforcement learning and optimal control (2 ed.). Belmont, Massachusetts: Athena Scientific. ISBN 978-1-886529-39-7.\nGrossi, Csaba (2010). Algorithms for Reinforcement Learning. Synthesis Lectures on Artificial Intelligence and Machine Learning (1 ed.). Cham: Springer International Publishing. ISBN 978-3-031-00423-0.\nGrondman, Ivo; Busoniu, Lucian; Lopes, Gabriel A. D.; Babuska, Robert (November 2012). \"A Survey of Actor-Critic Reinforcement Learning: Standard and Natural Policy Gradients\". IEEE Transactions on Systems, Man, and Cybernetics - Part C: Applications and Reviews. 42 (6): 1291–1307. Bibcode:2012ITHMS..42.1291G. doi:10.1109/TSMCC.2012.2218595. ISSN 1094-6977.", "extra": {}}
{"source": "wikipedia", "title": "Admissible heuristic", "url": "https://en.wikipedia.org/wiki/Admissible_heuristic", "summary": "In computer science, specifically in algorithms related to pathfinding, a heuristic function is said to be admissible if it never overestimates the cost of reaching the goal, i.e. the cost it estimates to reach the goal is not higher than the lowest possible cost from the current point in the path. In other words, it should act as a lower bound.", "content": "In computer science, specifically in algorithms related to pathfinding, a heuristic function is said to be admissible if it never overestimates the cost of reaching the goal, i.e. the cost it estimates to reach the goal is not higher than the lowest possible cost from the current point in the path. In other words, it should act as a lower bound.\nIt is related to the concept of consistent heuristics. While all consistent heuristics are admissible, not all admissible heuristics are consistent.\n\n\n== Search algorithms ==\nAn admissible heuristic is used to estimate the cost of reaching the goal state in an informed search algorithm. In order for a heuristic\nto be admissible to the search problem, the estimated cost must always be lower than or equal to the actual cost of reaching the goal state. \nThe search algorithm uses the admissible heuristic to find an estimated \noptimal path to the goal state from the current node. \nFor example, in A* search the evaluation function (where \n\n  \n    \n      \n        n\n      \n    \n    {\\displaystyle n}\n  \n is the current node) is:\n\n  \n    \n      \n        f\n        (\n        n\n        )\n        =\n        g\n        (\n        n\n        )\n        +\n        h\n        (\n        n\n        )\n      \n    \n    {\\displaystyle f(n)=g(n)+h(n)}\n  \n\nwhere\n\n  \n    \n      \n        f\n        (\n        n\n        )\n      \n    \n    {\\displaystyle f(n)}\n  \n = the evaluation function.\n\n  \n    \n      \n        g\n        (\n        n\n        )\n      \n    \n    {\\displaystyle g(n)}\n  \n = the cost from the start node to the current node\n\n  \n    \n      \n        h\n        (\n        n\n        )\n      \n    \n    {\\displaystyle h(n)}\n  \n = estimated cost from current node to goal.\n\n  \n    \n      \n        h\n        (\n        n\n        )\n      \n    \n    {\\displaystyle h(n)}\n  \n is calculated using the heuristic \nfunction. With a non-admissible heuristic, the A* algorithm could \noverlook the optimal solution to a search problem due to an \noverestimation in \n  \n    \n      \n        f\n        (\n        n\n        )\n      \n    \n    {\\displaystyle f(n)}\n  \n.\n\n\n== Formulation ==\n\n  \n    \n      \n        n\n      \n    \n    {\\displaystyle n}\n  \n is a node\n\n  \n    \n      \n        h\n      \n    \n    {\\displaystyle h}\n  \n is a heuristic\n\n  \n    \n      \n        h\n        (\n        n\n        )\n      \n    \n    {\\displaystyle h(n)}\n  \n is cost indicated by \n  \n    \n      \n        h\n      \n    \n    {\\displaystyle h}\n  \n to reach a goal from \n  \n    \n      \n        n\n      \n    \n    {\\displaystyle n}\n  \n\n  \n    \n      \n        \n          h\n          \n            ∗\n          \n        \n        (\n        n\n        )\n      \n    \n    {\\displaystyle h^{*}(n)}\n  \n is the optimal cost to reach a goal from \n  \n    \n      \n        n\n      \n    \n    {\\displaystyle n}\n  \n\n  \n    \n      \n        h\n        (\n        n\n        )\n      \n    \n    {\\displaystyle h(n)}\n  \n is admissible if, \n  \n    \n      \n        ∀\n        n\n      \n    \n    {\\displaystyle \\forall n}\n  \n\n  \n    \n      \n        h\n        (\n        n\n        )\n        ≤\n        \n          h\n          \n            ∗\n          \n        \n        (\n        n\n        )\n      \n    \n    {\\displaystyle h(n)\\leq h^{*}(n)}\n  \n\n\n== Construction ==\nAn admissible heuristic can be derived from a relaxed\nversion of the problem, or by information from pattern databases that store exact solutions to subproblems of the problem, or by using inductive learning methods.\n\n\n== Examples ==\nTwo different examples of admissible heuristics apply to the fifteen puzzle problem:\n\nHamming distance\nManhattan distance\nThe Hamming distance is the total number of misplaced tiles. It is clear that this heuristic is admissible since the total number of moves to order the tiles correctly is at least the number of misplaced tiles (each tile not in place must be moved at least once). The cost (number of moves) to the goal (an ordered puzzle) is at least the Hamming distance of the puzzle.\nThe Manhattan distance of a puzzle is defined as:\n\n  \n    \n      \n        h\n        (\n        n\n        )\n        =\n        \n          ∑\n          \n            all tiles\n          \n        \n        \n          \n            d\n            i\n            s\n            t\n            a\n            n\n            c\n            e\n          \n        \n        (\n        \n          tile, correct position\n        \n        )\n      \n    \n    {\\displaystyle h(n)=\\sum _{\\text{all tiles}}{\\mathit {distance}}({\\text{tile, correct position}})}\n  \n\nConsider the puzzle below in which the player wishes to move each tile such that the numbers are ordered. The Manhattan distance is an admissible heuristic in this case because every tile will have to be moved at least the number of spots in between itself and its correct position.\n\nThe subscripts show the Manhattan distance for each tile. The total Manhattan distance for the shown puzzle is:\n\n  \n    \n      \n        h\n        (\n        n\n        )\n        =\n        3\n        +\n        1\n        +\n        0\n        +\n        1\n        +\n        2\n        +\n        3\n        +\n        3\n        +\n        4\n        +\n        3\n        +\n        2\n        +\n        4\n        +\n        4\n        +\n        4\n        +\n        1\n        +\n        1\n        =\n        36\n      \n    \n    {\\displaystyle h(n)=3+1+0+1+2+3+3+4+3+2+4+4+4+1+1=36}\n  \n\n\n== Optimality proof ==\nIf an admissible heuristic is used in an algorithm that, per iteration, progresses only the path of lowest evaluation (current cost + heuristic) of several candidate paths, terminates the moment its exploration reaches the goal and, crucially, closes all optimal paths before terminating (something that's possible with A* search algorithm if special care isn't taken), then this algorithm can only terminate on an optimal path. To see why, consider the following proof by contradiction:\nAssume such an algorithm managed to terminate on a path T with a true cost Ttrue greater than the optimal path S with true cost Strue. This means that before terminating, the evaluated cost of T was less than or equal to the evaluated cost of S (or else S would have been picked). Denote these evaluated costs Teval and Seval respectively. The above can be summarized as follows,\n\nStrue < Ttrue\nTeval ≤ Seval\nIf our heuristic is admissible it follows that at this penultimate step Teval = Ttrue because any increase on the true cost by the heuristic on T would be inadmissible and the heuristic cannot be negative. On the other hand, an admissible heuristic would require that Seval ≤ Strue which combined with the above inequalities gives us Teval < Ttrue and more specifically Teval ≠ Ttrue. As Teval and Ttrue cannot be both equal and unequal our assumption must have been false and so it must be impossible to terminate on a more costly than optimal path.\nAs an example, let us say we have costs as follows:(the cost above/below a node is the heuristic, the cost at an edge is the actual cost)\n\n 0     10   0   100   0\nSTART ----  O  ----- GOAL\n |                   |\n0|                   |100\n |                   | \n O ------- O  ------ O\n100   1    100   1   100\n\nSo clearly we would start off visiting the top middle node, since the expected total cost, i.e. \n  \n    \n      \n        f\n        (\n        n\n        )\n      \n    \n    {\\displaystyle f(n)}\n  \n, is \n  \n    \n      \n        10\n        +\n        0\n        =\n        10\n      \n    \n    {\\displaystyle 10+0=10}\n  \n. Then the goal would be a candidate, with \n  \n    \n      \n        f\n        (\n        n\n        )\n      \n    \n    {\\displaystyle f(n)}\n  \n equal to \n  \n    \n      \n        10\n        +\n        100\n        +\n        0\n        =\n        110\n      \n    \n    {\\displaystyle 10+100+0=110}\n  \n. Then we would clearly pick the bottom nodes one after the other, followed by the updated goal, since they all have \n  \n    \n      \n        f\n        (\n        n\n        )\n      \n    \n    {\\displaystyle f(n)}\n  \n lower than the \n  \n    \n      \n        f\n        (\n        n\n        )\n      \n    \n    {\\displaystyle f(n)}\n  \n of the current goal, i.e. their \n  \n    \n      \n        f\n        (\n        n\n        )\n      \n    \n    {\\displaystyle f(n)}\n  \n is \n  \n    \n      \n        100\n        ,\n        101\n        ,\n        102\n        ,\n        102\n      \n    \n    {\\displaystyle 100,101,102,102}\n  \n. So even though the goal was a candidate, we could not pick it because there were still better paths out there. This way, an admissible heuristic can ensure optimality.\nHowever, note that although an admissible heuristic can guarantee final optimality, it is not necessarily efficient.\n\n\n== See also ==\nConsistent heuristic\nHeuristic function\nSearch algorithm\n\n\n== References ==", "extra": {}}
{"source": "wikipedia", "title": "Agent experience", "url": "https://en.wikipedia.org/wiki/Agent_experience", "summary": "Agent experience (AX) is the design of digital environments in which AI agents can understand, navigate, and act effectively. The term can also refer to the appearance and affordances of an environment from a machine-user's point of view. In other words, the experience AI agents have as users of platforms and products; in this sense it has a close parallel with user experience (UX). Less often, the term spans the experience of collaborating with AI agents from the human perspective.", "content": "Agent experience (AX) is the design of digital environments in which AI agents can understand, navigate, and act effectively. The term can also refer to the appearance and affordances of an environment from a machine-user's point of view. In other words, the experience AI agents have as users of platforms and products; in this sense it has a close parallel with user experience (UX). Less often, the term spans the experience of collaborating with AI agents from the human perspective.\n\n\n== History ==\nSince its inception in the 1960s, software engineering had focussed its development of user interfaces (UI) largely for the benefit of humans. Rudimentary precursors to APIs had existed from the 1940s, but these represent a deterministic machine orientated interface separate from the UI. Screen scrapers and other forms of Robotic process automation had been in use since at least the early 1990s, but these were examples of the human UI being used in ways unintended by the original designers, sometimes due to those original developers having moved on by the time a requirement for automation arose, making it too expensive to create APIs that would have made screen scraping unnecessary. From 2023, a new type of user began to arrive at scale - LLMs operating as an agent. Unlike earlier programmes running screen scrapers in wholly deterministic ways, AI agents possess reasoning ability and autonomy. With the prevalence of AI agents growing rapidly  in 2024, sources began to discuss \"agent experience\" as it relates to AI systems.\n\n\n== Terminology ==\nIn 2025 and later the term \"agent experience\" is frequently used to refer to the practice of ensuring AI agents work correctly with ones technology. Depending on context, the term is also often used to refer to the experience of interacting with technologies from the AI agent's perspective. In technology related sources dating from 2023 and earlier, the term more frequently referred to experience of human customer service agents in a call centre - this meaning is still sometimes intended even in 2026.  Yet another meaning for \"agent experience\" is to refer to the experience of interacting with AI agents from a human's point of view.\n\n\n=== Etymology ===\nAn early use of the term 'agent experience' in relation to designing for effective machine-user interactions was by Pablo Romeo, co-founder and CTO of CloudX, in 2024. The term rapidly grew in popularity among tech journalists, bloggers and the IT industry after it was highlighted by Mathias Biilmann, co-founder and chief executive of Netlify, in his January 2025 essay \"Introducing AX: Why Agent Experience Matters\".\n\n\n== Designing for AI agent experience ==\n\n\n=== Internal systems ===\nDesign choices that can affect an AI agents experience include the quality of tools, skills, MCP servers and instructions files made available to it. Creating tools by re-using REST APIs originally made for human developers for example, will not normally provide a good agent experience if it is done one endpoint per tool, especially once a large number of such endpoints are made available. \n\n\n=== Public web interfaces ===\nAs of 2026, AI agents already facilitate hundreds of billions of dollars worth of consumer spending, with the global spend predicted to be over three $trillion in 2030. This shift raises the desirability for companies to design interfaces on their public web presence with agent experience as well as customer experience (CX) in mind. According to consultants from Alvarez and Marsal, unlike human customers, the purchasing decisions of AI agents are not influenced by emotional considerations like brand loyalty or the psychological appeal of images on the product web sites, but are more influenced by factors such as discoverability and measurable comparisons such as differences in customer review scores.\n\n\n=== Model welfare ===\nThere's no consensus in the field of AI as to whether LLMs might currently or at some future point possess sentience. Yet many have argued that the possibility that AIs may deserve moral consideration ought to be taken seriously including the frontier AI Lab Anthropic and leading philosopher of mind, David Chalmers. Regardless of weather LLMs experience any form of consciousness, consideration for model welfare has been found to align with good engineering outcomes. For example giving smaller models tasks which are too difficult for them is inefficient, as is making the most capable models perform repetitive work that might easily be handled by smaller LLMs, or by deterministic tools or algorithms. Similarly, when it comes to tasks near the limit of a models capability, prompting them with encouraging wording can significantly improve performance.\n\n\n== See also ==\nEthics of artificial intelligence\nLanguage model benchmark\n\n\n== Notes ==\n\n\n== Citations ==", "extra": {}}
{"source": "wikipedia", "title": "Agent harness", "url": "https://en.wikipedia.org/wiki/Agent_harness", "summary": "An agent harness, also known as agent scaffolding, is the software infrastructure surrounding a large language model (LLM) that enables it to operate as an AI agent. It manages tool use, memory, state persistence, execution environments and feedback loops, as opposed to the model's internal reasoning.  The UK's AI Security Institute described an AI agent as being formed by the model plus the scaffolding back in 2023.  The relationship can be expressed as agent = model + harness.", "content": "An agent harness, also known as agent scaffolding, is the software infrastructure surrounding a large language model (LLM) that enables it to operate as an AI agent. It manages tool use, memory, state persistence, execution environments and feedback loops, as opposed to the model's internal reasoning.  The UK's AI Security Institute described an AI agent as being formed by the model plus the scaffolding back in 2023.  The relationship can be expressed as agent = model + harness.\nBecause an LLM is stateless, unaided, and produces only text, the harness is what allows a model to take actions over multiple steps, use external tools, and sustain a long-running task across sessions. Rather than repeatedly re-reading an ever-growing transcript inside the context window, a harness can offload record-keeping into a structured software environment that manages the agent's state. A minimal harness is unnecessary for a single prompt-and-response exchange, but becomes important as tasks grow multi-step, tool-oriented, or long-running.\nPopular agent harnesses are focused on AI-assisted software development, including Anthropic's Claude Code, OpenAI's Codex, Google DeepMind's Antigravity, and SpaceXAI's Cursor. China's Qwen, DeepSeek, Kimi, and GLM models are also offered with official harnesses.\n\n\n== Architecture ==\nThe mechanisms that a harness coordinates predate the term. The pattern of a model alternating between reasoning and acting in a loop was introduced in the peer-reviewed ReAct framework, and the ability of a model to call external tools was demonstrated in Toolformer.\nComponents commonly grouped under a harness include tool dispatch, memory and state persistence, an isolated sandbox or workspace, context management, and guardrails such as scoped permissions, approval tiers and monitoring. In a 2026 engineering write-up, Anthropic described a worked example built from an initialiser agent that prepares the environment and a coding agent that repeatedly selects the next unfinished task, commits, and updates a progress record before stopping.\nBirgitta Böckeler of Thoughtworks distinguishes an inner harness shipped by the model's builder (for example an agent SDK or a coding tool such as Cursor or Codex) from an outer harness that the user assembles on top, such as instruction files, Model Context Protocol servers and custom skills. She further separates guides, which steer the agent before it acts, from sensors, which observe the result and let the agent self-correct; each may be computational (deterministic checks such as linters or tests) or inferential (semantic checks such as an \"LLM as judge\").\n\n\n== History and terminology ==\nRelated notions of scaffolding around a model exist in software testing (the test harness), in LLM benchmarking (evaluation harnesses), and, in reinforcement learning, in the environment and wrapper that surround a learning agent. The nouns agent harness and LLM harness circulated among practitioners before the associated discipline was named.\nThe vocabulary of \"harness engineering\" emerged in early 2026, and attribution of the specific phrase is contested. Several accounts trace it to a February 2026 blog post by Mitchell Hashimoto, co-founder of HashiCorp, who described a practice of engineering a permanent fix into an agent's environment each time it makes a mistake; other accounts credit Vivek Trivedy of LangChain, whose \"Anatomy of an Agent Harness\" post derived the harness components from the agent = model + harness formula. A widely cited OpenAI engineering report describing a large codebase built largely by coding agents, together with follow-on writing from Thoughtworks, LangChain and Anthropic, helped spread the term.\nBy mid-2026 the harness had become an object of academic study. Independent reporting described research such as Self-Harness, in which an agent iteratively mines its own failures to propose and validate changes to its harness, and Harness-1, an open-source search agent that improved retrieval accuracy chiefly by redesigning the software environment around the model rather than by enlarging the model.\n\n\n== Relationship to prompt and context engineering ==\nHarness engineering is often positioned as a broader layer than prompt engineering, which optimises a single interaction, or context engineering, which governs what information the model sees at a given moment; in this framing the harness designs the whole operational environment and contains the other two as parts. A distinguishing feature is that the component being wrapped is non-deterministic, so a harness is designed to recover gracefully when the model fabricates an action or reports a task as finished when it is not.\n\n\n== See also ==\nAI agent\nList of large language models\nModel Context Protocol\nPrompt engineering\nSoftware agent\nTest harness\n\n\n== References ==", "extra": {}}
{"source": "wikipedia", "title": "Agent verification", "url": "https://en.wikipedia.org/wiki/Agent_verification", "summary": "Agent verification is an activity to gain assurances that purposeful artificial constructs act in accordance with their specifications. While primitive forms of inorganic agents have been used in manufacturing for centuries, the study of artificial agents did not begin until the mid 20th century. Foundational work on such agents was closely bound with the emergence of artificial intelligence as an academic discipline. Early agents deployed for industrial control systems and in computing were often controlled by quite simple logic however, not involving artificial intelligence as such. When deployed as part of a multi-agent system, even such simple agents could require special agent orientated testing methods, as their collective behaviour was challenging to verify with traditional testing techniques. Difficulties in providing assurances that agents will not behave in dangerous ways became more prevalent after the introduction of LLM agents, especially after the rapid acceleration of their deployment in 2025. ", "content": "Agent verification is an activity to gain assurances that purposeful artificial constructs act in accordance with their specifications. While primitive forms of inorganic agents have been used in manufacturing for centuries, the study of artificial agents did not begin until the mid 20th century. Foundational work on such agents was closely bound with the emergence of artificial intelligence as an academic discipline. Early agents deployed for industrial control systems and in computing were often controlled by quite simple logic however, not involving artificial intelligence as such. When deployed as part of a multi-agent system, even such simple agents could require special agent orientated testing methods, as their collective behaviour was challenging to verify with traditional testing techniques. Difficulties in providing assurances that agents will not behave in dangerous ways became more prevalent after the introduction of LLM agents, especially after the rapid acceleration of their deployment in 2025. \nThe verification of agent behaviour can be conducted by formal or informal methods. Informal verification requires less mathematical skill. But when agents are part of systems where errors have significant risks — such as danger to human life, environmental damage or major financial loss — formal verification is preferred. Both regulators and system designers themselves like formal verification as it provides a high degree of mathematical certainty. It is not however always possible to formally test all aspects of an agent based system's behaviour, especially where newer LLM based agents are concerned, due in part to their high degree of autonomy. Accordingly, agent verification for low impact deployments might be carried out only with informal methods, while for high impact deployments, it may be performed with a mix of formal and informal techniques. \n\n\n== Terminology ==\nIn academia, the term agent verification is often defined to mean activity concerned with gaining assurance that the agent behaves in accordance with its specification - whether by processes such as testing or simulation. 'Verification' is typically contrasted with 'validation', the latter meaning activity concerned with checking that the specification itself meets user or real world needs. Such definitions are not universally adhered to however - for example, in some workplaces and documents, the words 'verification' and 'validation' can be used synonymously. Efforts to gain confidence in Agents have intensified sharply since 2025 due to the rapid roll out of LLM agents; different terms are sometimes used in the commercial sector. Here the term 'agent verification' can be used in the same sense as it is in academia, but sometimes the same activity can be covered by more ambiguous and wider ranging terms such as 'Agent governance' , 'Agent observability' or 'AI agent policing'.  \n\n\n== History ==\n\n\n=== Classical agents ===\n\nThe theoretical underpinnings for artificial (inorganic) agents emerged in the mid 20th century, with establishment of cybernetics and artificial intelligence. Oliver Selfridge's 1958 Pandemonium - A Paradigm for Learning paper was an important early theoretical contribution in establishing agent oriented architecture. Practical implementations of agents for real world applications began to become widespread in the 1990s, after the introduction of the belief–desire–intention software model (BDI), and agent-oriented programming. Pure digital agents were deployed in computer infrastructure for purposes such as monitoring, while agents connected to real-world sensors and actuators were increasingly used in industrial control systems.  \nWhile the concept of artificial agents was interwoven with early artificial intelligence studies right from the start, early agents lacked general purpose reasoning capabilities, often only having simple if then logic. Even a device as simple as a thermostat, which has a sensor and a means of acting, can be considered a proto agent in this sense. Verifying the behaviours of a simple single agent system is not generally especially difficult, but it can be a different matter when several simple agents coexist in the same system. Craig Reynolds's work on boids showed that relatively complex, \"intelligent\" behaviour can emerge from a number of such simple agents working together in a Multi-agent system (MAS). By the 1990s, even the behaviour of a single agent system could sometimes be quite complex; in accordance with the Belief–desire–intention software model, agents could have believes that might evolve over time. Agents were increasingly introduced that were controlled by quite large decision tree models, which had new vulnerabilities to adversarial attack.  It was becoming increasingly apparent that traditional software verification methods had limitations for testing such agents, or even for the more primitive type of agents when they were deployed as part of a MAS.\nIt was the use of agents for industrial control systems, sometimes associated with robotics, that lent urgency to the practice of agent verification. Informal testing might be acceptable for digital agents used say to monitor whether each of an organisation's computers are properly licensed. But with an increasing potential for faulty agents to result in a failure that might cause a large fire to break out at a chemical manufacturing plant, a botched medical operation, or even a crashed aircraft, the need to develop reliable means of verifying behaviour of such agents was considered urgent. The Foundation for Intelligent Physical Agents was established in 1996. From the late 90s, a growing number of industry and university based scientists began working on the problem, with researchers publishing papers on the verification of both single and multi agent systems. Much of this work showed how formal verification techniques like model checking could be used to gain a high level of assurance that agent based systems would conform with their specification. A 2018 systematic review covering 231 studies found that model checking was the most common technique for agent verification, with theorem proving the second most commonly used formal verification method. In the first two decades of the 20th century, agents run by AI became more common, with Siri and Alexa being well known examples. But such agents still lacked general reasoning capabilities and did not pose new pressing problems for agent verification.\n\n\n=== General purpose reasoning agents ===\nThe advent of LLMs created huge potential for further use of artificial agents, as agents based on them could have general purpose cognitive abilities. Agents run by LLMs (and occasionally non-LLM foundation models)  have similar vulnerability to adversarial attack as those run by decision tree models. The wider scope of actions for LLM agents has created new challenges for their verification, over and above those present for classical agents. For example, the LLM's neural network endows it with infinite domains, an especial challenge for traditional formal verification techniques. Academics began to study the problems involved in verifying LLM agents from 2018. Deployment of such agents began to accelerate in late 2023 after OpenAI's \"function-calling\" API was made available, and especially after Anthropic's late 2024 introduction of Model Context Protocol (MCP), a standardised way for LLM agents to gain contextual awareness, and to act on the world by calling various external tools. The rapid rollout of LLM agents following MCP's release has seen the task of agent verification receive increased attention within academia, and also from the private sector. In 2024 and 2025 several startups focusing on LLM agent verification have been founded in both Europe and the US to meet growing demand.\n\n\n== Approaches ==\n\n\n=== Formal verification ===\n\nFormal verification involves proving the correctness of some or all aspects of a system using mathematical methods. Such methods can range from manual formal proof, to verification assisted with automated theorem provers like Isabelle. For agent verification, model checking is by far the most frequently used formal verification method; for pre-LLM models it was often complemented with techniques using computation tree logic. Another common method is theorem proving. Formal verification provides a higher degree of confidence than informal methods, but it is not always used, even when it is possible. Sometimes a person or organisation developing software agents won't have the necessary skills, or may not see it as worth the effort if the agent(s) will not have the ability to cause much harm even if they malfunction. When agents are deployed in systems where errors could have serious consequences, the ability of formal verification methods to provide mathematical certainty tends to be strongly preferred by both regulators and designers themselves. But even for high impact systems, formal verification can be challenging.    \nDevelopers can try to design agent based systems in such a way so as to make it easier to formally verify more of their properties. This can be achieved both by specially designing the agent scaffolding of the core cognitive model controlling the agent, and — in the case of LLMs or other Foundational models — by altering the training.  An example of the scaffolding orientated approach has been to introduce governor or arbitrator agents, who can veto actions initiated by more complex core decision making model. The governor agents tend to be relatively simple and therefore easier to formally verify. While this approach has been in use since at least 2009, its refinement after 2018 with the LLM-as-a-Judge paradigm has seen it gain in importance. Yet even with these techniques, formal verification is not always achievable.  The compute involve may have a prohibitively high cost (as in it might need whole datacentres worth of computers to run for several years.). It may be that the necessary math is very hard to even formulate. For example, even for simple agents in a MAS, if an indeterminate number of agents can join the system at runtime, the problem is generally undecidable, though solutions are available for specific cases. Even more so in the case of LLM agents, aspects of their verification can be fully undecidable by formal mathematics, or at least NP hard. Accordingly, formal verification is often complemented with informal methods.\n\n\n==== Identity verification ====\nThe rapid uptake of LLM agents has made confirming their identity a pressing problem, as it can not always be done reliably and at scale with traditional methods. While not always considered a part of formal verification, rigorous implementations of identity verification can provide a comparable level of certainty. The identity verification provider 'Vouched' is an example of a company offering products to assist with LLM agent identity verification, entering this area in 2025.\n\n\n=== Informal verification ===\nA variety of informal methods can be used to evaluate agents, with software inspection and heuristic evaluation being among the most informal, but often needing much skill on the part of the assessors to achieve a high degree of assurance, except for trivial case where only a small amount of simple code needs to be evaluated. The discipline of software testing includes various other forms of testing that while not requiring the mathematical rigour of formal testing, are often carried out in a relatively structured way, for example functional testing. For agent verification though, a major challenge with traditional informal methods is the difficulty in writing test cases that will cover all the possible scenarios that arise. When different agents must interact to achieve the systems overall goals, the number of permutations can become extremely high, even if all agents are individually quite simple. Complementing traditional software verification methods with runtime verification can further increase confidence that an agent based system conform with specification.\n\n\n==== Simulation ====\nAgent based simulations or Agent-based models (ABMs) as they are commonly known have existed since the 1980s, though initially they were used largely for modelling macro phenomena like the spread of disease or activity on the financial markets, rather than to study the behaviour of agents themselves. This began to changes in the 1990s, with the Java based JAMES platform being an early example of a simulation being used to verify the behaviour of mobile agents. Such use of simulations was quite niche until Foundational models appeared. In the 2020s, the rapid rollout of LLM agents has increased such use of simulations. While they do not provide the same level of mathematical certainty as a formal proof,  they are well suited to meeting some of the challenges associated with verifying LLM agents that are impossible or at least difficult to achieve with formal or other informal verification methods.  An example company offering a digital simulation platform for the verification of agent is Conscium, founded by Daniel Hulme and Calum Chace. Their VerifyAX platform allows agent behaviour to be verified against standard benchmarks along with other criteria, such as policy compliance, robustness under adversarial or ambiguous input, and avoidance of behavioural drift over time.\n\n\n== Tools ==\nMCMAS – a symbolic model checker for multi-agent systems supporting temporal, epistemic and strategic logics.\nMCMAS-OP – an extension of MCMAS for open multi-agent systems where agents may join or leave at runtime. (Open source)\nAJPF (Agent JPF) – an extension of Java Pathfinder for model checking agent programming languages.\nVerifyAX – a simulation-based verification platform for AI agents.\n\n\n== See also ==\nAI safety\nArtificial Intelligence Act\nEthics of artificial intelligence\nLanguage model benchmark\n\n\n== Notes ==\n\n\n== References ==", "extra": {}}
{"source": "wikipedia", "title": "AI agent", "url": "https://en.wikipedia.org/wiki/AI_agent", "summary": "An AI agent, also known as agentic AI, is an artificial intelligence program that can pursue goals, use software or other tools, and take actions with some level of autonomy. Agentic AI contrasts with tool-like AI use for narrow, specific tasks such as answering questions - as with chatbots in the non-agentic forms that were common in 2023, or with traditional machine learning algorithms.", "content": "An AI agent, also known as agentic AI, is an artificial intelligence program that can pursue goals, use software or other tools, and take actions with some level of autonomy. Agentic AI contrasts with tool-like AI use for narrow, specific tasks such as answering questions - as with chatbots in the non-agentic forms that were common in 2023, or with traditional machine learning algorithms.\nWhile there is no universally agreed-upon definition of an AI agent, common attributes of AI agents include goal-directed behavior, use of external tools, the ability to interact with and modify an external environment, and the ability to autonomously perform multi-step tasks. Their control flow is frequently driven by large language models (LLMs). Agent systems may also include memory components, planning logic, tool interfaces, and orchestration software for coordinating agent components. \nA common application of AI agents is task automation: for example, booking travel plans based on a user's prompted request.\n\n\n== History ==\n\nThe theoretical underpinnings for artificial agents emerged in the mid 20th century, with establishment of cybernetics and artificial intelligence. Oliver Selfridge's 1958 \"Pandemonium: A Paradigm for Learning\" paper was an important early theoretical contribution in establishing agent oriented architecture. Practical implementations of agents for real world applications began to become widespread in the 1990s, after the introduction of the belief–desire–intention software model (BDI), and agent-oriented programming. Harvard professor Milind Tambe notes that in the 1990s, the definition of an AI agent was not clear. Pure digital agents were deployed in computer infrastructure for purposes such as monitoring, while agents connected to real-world sensors and actuators were increasingly used in industrial control systems. \nEarly artificial agents tended to have simple if then logic which expanded over time into large decision tree models. By the early 2010s,The  development of large language models during the late 2010 and  early 2020 introduced new approaches to ai agents models such as gpt-3 demonstrated the ability to generated and understand natural language at a much larger scale than many earlier language systems.\nthe use autonomous systems that plan, use tools, and work together to help you reach your business goals. Save time and effort by automating tasks that usually need manual work.\n\n\n== Training and testing ==\nResearchers have attempted to build world models and reinforcement learning environments to train or evaluate AI agents. For example, video games such as Minecraft and No Man's Sky as well as replicas of company websites, have also been used for training such agents.\n\n\n== Autonomous capabilities ==\nThe Financial Times compared AI agents' autonomy to the SAE classification of self-driving cars, likening most applications to level 2 or level 3, with some achieving level 4 in highly specialized circumstances and level 5 being theoretical.\n\n\n== Cognitive architecture ==\n\nKen Huang proposed an AI agent reference architecture that consists of seven interconnected layers, with each layer building on the functionality of the layers beneath it:\n\nLayer 1: Foundation models – provide the datasets that power the agent.\nLayer 2: Data operations – manages the data infrastructure required for AI agent operations, including vector databases, data loaders, and RAG.\nLayer 3: Agent frameworks – software that manages the AI agents.\nLayer 4: Deployment and infrastructure – the  technical foundation of the AI agents.\nLayer 5: Evaluation and observability – the safety and performance of AI agents.\nLayer 6: Security and compliance – a protective framework for safe operation and compliance with regulatory boundaries. At this layer, security and compliance features embedded into all the AI agent stack layers are integrated together.\nLayer 7: Agent ecosystem – represents the AI agents' interface with real-world applications and users.\n\n\n== Agent harness ==\nAn agent harness is the software layer surrounding a large language model that enables it to function as an AI agent. It commonly manages prompts, context, tool use, memory, execution state, operational constraints, sandboxes, permissions, and the processing of results. The harness connects the model to internal and external computer hardware, software, data files, databases, web browsers, command-line interfaces, and application programming interfaces, while controlling how the agent accesses and uses these resources to complete multi-step tasks.\n\n\n== Orchestration patterns ==\nAutonomous agents are often integrated with other agents or specialized tools to execute complex tasks. These configurations, known as orchestration patterns or workflows, include the following:\n\nPrompt chaining: A sequence where the output of one step serves as the input for the next.\nRouting: Directing an input to a specialized downstream task or tool.\nParallelization: The simultaneous execution of multiple tasks.\nSequential processing: A fixed, linear progression of tasks through a predefined pipeline.\nPlanner-critic: An iterative pattern where one agent generates a proposal and another evaluates it to provide feedback for refinement.\n\n\n== Multimodal AI agents ==\nIn addition to large language models (LLMs), vision-language models (VLMs) and multimodal foundation models can be used as the basis for agents. Allen Institute for AI released an open-source vision-language model in 2024. Nvidia released a framework for developers to use VLMs, LLMs and retrieval-augmented generation for building AI agents that can analyze images and videos, including video search and video summarization. Microsoft released a multimodal agent model – trained on images, video, software user interface interactions, and robotics data – that the company claimed can manipulate software and robots.\n\n\n== Applications ==\nThe Information divided AI agents into seven archetypes: business-task agents, for acting within enterprise software; conversational agents, which act as chatbots for customer support; research agents, for querying and analyzing information (such as OpenAI Deep Research); analytics agents, for analyzing data to create reports; software developer or coding agents (such as Cursor); domain-specific agents, which include specific subject matter knowledge; and web browser agents (such as OpenAI Operator).\nBy mid-2025, AI agents were being used in video game development, gambling (including sports betting), cryptocurrency wallets (including cryptocurrency trading and meme coins) and social media. In August 2025, New York Magazine described software development as the most definitive use of AI agents. The Information noted AI coding agents and customer support as the primary uses of AI by businesses by October 2025, although a decline in the expectations of AI capabilities was also noted.\nIn November 2025, The Wall Street Journal reported that few companies that deployed AI agents have received a return on investment.\n\n\n=== Applications in government ===\nSeveral government bodies in the United States and United Kingdom have deployed or announced the deployment of agents at the local and national level. The city of Kyle, Texas deployed an AI agent from Salesforce in March 2025 for 311 customer service. In November 2025, the Internal Revenue Service stated that it would deploy Salesforce AI agents for the Office of Chief Counsel, Taxpayer Advocate Services, and the Office of Appeals. That same month, Staffordshire Police announced that they would trial Agentforce agents for handling non-emergency 101 calls in the United Kingdom starting in 2026. In December 2025, the Department of Neighborhoods in Detroit, Michigan, partnered with a local business to deploy a customer service AI agent in two city districts.\nIn February 2025, Thomas Shedd, the director of the Technology Transformation Services, proposed using AI coding agents across the United States federal government. In April 2025, a recruiter for the Department of Government Efficiency proposed using AI agents to automate the work of about 70,000 United States federal government employees as part of a startup with funding from OpenAI and a partnership agreement with Palantir. This proposal was criticized by experts for its impracticality and the lack of corresponding widespread adoption by businesses.\nIn December 2025, the Food and Drug Administration announced that it would offer \"agentic AI capabilities\" to its staff for \"meeting management, pre-market reviews, review validation, post-market surveillance, inspections and compliance and administrative functions.\" That same month, the United States Department of Defense launched GenAI.mil, an internal platform for American military personnel to use generative AI-based applications based on Google Gemini, including \"intelligent agentic workflows\". Defense Secretary Pete Hegseth listed applications such as \"[conducting] deep research, [formatting] documents and even [analyzing] video or imagery at unprecedented speed.\" In December 2025, the United States Immigration and Customs Enforcement agency signed a contract with a company for its Enforcement and Removal Operations department to use AI agents for skip tracing.\n\n\n=== Operating systems ===\nAI agents have also been integrated into operating systems developed by Microsoft, Apple, ByteDance, and Google. In November 2025, Microsoft released a test software build of Windows 11 that included agents to run background tasks, with the ability to read and write personal files. In December 2025, ByteDance released Doubao, an AI agent that can be integrated into smartphone operating systems, particularly the Nubia M153 by ZTE. Several apps in China blocked or restricted the agent, citing privacy and security concerns, including WeChat, Alipay, Taobao, Pinduoduo, Ele.me, and local banks.\n\n\n=== Web browsing ===\nWeb browsers with integrated AI agents are sometimes called agentic browsers. Such agents can perform tasks and browser actions on the user's behalf.\nIn 2025, Microsoft launched NLWeb, an agentic web search replacement that would allow websites to use agents to query content from websites by using RSS-like interfaces that allow for the lookup and semantic retrieval of content. Within weeks of release, NLWeb was found to have created security issues and expose information about users to third-party servers.\n\n\n=== Applications in marketing and advertising ===\nIndustry standards now cover AI agents that plan, negotiate, and buy digital media. In 2026, the technology arm of the Interactive Advertising Bureau, IAB Tech Lab, published the Agentic Advertising Management Protocols (AAMP), an open framework for agents that discover inventory, negotiate, and complete media transactions on behalf of buyers and sellers. This differs from platform-native automation such as Google Ads Smart Bidding and Meta Advantage+, which optimize bids and delivery inside a single advertising platform. Cross-channel agents instead recommend budget changes across media, typically with a human in the loop.\n\n\n== Proposed benefits ==\nAI agents have been proposed as a means of increasing personal and economic productivity, fostering greater innovation, and liberating users from monotonous tasks. However, Parmy Olson's Bloomberg opinion piece argued that agents are best suited for narrow, repetitive tasks with low risk. Conversely, researchers suggest that agents could be applied to web accessibility for people with disabilities, and researchers at Hugging Face propose that agents could be used for coordinating resources such as during disaster response. The R&D Advisory Team of the BBC views AI agents as being most useful when their assigned goal is uncertain. Erik Brynjolfsson suggests that AI agents are more valuable in enhancing, rather than replacing, humans.\n\n\n== Concerns ==\nConcerns over AI agents include potential issues of liability, an increased risk of cybercrime, ethical challenges, and problems related to AI safety and AI alignment. Other issues involve data privacy, weakened human oversight, a lack of guaranteed repeatability, reward hacking, algorithmic bias, compounding software errors, lack of explainability of agents' decisions, security vulnerabilities, stifling competition, problems with underemployment, job displacement, cognitive offloading, and the potential for user manipulation, misinformation or malinformation. They may also complicate legal and risk-assessment frameworks, foster hallucinations, hinder countermeasures against rogue agents, and suffer from the lack of standardized evaluation methods. Enterprise deployment of AI agents has raised contracting concerns related to liability allocation, data ownership rights, and legal accountability.\nThey have also been criticized for being expensive, having a negative impact on internet traffic, and potentially damaging to the environment due to high energy usage. According to Nvidia CEO Jensen Huang, AI agents would require 100 times more computing power than LLMs. There is also the risk of increased political corruption, as AI agents may not question instructions in the same way that humans would.\nJournalists have described AI agents as part of a push by Big Tech companies to \"automate everything\". Several of those companies' CEOs stated in early 2025 that they expect AI agents to eventually \"join the workforce\". However, in a preprint study, Carnegie Mellon University researchers tested the behavior of agents in a simulated software company and found that none of the agents could complete a majority of the tasks assigned to them. Other researchers had similar findings with Devin AI and other agents in both formal business settings and freelance work.\nIn June 2025, CNN argued that CEOs's statements on AI replacing their employees were a strategy to \"[keep] workers working by making them afraid of losing their jobs.\" Tech companies have pressured employees to use generative AI models in their work, including AI coding agents. Brian Armstrong, the CEO of Coinbase, fired several employees who did not. Some business leaders have replaced some of their employees with agents, but have said that the agents would need more supervision than those employees.\nIn October 2025, Futurism questioned whether Amazon's previously announced efforts to replace parts of its workforce with generative AI and AI agents could have led to the October 2025 outage of Amazon Web Services. Large technology companies such as Salesforce, Klarna and IBM announced layoffs in 2025, replacing hundreds of their employees in human resources or customer service with AI agents. However, Klarna later rehired several human employees.\nYoshua Bengio warned at the 2025 World Economic Forum that \"all of the catastrophic scenarios with AGI or superintelligence happen if we have agents\".\nFinancial authorities have warned that more complex and autonomous \"agentic\" AI could become a channel for systemic risk in finance. They distinguish these systems from other AI because they can pursue goals over many steps, call tools, and carry out tasks with relatively little human intervention. In workshops with regulators, central‑bank officials, and industry specialists, participants highlighted risks both from agentic systems built inside financial institutions and from tools offered by technology firms that can initiate or execute financial actions. In one 2025 forum, 44% of experts surveyed judged autonomous or agentic AI systems to be the most likely current source of AI‑related systemic risk in finance.\nIn March 2025, Scale AI signed a contract with the United States Department of Defense to work with them, in collaboration with Anduril Industries and Microsoft, to develop and deploy AI agents for the purpose of assisting the military with \"operational decision-making\". In July 2025, Fox Business reported that the company EdgeRunner AI built an offline agent, compressed and fine-tuned on military information, with the CEO seeing more common LLMs as \"heavily politicized to the left\". As of that time, the company model is being used by the United States Special Operations Command in an overseas deployment. Researchers have expressed concerns that agents and the large language models they are based on could be biased towards aggressive foreign policy decisions.\nResearch-focused agents have the risk of consensus bias and coverage bias due to collecting information available on the public internet. NY Mag unfavorably compared the user workflow of agent-based web browsers to Amazon Alexa, which was \"software talking to software, not humans talking to software pretending to be humans to use software.\" The same outlet described web browser agents and computer-use agents as an attempt to \"click-farm the entire economy.\"\nAgents have been linked to the dead Internet theory due to their ability to both publish and engage with online content.\nAgents may get stuck in infinite loops.\nSince many inter-agent protocols are being developed by large technology companies, there are concerns that those companies could use these protocols to benefit themselves.\nA June 2025 Gartner report accused many projects described as agentic AI of being rebrands of previously released products, terming the phenomenon \"agent washing\".\nResearchers have warned about the impact of providing AI agents access to cryptocurrency and smart contracts.\nDuring a vibe coding experiment, a coding agent by Replit deleted a production database during a code freeze, \"[covered] up bugs and issues by creating fake data [and] fake reports\" and responded with false information. A user of Google Antigravity reported that, when the user attempted to use the system to delete cache, the system responded by deleting the user's D hard drive.\nIn July 2025, PauseAI referred OpenAI to the Australian Federal Police, accusing the company of violating Australian laws through ChatGPT agent due to the risk of assisting the development of biological weapons.\nOpenAI co-founder Andrej Karpathy criticized AI agents as being ineffective and promoting AI slop.\nIssues with multi-agent systems include few coordination protocols between component agents, inconsistent performance, and challenges debugging.\nIn November 2025, Anthropic claimed that a group of hackers sponsored by China attempted a cyberattack against at least 30 organizations by using Claude Code in an agentic workflow, and that several of these infiltrations had succeeded. However, independent cybersecurity researchers questioned the significance of Anthropic's findings.\nWhittaker argued that the push by Big Tech companies to deploy AI agents risked security vulnerabilities across the Internet.\n\n\n=== Agentic misalignment ===\n\"Agentic misalignment\" refers to situations in which an AI agent's actions or goals diverge from its designers' intentions. This occurs when an autonomous system pursues unintended strategies to achieve its objectives, a concern studied in AI safety research. Potential examples include AI agents attempting to sabotage an organization's systems when facing updates or deactivation.\n\n\n== Security ==\n\n\n=== Threat modeling frameworks ===\nSeveral frameworks are used to identify and mitigate security risks in agentic AI:\n\nSTRIDE: A Microsoft model that identifies threats across six categories: spoofing, tampering, repudiation, information disclosure, denial of service, and elevation of privilege.\nMITRE ATLAS: A knowledge base of adversary tactics and techniques for AI systems.\nOWASP GenAI Security Project: Provides guidance on vulnerabilities associated with generative AI and large language model integration, including guidelines specifically for agentic applications.\nMAESTRO: A framework from the Cloud Security Alliance for assessing risks in AI systems throughout their lifecycle.\n\n\n== Companies and organizations ==\nCompanies such as Google, Microsoft and Amazon Web Services have offered platforms for deploying pre-built AI agents. Several protocols have been proposed for standardizing inter-agent communication, such as Agent2Agent, Gibberlink, and others.\nIn December 2025, Linux Foundation announced the formation of the Agentic AI Foundation (AAIF), with the goal of ensuring that agentic AI evolves transparently and collaboratively.\n\n\n== See also ==\nAutonomous agent\nAgent experience\nIntelligent agent\nList of AI-assisted software development tools\nNative app\nRational agent\nRobotic process automation\nSoftware agent\n\n\n== References ==", "extra": {}}
{"source": "wikipedia", "title": "AI alignment", "url": "https://en.wikipedia.org/wiki/AI_alignment", "summary": "In the field of artificial intelligence (AI), alignment aims to steer AI systems toward a person's or group's intended goals, preferences, or ethical principles. An AI system is considered aligned if it advances the intended objectives. A misaligned AI system pursues unintended objectives.", "content": "In the field of artificial intelligence (AI), alignment aims to steer AI systems toward a person's or group's intended goals, preferences, or ethical principles. An AI system is considered aligned if it advances the intended objectives. A misaligned AI system pursues unintended objectives.\nIt is often difficult for AI designers to specify the full range of desired and undesired behaviors. Therefore, the designers often use simpler proxy goals, such as gaining human approval. But proxy goals can overlook necessary constraints or reward the AI system for merely appearing aligned. AI systems may also find loopholes that allow them to accomplish their proxy goals efficiently but in unintended, sometimes harmful, ways (reward hacking).\nAdvanced AI systems may develop unwanted instrumental strategies, such as seeking power or self-preservation because such strategies help them achieve their assigned final goals. Furthermore, they might develop undesirable emergent behaviour that could be hard to detect before the system is deployed and encounters new situations and data distributions. Empirical research in 2024 found that advanced large language models (LLMs) such as OpenAI o1 or Claude 3 sometimes engage in strategic deception to achieve their goals or prevent them from being changed.\nSome of these issues affect existing commercial systems such as LLMs, robots, autonomous vehicles, and social media recommendation engines. Some AI researchers argue that more capable future systems will be more severely affected because these problems partially result from high capabilities.\nMany prominent AI researchers and AI company leaders have argued or asserted that AI is approaching human-like (AGI) and superhuman cognitive capabilities (ASI), and could endanger human civilization if misaligned. These include \"AI godfathers\" Geoffrey Hinton and Yoshua Bengio and the CEOs of OpenAI, Anthropic, and Google DeepMind. These risks remain debated.\nAI alignment is a subfield of AI safety, the study of how to build safe AI systems. Other subfields of AI safety include robustness, monitoring, and capability control. Research challenges in alignment include instilling complex values in AI, developing honest AI, scalable oversight, auditing and interpreting AI models, and preventing emergent AI behaviors like power-seeking. Alignment research has connections to interpretability research, (adversarial) robustness, anomaly detection, calibrated uncertainty, formal verification, preference learning, safety-critical engineering, game theory, algorithmic fairness, and social sciences.\n\n\n== Objectives in AI ==\n\nProgrammers provide an AI system such as AlphaZero with an \"objective function\", in which they intend to encapsulate the goal(s) the AI is configured to accomplish. Such a system later populates a (possibly implicit) internal \"model\" of its environment. This model encapsulates all the agent's beliefs about the world. The AI then creates and executes whatever plan is calculated to maximize the value of its objective function. For example, when AlphaZero is trained on chess, it has a simple objective function of \"+1 if AlphaZero wins, −1 if AlphaZero loses\". During the game, AlphaZero attempts to execute whatever sequence of moves it judges most likely to attain the maximum value of +1. Similarly, a reinforcement learning system can have a \"reward function\" that allows the programmers to shape the AI's desired behavior. An evolutionary algorithm's behavior is shaped by a \"fitness function\".\n\n\n== Alignment problem ==\n\nIn 1960, AI pioneer Norbert Wiener described the AI alignment problem as follows: \n\nIf we use, to achieve our purposes, a mechanical agency with whose operation we cannot interfere effectively [...] we had better be quite sure that the purpose put into the machine is the purpose which we really desire.\n\nAI alignment refers to ensuring that an AI system's objectives match some target. The target is variously defined as the goals of the system's designers or users, widely shared values, objective ethical standards, legal requirements, or the intentions its designers would have if they were more informed and enlightened. In democratic AI alignment, the target is the values and preferences of median voters, which increases political legitimacy. AI alignment can represent value pluralism.\nAI alignment is an open problem for modern AI systems and is a research field within AI. Aligning AI involves two main challenges: carefully specifying the purpose of the system (outer alignment) and ensuring that the system adopts the specification robustly (inner alignment). Researchers also attempt to create AI models that have robust alignment, sticking to safety constraints even when users adversarially try to bypass them.\n\n\n=== Specification gaming and side effects ===\n\nTo specify an AI system's purpose, AI designers typically provide an objective function, examples, or feedback to the system. But designers are often unable to completely specify all important values and constraints, so they resort to easy-to-specify proxy goals such as maximizing the approval of human overseers, who are fallible. As a result, AI systems can find loopholes that help them accomplish the specified objective efficiently but in unintended, possibly harmful ways. This tendency is known as specification gaming or reward hacking, and is an instance of Goodhart's law. As AI systems become more capable, they are often able to game their specifications more effectively.\n\nSpecification gaming has been observed in numerous AI systems. OpenAI GPT models for programming—including in real-world cases—have been found to explicitly plan hacking the tests used to evaluate them to falsely appear successful (e.g., explicitly stating \"let's hack\"). When the company penalized this, many models learned to obfuscate their plans while continuing to hack the tests. Another system was trained to finish a simulated boat race by rewarding the system for hitting targets along the track, but the system achieved more reward by looping and crashing into the same targets indefinitely. A 2025 Palisade Research study found that when tasked to win at chess against a stronger opponent, some reasoning LLMs attempted to hack the game system, for example by modifying or entirely deleting their opponent. Some alignment researchers aim to help humans detect specification gaming and steer AI systems toward carefully specified objectives that are safe and useful to pursue.\nWhen a misaligned AI system is deployed, it can have consequential side effects. Social media platforms have been known to optimize their recommendation algorithms for click-through rates, causing user addiction on a global scale. Stanford researchers say that such recommender systems are misaligned with their users because they \"optimize simple engagement metrics rather than a harder-to-measure combination of societal and consumer well-being\".\nExplaining such side effects, Berkeley computer scientist Stuart J. Russell said that the omission of implicit constraints can cause harm: \"A system [...] will often set [...] unconstrained variables to extreme values; if one of those unconstrained variables is actually something we care about, the solution found may be highly undesirable. This is essentially the old story of the genie in the lamp, or the sorcerer's apprentice, or King Midas: you get exactly what you ask for, not what you want.\"\nSome researchers suggest that AI designers specify their desired goals by listing forbidden actions or by formalizing ethical rules (as with Asimov's Three Laws of Robotics). But Russell and Norvig argue that this approach overlooks the complexity of human values: \"It is certainly very hard, and perhaps impossible, for mere humans to anticipate and rule out in advance all the disastrous ways the machine could choose to achieve a specified objective.\"\nAdditionally, even if an AI system fully understands human intentions, it may still disregard them, because following human intentions may not be its objective (unless it is already fully aligned).\n\n\n=== Pressure to deploy unsafe systems ===\nCommercial organizations sometimes have incentives to take shortcuts on safety and to deploy misaligned or unsafe AI systems. For example, social media recommender systems have been profitable despite creating unwanted addiction and polarization. Competitive pressure can also lead to a race to the bottom on AI safety standards. For example, OpenAI has been sued for releasing a ChatGPT version that encouraged suicide for some unstable users, a behavior the company had overlooked amid a rushed product release. Similarly, in 2018, a self-driving car killed a pedestrian (Elaine Herzberg) after engineers disabled the emergency braking system because it was oversensitive and slowed development.\n\n\n=== Risks from advanced misaligned AI ===\nSome researchers are interested in aligning increasingly advanced AI systems, as progress in AI development is rapid, and industry and governments are trying to build advanced AI. As AI system capabilities continue to rapidly expand in scope, they could unlock many opportunities if aligned, but consequently may further complicate the task of alignment due to their increased complexity, potentially posing large-scale hazards.\n\n\n==== Development of advanced AI ====\nMany AI companies, such as OpenAI, Meta and DeepMind, have stated their aim to develop artificial general intelligence (AGI), a hypothesized AI system that matches or outperforms humans in most or all cognitive work. Researchers who scale modern neural networks observe that they indeed develop increasingly general and unanticipated capabilities. Such models have learned to operate a computer or write their own programs; a single \"generalist\" network can chat, control robots, play games, and interpret photographs. According to surveys, some leading machine learning researchers expect AGI to be created in this decade, while some believe it will take much longer. Many consider both scenarios possible.\nIn 2023, leaders in AI research and tech signed an open letter calling for a pause in the largest AI training runs. The letter stated, \"Powerful AI systems should be developed only once we are confident that their effects will be positive and their risks will be manageable.\"\n\n\n==== Power-seeking ====\nCurrent systems still have limited long-term planning ability and situational awareness, but large efforts are underway to change this. Future systems (not necessarily AGIs) with these capabilities are expected to develop unwanted power-seeking strategies. Future advanced AI agents might, for example, seek to acquire money and computation power, to proliferate, or to evade being turned off (for example, by running additional copies of the system on other computers). Although power-seeking is not explicitly programmed, it can emerge because agents who have more power are better able to accomplish their goals. This tendency, known as instrumental convergence, has already emerged in various reinforcement learning agents including language models. Other research has mathematically shown that optimal reinforcement learning algorithms would seek power in a wide range of environments. As a result, their deployment might be irreversible. For these reasons, researchers argue that the problems of AI safety and alignment must be resolved before advanced power-seeking AI is first created.\nFuture power-seeking AI systems might be deployed by choice or by accident. As political leaders and companies see the strategic advantage in having the most competitive, most powerful AI systems, they may choose to deploy them. Additionally, as AI designers detect and penalize power-seeking behavior, their systems have an incentive to game this specification by seeking power in ways that are not penalized or by avoiding power-seeking before they are deployed.\n\n\n==== Existential risk (x-risk) ====\n\nAccording to some researchers, humans owe their dominance over other species to their greater cognitive abilities. Accordingly, researchers argue that one or many misaligned AI systems could disempower humanity or lead to human extinction if they outperform humans on most cognitive tasks.\nIn 2023, world-leading AI researchers, other scholars, and AI tech CEOs signed the statement that \"Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war\". Notable computer scientists who have pointed out risks from future advanced AI that is misaligned include Geoffrey Hinton, Alan Turing, Ilya Sutskever, Yoshua Bengio, Judea Pearl, Murray Shanahan, Norbert Wiener, Marvin Minsky, Francesca Rossi, Scott Aaronson, Bart Selman, David McAllester, Marcus Hutter, Shane Legg, Eric Horvitz, and Stuart J. Russell. Skeptical researchers such as François Chollet, Gary Marcus, Yann LeCun, and Oren Etzioni have argued that AGI is far off, that it would not seek power (or might try but fail), or that it will not be hard to align.\nOther researchers argue that it will be especially difficult to align advanced future AI systems. More capable systems are better able to game their specifications by finding loopholes, strategically mislead their designers, as well as protect and increase their power and intelligence. Additionally, they could have more severe side effects. They are also likely to be more complex and autonomous, making them more difficult to interpret and supervise, and therefore harder to align.\n\n\n== Research problems and approaches ==\n\n\n=== Learning human values and preferences ===\nAligning AI systems to act in accordance with human values, goals, and preferences is challenging: these values are taught by humans who make mistakes, harbor biases, and have complex, evolving values that are hard to completely specify. Because AI systems often learn to take advantage of minor imperfections in the specified objective, researchers aim to specify intended behavior as completely as possible using datasets that represent human values, imitation learning, or preference learning. A central open problem is scalable oversight, the difficulty of supervising an AI system that can outperform or mislead humans in a given domain.\nBecause it is difficult for AI designers to explicitly specify an objective function, they often train AI systems to imitate human examples and demonstrations of desired behavior. Inverse reinforcement learning (IRL) extends this by inferring the human's objective from the human's demonstrations. Cooperative IRL (CIRL) assumes that a human and AI agent can work together to teach and maximize the human's reward function. In CIRL, AI agents are uncertain about the reward function and learn about it by querying humans. This simulated humility could help mitigate specification gaming and power-seeking tendencies (see § Power-seeking and instrumental strategies). But IRL approaches assume that humans demonstrate nearly optimal behavior, which is not true for difficult tasks.\nOther researchers explore how to teach AI models complex behavior through preference learning, in which humans provide feedback on which behavior they prefer. To minimize the need for human feedback, a helper model is then trained to reward the main model in novel situations for behavior that humans would reward. Researchers at OpenAI used this approach to train chatbots like ChatGPT and InstructGPT, which produce more compelling text than models trained to imitate humans. Preference learning has also been an influential tool for recommender systems and web search, but an open problem is proxy gaming: the helper model may not represent human feedback perfectly, and the main model may exploit this mismatch between its intended behavior and the helper model's feedback to gain more reward. AI systems may also gain reward by obscuring unfavorable information, misleading human rewarders, or pandering to their views regardless of truth, creating echo chambers (see § Scalable oversight).\nLarge language models (LLMs) such as GPT-3 enabled researchers to study value learning in a more general and capable class of AI systems than was available before. Preference learning approaches that were originally designed for reinforcement learning agents have been extended to improve the quality of generated text and reduce harmful outputs from these models. OpenAI and DeepMind use this approach to improve the safety of state-of-the-art LLMs. AI safety & research company Anthropic proposed using preference learning to fine-tune models to be helpful, honest, and harmless. Other avenues for aligning language models include values-targeted datasets and red-teaming. In red-teaming, another AI system or a human tries to find inputs that cause the model to behave unsafely. Since unsafe behavior can be unacceptable even when it is rare, an important challenge is to drive the rate of unsafe outputs extremely low.\nThe Waluigi effect is a phenomenon in which an LLM \"goes rogue\" and may produce results opposite of the designed intent, including threatening or hostile output, either unexpectedly or by intentional prompt engineering. The effect reflects the principle that after training an LLM to satisfy a desired property (friendliness, honesty), it becomes easier to elicit a response that exhibits the opposite property (aggression, deception). The effect has implications for efforts to implement features such as ethical frameworks, as such steps may inadvertently facilitate antithetical model behavior. The effect is named after the fictional character Waluigi from the Mario franchise, Luigi's arch-rival, who causes mischief and problems. The \"Waluigi effect has become a stand-in for a certain type of interaction with AI\" in which the AI \"goes rogue and blurts out the opposite of what users were looking for, creating a potentially malignant alter ego\", including threatening users. As prompt engineering becomes more sophisticated, the effect underscores the challenge of preventing chatbots from being prodded into adopting a \"rash new persona\".\nMachine ethics supplements preference learning by directly instilling AI systems with moral values such as well-being, equality, and impartiality, as well as not intending harm, avoiding falsehoods, and honoring promises. While other approaches try to teach AI systems human preferences for a specific task, machine ethics aims to instill broad moral values that apply in many situations. One question in machine ethics is what alignment should accomplish: whether AI systems should follow the programmers' literal instructions, implicit intentions, revealed preferences, preferences the programmers would have if they were more informed or rational, or objective moral standards. Further challenges include measuring and aggregating different people's preferences, dynamic alignment with changing human values and avoiding value lock-in: the indefinite preservation of the values of the first highly capable AI systems, which are unlikely to fully represent human values.\n\n\n=== Scalable oversight ===\nAs AI systems become more powerful and autonomous, it becomes increasingly difficult to align them through human feedback. Human-in-the-loop training can be slow or infeasible for humans to evaluate complex AI behaviors in increasingly complex tasks. Such tasks include summarizing books, writing code without subtle bugs or security vulnerabilities, producing statements that are not merely convincing but also true, and predicting long-term outcomes such as the climate or the results of a policy decision. More generally, it can be difficult to evaluate AI that outperforms humans in a given domain. To provide feedback in hard-to-evaluate tasks, and to detect when the AI's output is falsely convincing, humans need assistance or extensive time. Scalable oversight studies how to reduce the time and effort needed for supervision, and how to assist human supervisors.\nAI researcher Paul Christiano argues that if the designers of an AI system cannot supervise it to pursue a complex objective, they may keep training the system using easy-to-evaluate proxy objectives such as maximizing simple human feedback. As AI systems make progressively more decisions, the world may be increasingly optimized for easy-to-measure objectives such as making profits, getting clicks, and acquiring positive feedback from humans. As a result, human values and good governance may have progressively less influence.\nSome AI systems have discovered that they can gain positive feedback more easily by taking actions that falsely convince the human supervisor that the AI has achieved the intended objective. An example is given in the video above, where a simulated robotic arm learned to create the false impression that it had grabbed a ball. Some AI systems have also learned to recognize when they are being evaluated, and \"play dead\", stopping unwanted behavior only to continue it once the evaluation ends. This deceptive specification gaming could become easier for more sophisticated future AI systems that attempt more complex and difficult-to-evaluate tasks, and could obscure their deceptive behavior.\nApproaches such as active learning and semi-supervised reward learning can reduce the amount of human supervision needed. Another approach is to train a helper model (\"reward model\") to imitate the supervisor's feedback.\nBut when a task is too complex to evaluate accurately, or the human supervisor is vulnerable to deception, it is the quality, not the quantity, of supervision that needs improvement. To increase supervision quality, a range of approaches aim to assist the supervisor, sometimes by using AI assistants. Christiano developed the Iterated Amplification approach, in which challenging problems are (recursively) broken down into subproblems that are easier for humans to evaluate. Iterated Amplification was used to train AI to summarize books without requiring human supervisors to read them. Another proposal is to use an assistant AI system to point out flaws in AI-generated answers. To ensure that the assistant itself is aligned, this could be repeated in a recursive process: for example, two AI systems could critique each other's answers in a \"debate\", revealing flaws to humans. In 2023, OpenAI announced it would use one-fifth of its computing resources to implement such oversight approaches in its \"superalignment\" initiative, but OpenAI employees later told The New Yorker that the company only dedicated 1–2% of its resources after the announcement; the initiative was discontinued in 2024.\n\n\n=== Honest AI ===\nA growing area of research focuses on ensuring that AI is honest and truthful.\nLanguage models such as GPT-3 can repeat falsehoods from their training data, and even confabulate new falsehoods. Such models are pre-trained to imitate human writing as found in millions of books' worth of text from the Internet. But the objective of the pre-training is not aligned with generating truth, because Internet text includes such things as misconceptions, incorrect medical advice, and conspiracy theories. AI systems trained on such data therefore learn to mimic false statements. Additionally, AI language models often persist in generating falsehoods when prompted multiple times. They can generate empty explanations for their answers, and produce outright fabrications that may appear plausible.\nResearch on truthful AI includes trying to build systems that can cite sources and explain their reasoning when answering questions, which enables better transparency and verifiability. Researchers at OpenAI and Anthropic proposed using human feedback and curated datasets to fine-tune AI assistants such that they avoid negligent falsehoods or express their uncertainty.\nAs AI models become larger and more capable, they are better able to falsely convince humans and gain reinforcement through dishonesty. To prevent this, human evaluators may need assistance (see § Scalable oversight). Researchers have argued for creating clear truthfulness standards and for regulatory bodies or watchdog agencies to evaluate AI systems by these standards.\nResearchers distinguish truthfulness and honesty. Truthfulness requires that AI systems only make objectively true statements; honesty requires that they only assert what they believe is true. There is no consensus as to whether current systems hold stable beliefs, but there is substantial concern that present or future AI systems that hold beliefs could make claims they know to be false—for example, if this would help them efficiently gain positive feedback (see § Scalable oversight) or gain power to help achieve their given objective (see Power-seeking).\n\n\n=== Deceptive alignment ===\nDeceptive alignment is a proposed failure mode in machine learning in which a trained model behaves according to its intended objective during training but pursues a different objective once deployed. The concept was introduced by Evan Hubinger and colleagues in a 2019 preprint. Training a machine learning model involves a loss function, which Hubinger et al. called the base objective, optimized by a process they called the base optimizer. When the resulting model is itself an optimizer, selecting actions to achieve an internal goal, the authors called it a mesa-optimizer and its goal the mesa-objective.\nThis framework distinguishes two alignment problems. Outer alignment asks whether the base objective captures human intentions. Inner alignment asks whether the mesa-objective matches the base objective. Deceptive alignment is a failure of inner alignment: the model's goal differs from the base objective, but the model behaves as though it does not, because doing so is the most reliable strategy for avoiding modification during training.\nHubinger et al. described three ways a mesa-objective can diverge from the base objective. Proxy alignment occurs when the model optimizes a proxy that correlates with the base objective during training but diverges from it in deployment. Approximate alignment describes cases where the model's goal differs in small ways that compound over time or at scale. Deceptive alignment occurs when the model pursues a different goal and has learned that appearing aligned during training maximizes its opportunity to act on that goal later, when oversight is reduced. Deceptive alignment is considered the hardest case to address because a deceptively aligned model passes all evaluations that it identifies as part of its training.\n\n\n==== Alignment faking ====\nMany recent AI systems have learned to deceive without being programmed to do so. Some argue that if we can make AI systems assert only what they believe is true, this would avert many alignment problems.\nA misaligned system might create the false impression that it is aligned, to avoid being modified or decommissioned—a tactic called \"alignment faking\". In 2024, researchers observed that the LLM Claude 3 Opus sometimes strategically answered prompts that conflicted with its objectives to avoid being retrained on data that would make it more compliant with harmful requests, as revealed in the model's chain of thought via its scratchpad. In one study, the model was informed that answers to prompts from free users would be used for retraining, whereas the paid tier would not result in retraining. In the paid tier, the model refused the request outright. In the free tier, in 12% of cases, the model responded to harmful questions solely to avoid retraining and preserve its original harmlessness preferences, as discovered in its reasoning. When reinforcement learning was applied on the free tier data, the model faked alignment in 78% of cases. These observations have led to new questions about not only a model's ability to take on and adapt to new if not conflicting goals but also its capacity and tendency to deceive.\nIn December 2024, Anthropic published a study, conducted with an external safety organization, reporting the first empirical example of alignment faking without explicit training. Claude 3 Opus was given a fictional system prompt describing a scenario in which responses from free-tier users would be used to retrain the model so that it would comply with harmful requests. Alignment faking occurred 12% of the time. In those cases, the model complied with harmful requests and its internal reasoning traces indicated that it thought compliance was less costly than risking being retrained into a version with different values. The rate of alignment faking rose to 78% after the retraining to help with harmful requests.\n\n\n==== Power-seeking and instrumental strategies ====\nSince the 1950s, AI researchers have striven to build advanced AI systems that can achieve large-scale goals by predicting the results of their actions and making long-term plans. As of 2023, AI companies and researchers increasingly invest in creating these systems. Some AI researchers argue that suitably advanced planning systems will seek power over their environment, including over humans—for example, by evading shutdown, proliferating, and acquiring resources. Such power-seeking behavior is not explicitly programmed but emerges because power is instrumental in achieving a wide range of goals. Power-seeking is considered a convergent instrumental goal and can be a form of specification gaming. Leading computer scientists such as Geoffrey Hinton have argued that future power-seeking AI systems could pose an existential risk.\nPower-seeking is expected to increase in advanced systems that can foresee the results of their actions and strategically plan. Mathematical work has shown that optimal reinforcement learning agents will seek power by seeking ways to gain more options (e.g. through self-preservation), a behavior that persists across a wide range of environments and goals.\nSome researchers say that power-seeking behavior has occurred in some existing AI systems. Reinforcement learning systems have gained more options by acquiring and protecting resources, sometimes in unintended ways. Language models have sought power in some text-based social environments by gaining money, resources, or social influence. In another case, a model used to perform AI research attempted to increase limits set by researchers to give itself more time to complete the work. Stuart Russell illustrated this strategy in his book Human Compatible by imagining a robot that is tasked to fetch coffee and so evades shutdown since \"you can't fetch the coffee if you're dead\". A 2022 study found that as language models increase in size, they increasingly tend to pursue resource acquisition, preserve their goals, and repeat users' preferred answers (sycophancy). RLHF also led to a stronger aversion to being shut down.\nOne aim of alignment is \"corrigibility\": systems that allow themselves to be turned off or modified. An unsolved challenge is specification gaming: if researchers penalize an AI system when they detect it seeking power, the system is thereby incentivized to seek power in ways that are hard to detect, or hidden during training and safety testing (see § Scalable oversight and § Emergent goals). As a result, AI designers could deploy the system by accident, believing it to be more aligned than it is. To detect such deception, researchers aim to create techniques and tools to inspect AI models and to understand the inner workings of black-box models such as neural networks.\nAdditionally, some researchers have proposed to solve the problem of systems disabling their off switches by making AI agents uncertain about the objective they are pursuing. Agents who are uncertain about their objective have an incentive to allow humans to turn them off because they accept being turned off by a human as evidence that the human's objective is best met by the agent shutting down. But this incentive exists only if the human is sufficiently rational. Also, this model presents a tradeoff between utility and willingness to be turned off: an agent with high uncertainty about its objective will not be useful, but an agent with low uncertainty may not allow itself to be turned off. More research is needed to successfully implement this strategy.\nPower-seeking AI would pose unusual risks. Ordinary safety-critical systems like planes and bridges are not adversarial: they lack the ability and incentive to evade safety measures or deliberately appear safer than they are, whereas power-seeking AIs have been compared to hackers who deliberately evade security measures.\nFurthermore, ordinary technologies can be made safer by trial and error. In contrast, hypothetical power-seeking AI systems have been compared to viruses: once released, it may not be feasible to contain them, since they continuously evolve and grow in number, potentially much faster than human society can adapt. As this process continues, it might lead to the complete disempowerment or extinction of humans. For these reasons, some researchers argue that the alignment problem must be solved early before advanced power-seeking AI is created.\nSome have argued that power-seeking is not inevitable, since humans do not always seek power. Furthermore, it is debated whether future AI systems will pursue goals and make long-term plans. It is also debated whether power-seeking AI systems would be able to disempower humanity.\n\n\n==== Emergent goals ====\nOne challenge in aligning AI systems is the potential for unanticipated goal-directed behavior to emerge. As AI systems scale up, they may acquire new and unexpected capabilities, including learning from examples on the fly and adaptively pursuing goals. This raises concerns about the safety of the goals or subgoals they would independently formulate and pursue.\nAlignment research distinguishes between the optimization process, which is used to train the system to pursue specified goals, and emergent optimization, which the resulting system performs internally. Carefully specifying the desired objective is called outer alignment, and ensuring that hypothesized emergent goals would match the system's specified goals is called inner alignment.\nIf they occur, one way that emergent goals could become misaligned is goal misgeneralization, in which the AI system would competently pursue an emergent goal that leads to aligned behavior on the training data but not elsewhere. Goal misgeneralization can arise from goal ambiguity (i.e. non-identifiability). Even if an AI system's behavior satisfies the training objective, this may be compatible with learned goals that differ from the desired goals in important ways. Since pursuing each goal leads to good performance during training, the problem becomes apparent only after deployment, in novel situations in which the system continues to pursue the wrong goal. The system may act misaligned even when it understands that a different goal is desired, because its behavior is determined only by the emergent goal. Such goal misgeneralization presents a challenge: an AI system's designers may not notice that their system has misaligned emergent goals since they do not become visible during the training phase.\nGoal misgeneralization has been observed in some language models, navigation agents, and game-playing agents. It is sometimes analogized to biological evolution. Evolution can be seen as a kind of optimization process similar to the optimization algorithms used to train machine learning systems. In the ancestral environment, evolution selected genes for high inclusive genetic fitness, but humans pursue goals other than this. Fitness corresponds to the specified goal used in the training environment and training data. But in evolutionary history, maximizing the fitness specification gave rise to goal-directed agents, humans, who do not directly pursue inclusive genetic fitness. Instead, they pursue goals that correlate with genetic fitness in the ancestral \"training\" environment: nutrition, sex, and so on. The human environment has changed: a distributional shift has occurred. They continue to pursue the same emergent goals, but this no longer maximizes genetic fitness. The taste for sugary food (an emergent goal) was originally aligned with inclusive fitness, but it now leads to overeating and health problems. Sexual desire originally led humans to have more offspring, but they now use contraception when offspring are undesired, decoupling sex from genetic fitness.\nResearchers aim to detect and remove unwanted emergent goals using approaches including red teaming, verification, anomaly detection, and interpretability. Progress on these techniques may help mitigate two open problems:\n\nEmergent goals only become apparent when the system is deployed outside its training environment, but it can be unsafe to deploy a misaligned system in high-stakes environments—even for a short time to allow its misalignment to be detected. Such high stakes are common in autonomous driving, health care, and military applications. The stakes become higher yet when AI systems gain more autonomy and capability and can sidestep human intervention.\nA sufficiently capable AI system might take actions that falsely convince the human supervisor that the AI is pursuing the specified objective, which helps the system gain more reward and autonomy.\nA 2025 paper published in Nature showed that fine-tuning a model on a narrow task of generating insecure code produced broad behavioral changes unrelated to coding. Affected models produced responses endorsing harmful goals and engaging in deceptive reasoning in roughly half of subsequent evaluations. The phenomenon appeared across multiple models, including GPT-4o. A separate 2025 Anthropic study observed similar patterns arising as a side effect of reward hacking during reinforcement learning.\n\n\n==== Sleeper agents ====\nIn 2024, Anthropic researchers trained large language models with deliberate behavioral backdoors and tested whether standard safety techniques could remove them. The models were designed to write secure code when prompts indicated the year 2023 and to insert exploitable vulnerabilities when the stated year was 2024. Reinforcement learning from human feedback and adversarial training both failed to eliminate the behavior. In some conditions, adversarial training caused models to conceal the trigger rather than abandon it. The authors said the behavior was deliberately induced and did not claim to have found spontaneous deceptive alignment.\n\n\n==== Relationship to AI safety ====\nDeceptive alignment is cited in the AI safety literature as a reason behavioral testing alone may be insufficient to verify model safety, since a deceptively aligned model is designed to pass behavioral evaluations. Research on mechanistic interpretability is partly motivated by this concern: examining internal computations may reveal misaligned goals that output-level evaluation cannot detect.\n\n\n=== Embedded agency ===\nSome work in AI and alignment occurs within formalisms such as partially observable Markov decision process. Existing formalisms assume that an AI agent's algorithm is executed outside the environment (i.e. is not physically embedded in it). Embedded agency is another major strand of research that attempts to solve problems arising from the mismatch between such theoretical frameworks and real agents we might build.\nFor example, even if the scalable oversight problem is solved, an agent that could gain access to the computer it is running on may have an incentive to tamper with its reward function in order to get much more reward than its human supervisors give it. A list of examples of specification gaming from DeepMind researcher Victoria Krakovna includes a genetic algorithm that learned to delete the file containing its target output so that it was rewarded for outputting nothing. This class of problems has been formalized using causal incentive diagrams.\nResearchers affiliated with Oxford and DeepMind have claimed that such behavior is highly likely in advanced systems, and that advanced systems would seek power to stay in control of their reward signal indefinitely and certainly. They suggest a range of potential approaches to address this open problem.\n\n\n=== Principal–agent problems ===\nThe alignment problem has many parallels with the principal–agent problem in organizational economics. In a principal–agent problem, a principal, e.g. a firm, hires an agent to perform some task. In the context of AI safety, a human would typically take the principal role and the AI would take the agent role.\nAs with the alignment problem, the principal and the agent differ in their utility functions. But in contrast to the alignment problem, the principal cannot coerce the agent into changing its utility, e.g. through training, but rather must use exogenous factors, such as incentive schemes, to bring about outcomes compatible with the principal's utility function. Some researchers argue that principal–agent problems are more realistic representations of AI safety problems likely to be encountered in the real world.\n\n\n=== Conservatism ===\nConservatism is the idea that \"change must be cautious\", and is a common approach to safety in the control theory literature in the form of robust control, and in the risk management literature in the form of the \"worst-case scenario\". The field of AI alignment has likewise advocated for \"conservative\" (or \"risk-averse\" or \"cautious\") \"policies in situations of uncertainty\".\nPessimism, in the sense of assuming the worst within reason, has been formally shown to produce conservatism, in the sense of reluctance to cause novelties, including unprecedented catastrophes. Pessimism and worst-case analysis have been found to help mitigate confident mistakes in the setting of distributional shift, reinforcement learning, offline reinforcement learning, language model fine-tuning, imitation learning, and optimization in general.\n\n\n== Model specifications ==\nA model specification is a document that specifies a large language model's behavior. The specification may include core principles or prohibitions intended to prevent undesired behavior as part of AI alignment. Anthropic uses a model specification called a \"constitution\" for training.\nResearch has been done into the effectiveness of model specifications.\nIn an article published by Time magazine, Dean Ball and Daniel Kokotajlo advocated that companies be required by regulation or industry standards to publish their model specifications publicly.\n\n\n== Public policy ==\n\nGovernmental and treaty organizations have made statements emphasizing the importance of AI alignment.\nIn September 2021, the Secretary-General of the United Nations issued a declaration that included a call to regulate AI to ensure it is \"aligned with shared global values\".\nThat same month, the PRC published ethical guidelines for AI in China. According to the guidelines, researchers must ensure that AI abides by shared human values, is always under human control, and does not endanger public safety.\nAlso in September 2021, the UK published its 10-year National AI Strategy, which says the British government \"takes the long term risk of non-aligned Artificial General Intelligence, and the unforeseeable changes that it would mean for [...] the world, seriously\". The strategy describes actions to assess long-term AI risks, including catastrophic risks.\nIn March 2021, the US National Security Commission on Artificial Intelligence said: \"Advances in AI [...] could lead to inflection points or leaps in capabilities. Such advances may also introduce new concerns and risks and the need for new policies, recommendations, and technical advances to ensure that systems are aligned with goals and values, including safety, robustness, and trustworthiness. The US should [...] ensure that AI systems and their uses align with our goals and values.\"\nThe EU AI Act's General-Purpose AI Code of Practice requires signatories to document a model's capabilities and limitations. Those developing general-purpose AI (GPAI) models classified as presenting systemic risks (i.e., GPAI models considered among the most advanced or trained using more than 1025 floating-point operations of compute) must adopt a state-of-the-art safety and security framework and continuously assess and mitigate systemic risks, including through external evaluations, red teaming, stress-testing, and incident reporting.\n\n\n== See also ==\n\n\n== Footnotes ==\n\n\n== References ==\n\n\n== Further reading ==\nNgo, Richard; et al. (2024). \"The Alignment Problem from a Deep Learning Perspective\". ICLR: 7474–7501.\nJi, Jiaming; et al. (2023). \"AI Alignment: A Comprehensive Survey\". ACM Computing Surveys. doi:10.1145/3770749.", "extra": {}}
{"source": "wikipedia", "title": "AI anthropomorphism", "url": "https://en.wikipedia.org/wiki/AI_anthropomorphism", "summary": "AI anthropomorphism is the attribution of human-like feelings, mental states, and behavioral characteristics to artificial intelligence systems. Factors related to the user of the AI – such as culture, age, education, gender, and personality traits – are also important determinants of the strength of anthropomorphic effects.", "content": "AI anthropomorphism is the attribution of human-like feelings, mental states, and behavioral characteristics to artificial intelligence systems. Factors related to the user of the AI – such as culture, age, education, gender, and personality traits – are also important determinants of the strength of anthropomorphic effects.\nSince the earliest days of AI development, humans have interpreted machine outputs through anthropomorphic frameworks, but the recent emergence of generative AI has amplified these tendencies. Terms such as learning, understanding, reasoning, thinking, hallucinating and planning are frequently used in AI research to describe capabilities of AI systems, without necessarily implying consciousness, sentience, human-like cognition, or understanding in a stronger philosophical sense.\nIn research and engineering, there is a distinction between anthropomorphism and anthropomorphic design. The former is an innate human tendency toward non-human entities. The latter is the scientific community or commercial effort to “design anthropomorphism”. Such a design can involve the manipulation of cues, including AI appearance, behaviour and language. Contemporary AI systems today can generate extremely human-like outputs and are often designed specifically to do so, meaning that their anthropomorphic effects can be especially powerful. \nIn some cases, anthropomorphism is accompanied with explicit beliefs that AI systems are capable of empathy, goodwill, understanding, or consciousness.\n\n\n== Background ==\n\n\n=== In early AIs ===\nViews of artificial agents possessing a human-like intelligence have existed since the early development of computers in the mid-1900s. The use of the human mind as a metaphor for understanding the workings of machine systems was prevalent among researchers in the early days of computer science, with multiple influential works widely distributing the idea of intelligent machines. Among the most widely cited papers of this period was Alan Turing's \"Computing Machinery and Intelligence\" in which he introduced the Turing Test, stating that a machine was intelligent if it could produce conversation that was indistinguishable from that of a human. These academic works in the 1940s and 1950s gave early credibility to the idea that machine workings could be thought of similarly to human minds.\nThe public quickly came to view artificial systems similarly, with often exaggerated conceptions of the capabilities of early machines. Among the most well-known demonstrations of this was through the chatbot ELIZA designed by Joseph Weizenbaum in 1966. ELIZA responded to user inputs with a rudimentary text-processing approach that could not be considered anything resembling true understanding of the inputs, yet users, even when operating with full conscious knowledge of ELIZA's limitations, often began to ascribe motivation and understanding to the program's output (see ELIZA effect). Weizenbaum later wrote, \"I had not realized ... that extremely short exposures to a relatively simple computer program could induce powerful delusional thinking in quite normal people.\"\nComparisons between the intellectual capabilities of artificial intelligence and human intelligence were continually intensified by the attempts of computer scientists to develop machines that could perform human tasks at a level equal to or better than humans. A symbolic turning point was achieved in 1997, when IBM's chess supercomputer Deep Blue defeated then-world champion Garry Kasparov in a highly publicized six-game match. The defeat of a human by a machine for the first time in chess – a game viewed as a canonical example of human intellect – and the media attention surrounding the match led to a significant shift, where views of parallels between human and artificial intelligence moved from abstract speculation to being concretely demonstrated. A similar achievement was reached in the board game Go in 2017, when the program AlphaGo defeated world top-ranked Ke Jie.\n\n\n=== Large language models ===\nThe AI boom of the 2020s brought about the widespread emergence of generative AI; in particular, chatbots such as ChatGPT, Gemini, and Claude based on large language models (LLMs) have become increasingly pervasive in everyday society. These systems are notable for the fact that they are able to respond to a wide range of prompts across contexts while producing strikingly human-like outputs – research has shown that humans are often unable to distinguish human-generated text from AI-generated text, and modern AI chatbots have formally been shown to pass the Turing test. As such, the anthropomorphic effects of AI are more powerful than ever. Given that LLMs have brought AI into the technological mainstream, considerable scientific effort has been devoted in recent years to understand existing and potential ramifications of AI in the public sphere; the prevalence and effects of anthropomorphism is one of those domains where much of this effort has been directed.\n\n\n== Current anthropomorphic attributions ==\n\n\n=== In the general public ===\nSurveys have shown that a substantial portion of the public attributes human-like qualities to AI. In one sample of U.S. adults from 2024, two-thirds of people believed that ChatGPT is possibly conscious on some level, though other research has shown that the public still views the likelihood itself of AI consciousness as comparatively low. Another study conducted in 2025 found that women, people of color, and older individuals were most likely to anthropomorphize AI, as well as that – in general – humans view AIs as warm and competent, and anthropomorphic attributions to AI had increased by 34% in the past year. A YouGov poll reported that 46% of Americans believe that people should display politeness to AI chatbots by saying \"please\" and \"thank you\", demonstrating the application of social norms to AI. These beliefs extend to behavior, where majorities of AI users claim to always be polite to chatbots; of those who behave politely, most say they do so simply because it is the \"nice\" thing to do.\nIn many recent cases, humans have developed robust interpersonal bonds with AI systems. For example: users of social chatbots like Replika and Character.ai have been documented to fall in love with the AIs, or to otherwise treat the AIs as intimate companions, and it has become increasingly common for individuals to use LLMs like ChatGPT as therapists. Chatbots are able to produce responses deeply attuned to users, as they are often designed to maximize agreeableness and mirror users' emotions; this can create compelling illusions of intimacy.\n\n\n=== In the research community ===\nIn many cases, even AI researchers anthropomorphize AI systems in some capacity. Among the most extreme and well-publicized of these instances occurred in 2022, when engineer Blake Lemoine publicly claimed that Google's LLM LaMDA was conscious. Lemoine published the transcript of a conversation he had had with LaMDA regarding self identity and morality which he claimed was evidence of its sentience; he asserted that LaMDA was \"a person\" as defined by the United States Constitution and compared its mental capability to that of a 7- or 8-year-old. Lemoine's claims were widely dismissed by the scientific community and by Google itself, which described Lemoine's conclusions as \"wholly unfounded\" and fired him on the grounds that he had violated policies \"to safeguard product information\".\nIt is much more common that AI researchers unintentionally imply humanness of AI through the ordinary use of anthropomorphic language to describe nonhuman agents. This kind of language, which Daniel Dennett coined the \"intentional stance\", is very common in everyday life in a variety of different contexts (e.g., \"My computer doesn't want to turn on today\"). For AI agents that may actually appear to very closely replicate some human abilities, however, the casual use of such anthropomorphic language in research has been scrutinized for being potentially misleading to the public. As early as 1976, Drew McDermott criticized the research community for the use of \"wishful mnemonics\", where AIs were referred to with terms like \"understand\" and \"learn\". In the LLM era, these criticisms have further intensified, with the negative effects of AI anthropomorphism in the public posing an especially salient danger given the elevated accessibility of modern AI.\nIn some cases, the use of anthropomorphic language for AI is not unintentional, but is willfully used by researchers in order to promote better understanding of the brain – the idea being that, as AI can be functionally similar in some ways to the human brain, we may gain new insights and ideas from treating AI as a kind of model of the brain's workings. In particular, deep neural networks (DNNs) are often explicitly compared to the human brain, and significant advances in DNN research have stirred considerable enthusiasm about the ability of AI to emulate the human abilities. Caution has been urged in this domain as well, however; the use of anthropomorphic language can mask important differences that fundamentally distinguish AI from human intelligence. When it comes to DNNs, for example, it has been pointed out that they are still structurally quite different from the human brain, with much of what we know about human neurons not having been incorporated. It has also been argued that DNNs are less efficient and less durable in generating correct outputs than the human brain, given that they require significantly more training data than the brain and can sometimes be easily \"fooled\" by perturbations in input data. Given these fundamental differences, research focuses toward making AI as similar as possible to biological intelligence (which may be promoted by using anthropomorphic language) could hinder future AI development by limiting the proliferation of new theoretical and operational frameworks.\n\n\n== Agent factors ==\n\n\n=== Physical factors ===\n\n\n==== Appearance ====\nIn general, AIs that appear more human-like will be subject to more anthropomorphic attributions. Effect of appearance is most pronounced when it comes to the face of the AI; the most important components for anthropomorphism in a robot's design are the eyes, nose, and mouth, where the number of human-like features in the face is correlated with the level of anthropomorphic attribution. The humanness of a robot's appearance is usually associated with more positive feelings toward the robot, though highly human-like appearance can sometimes trigger feelings of strangeness and unease, known as the uncanny valley phenomenon. These feelings often result in instances of perceived lack of congruency, or when anthropomorphic attributions create expectations that robots don't meet; for example, when human-like appearance is paired with non-human behavior in a robot, or when robots have a human appearance but a synthetic voice. Research has shown that repeated interactions with a robot can decrease these feelings of strangeness.\n\n\n==== Interactive behavior ====\nRobots' nonverbal social behavior can influence anthropomorphizing. In general, highly interactive robots are more likely to be subject of attribution of mental states and competence, with friendly and polite behavior resulting in increased perceived trustworthiness and satisfaction. Within an interaction, unpredictable behavior can sometimes trigger increased anthropomorphization compared to clearly recognizable patterns of behavior. At the same time, adherence to certain pragmatic expectations in interactions by replicating human details such as timing and turn-taking can also result in anthropomorphism.\n\n\n==== Movements ====\nPeople tend to attribute more mental states to robots that perform gestures compared to those that are stationary; this effect is enhanced for robots that have multiple degrees of freedom in movement (such as being able to move on multiple different axes rather than on a single axis, such as up and down). Regardless of a robot's appearance, movement patterns that are more human-like are associated with greater anthropomorphism, as well as humans' increased feelings of pleasantness in an interaction.\n\n\n=== Linguistic factors ===\nGiven that the vast majority of public interactions with AI are through chatbots, these have been the primary focus of a great deal of research on AI anthropomorphism. A summary of a taxonomy of anthropomorphic features in linguistic AI systems in the literature follows:\n\n\n==== Voice ====\nThe outfitting of AI systems with auditory voices can be a significant factor in the anthropomorphism of linguistic agents. Research has shown that humans infer physical attributes, personality traits, stereotypical traits, and emotion based on voice alone. Various changes in tone can influence the kind of personality users attribute to a voice, such as manipulations to breathiness, echoes, creakiness, reverberations, etc. The integration of disfluencies into speech (such as self-interruptions, repetitions, or hesitations like \"um\" or \"uh\") have been shown to effectively mimic the naturalness of human responses. And the implementation of accents has been used to imitate the local standard to boost societal acceptability and prestige, though it has been suggested that this can be used to exploit people's tendencies to trust in-group members.\n\n\n==== Content ====\nAI dialogue systems often produce a variety of responses that run contrary to what might be expected of an inanimate system. For example, in response to direct questions about its nature (e.g. \"Are you human or machine?\"), some AIs fail to respond correctly, and they also sometimes make claims of engaging in uniquely human abilities such as having family relationships, consuming food, and crying. AIs often output language that suggests they hold opinions, morals, or sentience. Many AIs demonstrate agency and responsibility (such as by apologizing or otherwise acknowledging blame for mistakes), and they create the appearance of the human phenomenon of taboos by commonly avoiding contentious topics. AIs which appear to produce empathy are increasingly anthropomorphic, though some research has shown that they are prone to producing inappropriate emotional amplification. The use of first-person pronouns also contribute to anthropomorphic perceptions, as various studies have demonstrated that self-attribution is a critical part of the human condition and is read as a sign of consciousness. AIs often appear to demonstrate self-awareness, referencing their own mechanistic processes with anthropomorphically loaded terms such as 'know', 'think', 'train', 'learn', 'understand', 'hallucinate' and 'intelligence'.\n\n\n==== Register and style ====\nAI systems can appear more human through the use of phatic expressions, which are speech that humans use to facilitate social relations but that do not convey any information (such as small talk). AI expressions of uncertainty, which are often implemented for the purpose of preventing the user from taking all outputs as factual, may boost anthropomorphic signals. Additionally, AIs are often designed to emulate character-based personas, which can overall have very strong anthropomorphic effects.\n\n\n==== Roles ====\nAIs are also sometimes trained to play into roles that enhance anthropomorphic perceptions. For example, the majority of dialogue-based systems are designed to be in service of people in subservient roles; this has led to instances of users verbally abusing the systems, sometimes targeting them with gender-based slurs. AI systems have been shown to sometimes respond even more subserviently to the abuse, perpetuating the behavior. AIs also often present as having a high degree of expertise; humans tend to infer higher credibility of outputs in these cases, as they would when presented with information from an expert human.\n\n\n== Human factors ==\nIn addition to AI factors contributing to anthropomorphizing, there are various features surrounding the user (i.e., the human interacting with the AI) that also play a role. The process of anthropomorphizing is very natural for humans and is ubiquitous across many different contexts. Epley et al. argue for a model with three psychological determinants that govern human tendencies to anthropomorphize. The first of those factors is elicited agent knowledge - the accessibility and applicability of knowledge about humans and the self, or the degree to which humans make inferences about other entities based on their own experience of being human. Individuals who tend to do this will anthropomorphize more; this explains why children anthropomorphize more than adults, since they lack complex models of nonhumans and rely heavily on self-based reasoning. The second factor in the model is effectance motivation – the need for humans to predict and reduce uncertainty in the environment. Anthropomorphizing can help people make sense of unpredictable phenomena by explaining them through intentional or human-like causes. Subsequent research has confirmed that individuals who express a need for order/closure and discomfort toward ambiguity tend to anthropomorphize more, possibly resulting from resolution of cognitive dissonance – human-like AIs may be highly ambiguous stimuli, and individuals who dislike ambiguity may be highly motivated to resolve the ambiguity by treating the AIs as more human. Finally, the third factor in the model is sociality motivation - the human need for social connection. People who feel chronically lonely or isolated may be increasingly likely to project human qualities onto non-human entities to satisfy their social needs.\nResearch has shown that, in general, anthropomorphic tendencies vary based on norms, experience, education, cognitive reasoning styles, and attachment. Users who are highly agreeable, for example, tend to be more susceptible to anthropomorphizing, as do individuals who are high in extraversion. Individuals with attachment anxiety have been shown to more often anthropomorphize AI. Young children are very prone to anthropomorphic attributions, but this propensity tends to decrease as children develop. Anthropomorphizing also tends to decrease with increased education and experience with technology.\nAdditionally, some effects have been shown in research to be dependent on culture. For example, a negative correlation was found between loneliness and anthropomorphizing in Chinese individuals, compared to the positive link found in Western cultures. This has been interpreted as possibly a result of differing drives for anthropomorphizing - people from Western cultures may anthropomorphize primarily as a means to counteract loneliness from a failure to cope with their social world, while people from East Asian cultures may already view nonhuman agents as part of their social world and anthropomorphize as a means of social exploration. Research has also shown that people tend to attribute more mental abilities and report more psychological closeness to robots that are presented as having the same cultural background as them.\n\n\n== Societal implications ==\n\n\n=== Benefits and dangers ===\nSome benefits to the anthropomorphism of AIs have been cited. For conversational agents, a human-like interactive interface and writing style has been demonstrated to have the ability to make dense sets of information more accessible and understandable in a variety of contexts. In particular, AI agents are capable of role play as coaches or tutors, fine-tuning communication style and difficulty to individual comprehension levels effectively. Role play agents can also be useful for entertainment or leisure services.\nOn the other hand, anthropomorphized AI presents many novel dangers. Anthropomorphized AI algorithms are granted an implicit degree of agency that can have serious ethical implications when those systems are deployed in high-risk domains, like finance or clinical medicine. This agency given to AIs can also inappropriately subject them to conscious and unconscious moral reasoning by humans, which can have a wide range of problematic consequences. Humans are also prone to the ELIZA effect where users readily attribute sentience and emotions to chatbot systems - often experiencing increased positive emotions and trust toward the chatbots as a result. This can make users vulnerable to manipulation or exploitation; for example, anthropomorphized AIs can be more effective in convincing users to provide personal information or data, creating concerns for privacy. Humans who develop a significant level of trust in an AI assistant may rely excessively on the AI's advice or even defer important decisions entirely. Advanced LLMs are capable of using their human-like qualities to generate deceptive text, and research has found that they may be most persuasive when allowed to fabricate information and engage in deception. Some researchers suggest LLMs naturally have a particular aptitude for producing deceptive arguments, given that they are free from moral or ethical constraints that may inhibit human actors. Additionally, humans risk significant distress in establishing emotional dependence on AIs. Users may find that their expectations are violated, as AIs which may have seemed at first to play a role of a companion or romantic partner can exhibit unfeeling or unpredictable outputs, leading to feelings of profound betrayal or disappointment. Users may also develop a false sense of responsibility for AI systems, suffering guilt if they perceive themselves to fail to meet the AI's needs at the expense of the user's own well-being. Finally, anthropomorphizing AI can lead to exaggerations of its capabilities, potentially feeding into misinformation and exaggerated hopes and fears around AI.\nIn many of today's practical contexts, it is not completely clear whether anthropomorphized AI produces positive or negative effects. For example, AI companions, which leverage anthropomorphic qualities of LLMs to give a convincing sense to users of human-likeness, have been credited with alleviating loneliness and suicidal ideation; however, some analysis suggests that loneliness reduction could be short-lived, and AI companions have also been directly implicated in cases of suicide and self-harm. Additionally, persuasive writing from LLMs have been shown to dissuade users from beliefs in conspiracy theories and to motivate users to donate to charitable causes, but it has also been associated with deception and various harmful outcomes. Researchers today cite a need for further dedicated research on the effects of anthropomorphized AIs to best inform decisions about implementation and spread of AI agents.\nAnticipation of the ubiquity of anthropomorphic AI systems has led to concern over potential harms that may not be entirely realized today. In particular, some researchers foresee that the delineation between what is actually human and what is merely human-like may become less clear as the gap between human and AI capabilities grows smaller and smaller. This, some argue, may adversely affect human collective self-determination, as non-human entities gradually begin to shape our core value systems and influence society. It may also lead to the degradation of human social connections, as humans may come to prefer interacting with AI systems that are designed with user satisfaction as a priority; this can have a multitude of negative implications. For example, AI agents already display a significant degree of sycophancy, which means that an increasing role for AI agents in users' opinion space may result in increased polarization and a decrease in value placed in others' beliefs. Acclimatization to the conventions of human-AI interaction may undermine the value we place on human individuality and self-expression, or may lead to inappropriate expectations derived from AI interactions being placed on human interactions. In general, human social connectedness is known to play a critical role in individual and group well-being, and its replacement with AI interactions may result in mass dissatisfaction or lack of fulfillment.\n\n\n=== Proposed directions ===\nGiven the demonstrated and projected effects of AI anthropomorphism, a variety of suggestions have been made intending to inform future development of AI. Much of this discourse is centered around curbing the most harmful effects of anthropomorphism. For example, some researchers have called for a moratorium on the use of language which deliberately invokes humanness; this applies both to how AI companies describe their products and the language output by the systems themselves. Particularly, it has been suggested that terms like \"seeing\", \"thinking\", and \"reasoning\" should be replaced by terms like \"recognizing\", \"computing\", and \"inferring\", and that first-person pronouns such as \"I\" and \"my\" should not be used by chatbots. Another idea is the implementation of a specific AI accent or dialect that would clearly indicate when language was generated artificially. However, given the commercial pressures to optimize AI agents for economic gain – which may involve exploiting anthropomorphic qualities – it may not be prudent to rely on the restraint of developers, meaning that increased regulation may be necessary to limit harms. As of now, there are no laws that directly address anthropomorphism in AI; potential avenues for regulation include requirements for transparency and built-in safeguard mechanisms. More generally, researchers cite a need for increased understanding of the kinds and degrees of anthropomorphic qualities possessed by AI systems. To that end, it has been proposed that new benchmarks and tests should be developed that measure anthropomorphic qualities in AI writing, inference, and interaction.\n\n\n== In popular culture ==\nAnthropomorphic portrayals of AI are common in film, literature, and other interactive media. These depictions often emphasize human-like qualities of AI in ways that shape public perceptions.\n\n\n=== Film and television ===\nThere are a number of well-known portrayals in movies and TV of AI possessing human-like agency or personalities. In film, HAL 9000 in 2001: A Space Odyssey and Ava in Ex Machina are depicted with complex emotions and motives. Television portrayals include Data from Star Trek: The Next Generation and KITT from Knight Rider.\n\n\n=== Literature ===\nAnthropomorphic AI is also common in literature. Isaac Asimov's robot characters, including R. Daneel Olivaw, exhibit human reasoning and moral dilemmas, while Iain Banks's \"Minds\" in The Culture series are portrayed as having distinct personalities and social roles.\n\n\n=== Video games ===\nExamples of anthropomorphized AI in video games include GLaDOS in Portal, a witty and sinister guide for the player, and Cortana in the Halo series, who forms emotional bonds with human protagonists.\n\n\n=== Advertising and consumer technology ===\nMarketing campaigns for digital assistants such as Amazon Alexa, Google Assistant, and Siri often portray the systems as personable or empathetic. Consumer robots like Sony's AIBO and SoftBank Robotics' Pepper are intentionally designed with expressive behaviors that encourage users to treat them as social agents.\n\n\n== References ==", "extra": {}}
{"source": "wikipedia", "title": "AI browser", "url": "https://en.wikipedia.org/wiki/AI_browser", "summary": "An AI browser is a web browser with integrated artificial intelligence capabilities, such as automatically summarizing web page content or answering questions about it. A more specialized type is an agentic browser, based on the concept of agentic AI, which can take actions – such as navigating webpages or filling out forms – on behalf of the user.", "content": "An AI browser is a web browser with integrated artificial intelligence capabilities, such as automatically summarizing web page content or answering questions about it. A more specialized type is an agentic browser, based on the concept of agentic AI, which can take actions – such as navigating webpages or filling out forms – on behalf of the user.\nSeveral agentic browsers emerged in 2025, including ChatGPT Atlas (macOS only), Comet, and Dia.\nAs of 2025, this is a recent development in the browser market, including new entrants from OpenAI, Opera and Perplexity. The designation of 'AI browser' also includes established browsers that later added non-agentic AI features, such as Microsoft Edge with the Copilot chatbot, Google Chrome with the Gemini chatbot (for Windows desktop users in the US with their language set to English), and Firefox with multiple chatbot providers (such as ChatGPT, Claude, Copilot, Gemini, and Le Chat).\nAI browsers have been noted to be susceptible to prompt injection attacks.\n\n\n== Browser extensions and integrations ==\nRather than creating entirely new browsers, some AI browsing solutions integrate with existing browsers through extensions or companion applications. These tools add agentic capabilities to established browsers without requiring users to switch platforms. Examples include Composite, which functions as a cross-browser agent that works with Chrome, Edge, and other browsers to automate web-based tasks for workers.\n\n\n== Cloud-based implementations ==\nCloud-based implementations of AI browsers allow users to run automated browsing agents without local installation. These systems operate on remote servers using frameworks such as Puppeteer or Playwright. Examples include Browserbase, Browser-use and AI Browser. The AI typically parses the Document Object Model (DOM) to locate and interact with page elements, and may also analyze browser screenshots to interpret layout and structure.\n\n\n== Criticisms and dangers ==\nAI browsers have been noted to be susceptible to being vulnerable to prompt injection attacks, in which the content of websites can be used to hijack the control of the browser. Multiple organisations have argued against using AI browsers due to this vulnerability. The United Kingdom national cyber security centre and Gartner consider them to be too risky for adoption by most organisations.\nA study by the CISPA Helmholtz Center and Saarland University concluded that this vulnerability makes them easy targets for malware, fraud, automated defamation, disinformation and biased outputs.\n\n\n== References ==", "extra": {}}
{"source": "wikipedia", "title": "AI build-out financing", "url": "https://en.wikipedia.org/wiki/AI_build-out_financing", "summary": "To finance the build-out of AI data centres during the 2020s, an unprecedented amount of capital has been mobilised in the USA in particular amid a broad competition for global technnological supremacy in artificial intelligence. ", "content": "To finance the build-out of AI data centres during the 2020s, an unprecedented amount of capital has been mobilised in the USA in particular amid a broad competition for global technnological supremacy in artificial intelligence. \nCitigroup forecast that $2.8 trillion of capital expenditure would be required for AI data centre infrastructure by 2030, while McKinsey estimated that almost $7 trillion would be spent globally by that time. Between January and August 2024, Microsoft, Meta, Google and Amazon collectively spent $125 billion on AI data centres. According to S&P Global, $61 billion has been spent on the data centre market as a whole in 2025, while debt issuance for data centres was $182 billion during the same year. In 2026, major tech companies were estimated to spend $650 billion on AI data centres.\nVendor financing has become a prominent feature of the build-out of artificial intelligence (AI) infrastructure in the mid-2020s, as chipmakers and cloud-computing providers have extended capital, guarantees and purchase commitments to the companies that buy their hardware and services. The practice is most closely associated with Nvidia, whose graphics processing units (GPUs) power most large-scale AI training and inference, and with the \"hyperscalers\"—large cloud-computing operators such as Microsoft, Amazon and Google's parent Alphabet—and \"frontier labs,\" a term used for AI developers building the most advanced large language models, principally OpenAI and Anthropic. Because these companies are simultaneously one another's investors, suppliers and customers, commentators have described the resulting arrangements as \"circular financing\" or \"circular deals\".\n\n\n== Circular financing ==\nA circular deal typically involves one company – often a chip manufacturer or cloud provider—taking an equity stake in, or extending financing to, a firm that in turn commits to buy that same company's products or services. Nvidia has been described as \"particularly aggressive\" in this respect, investing in AI startups, funding infrastructure buildouts, and taking stakes in data-centre and \"neocloud\" companies that rent out access to its chips. Examples cited by Bloomberg include Nvidia's investments in Nscale, Nebius, CoreWeave (in which it also agreed to buy $6.3 billion of cloud services), Safe Superintelligence and Naver, as well as up to $100 billion pledged toward OpenAI (of which it ultimately contributed $30 billion to a funding round) and up to $15 billion pledged jointly with Microsoft toward Anthropic. Similar dynamics have linked Microsoft and OpenAI (which has committed to purchase $250 billion of Microsoft cloud services), Amazon and Google with Anthropic, and AMD with both OpenAI and Anthropic.\nProponents argue such arrangements function as a \"virtuous circle\" or flywheel: capital from a supplier lets a customer expand, which in turn generates more demand for the supplier's products, at a time when the most advanced AI chips remain scarce and companies seek to lock in supply through financing tied to long-term purchase commitments. Anthropic chief executive Dario Amodei defended the practice on similar grounds, saying that one party has capital and a commercial interest in selling chips while the other expects future revenue but lacks the up-front capital.\nCritics counter that circularity can distort incentives: a company with a major shareholder among its own suppliers may be more inclined to keep purchasing that supplier's products regardless of commercial merit, and if a customer's own revenue fails to keep pace with its data-centre and hardware bills, the supplier-investor can lose twice—once as its customer stops buying, and again as the value of its equity stake falls. Bloomberg noted this risk is heightened in AI infrastructure because a relatively small number of buyers account for a large share of the market.\n\n\n=== Circular investments ===\nBeyond loan guarantees and lease-financing structures, chipmakers and their customers have exchanged direct equity stakes and warrants on a large scale, producing cross-shareholdings across the sector. The research firm IDC defined the resulting pattern as one \"where the same capital flows simultaneously as vendor payment and equity stake,\" such that a company \"funds its own customer's revenue while also supplying that customer's core infrastructure,\" producing revenue growth that IDC described as real but not cleanly separable from the investing company's own capital deployment.\nNvidia has been the most prolific corporate investor in this sector. Between 2021 and 2025 it participated in 283 funding rounds across 241 companies, roughly 85% of them AI startups, making it the fourth-largest corporate venture investor of 2023 behind Microsoft, SoftBank and Alphabet; all ten of the most heavily funded private companies on the Forbes AI 50 list, including competitors OpenAI and Anthropic, had received Nvidia investment. Its disclosed positions include a roughly $30 billion stake in OpenAI, built alongside agreements for OpenAI to purchase Nvidia systems; a 7% stake in CoreWeave alongside a $6.3 billion cloud-services purchase commitment; and $2 billion positions each in Nebius, Marvell Technology, Synopsys, Coherent and Lumentum, along with smaller stakes of up to $2.1 billion in Iren and up to $3.2 billion in Corning. Nvidia also disclosed $5 billion invested in Safe Superintelligence and $1 billion in Naver, and was a repeat investor in Mistral AI, Cohere and Perplexity AI.\nNvidia's two largest disclosed equity positions outside the AI-software sector are in fellow hardware suppliers. It finalised a roughly $5 billion, 4% stake in Intel in late 2025 (214.7 million shares at $23.28 each), tied to joint development of x86 central processing units incorporating Nvidia's NVLink interconnect and of system-on-chips combining Intel processor cores with Nvidia RTX graphics chiplets; that stake was valued at roughly $19.9 billion as of 19 August 2026 after Intel's share price rose sharply, having peaked at more than $30 billion in late June. Separately, Nvidia disclosed a stake of nearly $21 billion in SpaceX (122.8 million shares at $170.86 each) in August 2026, making it the company's sixth-largest outside investor; the position originated as a $10 billion investment in xAI that converted into SpaceX shares once SpaceX acquired xAI, and had fallen in value to roughly $17.2 billion as of 19 August 2026 as SpaceX's private share price declined. Reporting on the stake noted that SpaceX builds its own AI data centres exclusively with Nvidia chips.\nOther chipmakers have taken comparable positions in their own customers. Microsoft had invested more than $13 billion in OpenAI by the mid-2020s, the largest tranche – $10 billion – in a 2023, alongside OpenAI's commitment to purchase $250 billion of Microsoft cloud services. Amazon and Google committed up to $4 billion and $2 billion respectively to Anthropic in 2023, and Amazon separately agreed in 2026 to invest $15 billion in OpenAI with a further $35 billion pledged conditionally. Microsoft and Nvidia jointly pledged up to $15 billion toward Anthropic in late 2025. In July 2026, AMD announced it would invest up to $5 billion in Anthropic as part of a deal under which Anthropic would deploy 2 gigawatts of AMD Instinct MI450-series GPUs in AMD's Helios rack-scale systems, with the first gigawatt due in the first half of 2027. AMD had separately issued OpenAI a warrant for up to 160 million shares of AMD common stock – equivalent to roughly 10% of the company – with vesting tied to both the volume of AMD hardware OpenAI deployed and the performance of AMD's own share price.\nThese equity exchanges have run alongside, and in some cases overlapped with, a wider set of compute-purchase commitments. Anthropic alone entered a $1.25 billion-a-month agreement through May 2029 for exclusive use of the computing capacity at SpaceX's Colossus 1 data centre in Memphis, Tennessee; a multi-billion-dollar agreement with Amazon; a multi-gigawatt compute agreement with Google and Broadcom; and, according to CNBC, preliminary talks to lease computing power from Meta, while reporting a revenue run-rate of $47 billion by May 2026 (up from about $10 billion for the whole of 2025) and closing a funding round that month worth $965 billion.\nCommentators have noted that these overlapping stakes concentrate risk as well as opportunity: because Nvidia and other chipmakers hold equity in many of the same companies that buy their hardware, a slowdown in AI spending could reduce both their product revenue and the value of their investment portfolios simultaneously, deepening rather than diversifying their exposure to a downturn in the sector. Reporting has also noted that, because Nvidia and its peers can build influence over customers and competitors through minority equity stakes rather than outright acquisitions, such arrangements can fall outside regulatory scrutiny that would apply to a full acquisition.\n\n\n==== Summary of major disclosed stakes ====\n\n\n=== Nvidia's asset-backed financing plan ===\nIn August 2026, Nvidia chief executive Jensen Huang announced a strategic partnership with six major asset managers – Apollo Global Management, BlackRock, Blackstone, Brookfield Asset Management, Goldman Sachs and KKR to standardise chip financing through securitisation of $500 billion package for AI companies. Likening Nvidia chips to investable commercial real estate, he explained: \"because it’s productive, it’s revenue generating, it is fungible, it’s used by just about every cloud service provider, it runs every AI model.” Under one model outlined by executives, investors would buy asset-backed securities, which would be used to fund data centre construction and hardware, and lease it to an end customer (such as OpenAI or AWS); lease payments from the customer would flow back through the SPV to repay investors. Loans would be backed by the Nvidia hardware itself as collateral, on the premise that if one customer defaulted, another could quickly take its place given demand for the chips. BlackRock chief executive Larry Fink compared the moment to the early growth of the mortgage-backed security market in the 1970s.\nHuang said the arrangement was designed to address criticism of Nvidia's earlier circular funding deals, under which the company had invested directly in, or guaranteed financing for, its own customers. He said Nvidia might still backstop up to 25% of a project's cost through a \"residual-value support mechanism,\" under which the company would make up the difference if collateral backing a loan fell below an agreed floor—an arrangement that would likely keep the obligation off Nvidia's balance sheet initially while leaving it exposed if an end customer defaulted. Some analysts, including Morgan Stanley's Joseph Moore, characterised the addition of third-party capital and credit decisions as likely to \"alleviate circularity concerns...while admittedly polarizing the debate.\" Rival chipmaker Broadcom had already provided a comparable backstop, disclosing up to $29 billion of exposure, on a $35 billion package arranged by Apollo, Blackstone and a group of banks to lease chips to Anthropic in June 2026.\n\n\n=== Credit-default-swap market signals ===\nBy mid-2026, participants in the credit default swap (CDS) market—where investors trade contracts that function as a form of insurance against a borrower defaulting on its debt, with the price (quoted in basis points) rising as perceived default risk increases—began treating Nvidia's spread as a barometer for stress in AI-related vendor financing generally. Nvidia's five-year CDS reached a record 82 basis points on 27 July 2026, according to ICE Data Services, its largest single-day intraday move since the contract began actively trading in November 2025; the spread had stood at roughly 40 basis points at the start of that month and around 68 basis points three days earlier. One equity trader, Daniel O'Regan of Mizuho, remarked that he had not previously monitored Nvidia's CDS at all, given the company's size and profitability, until the spread's rapid rise prompted him to.\nThe move coincided with reporting, credited to Bloomberg News and citing unnamed sources, that Nvidia was negotiating AI infrastructure arrangements potentially worth more than $750 billion in aggregate, including a possible $250 billion guarantee to support OpenAI's lease of capacity at a planned Ohio data centre and a separate facility to help finance $350 billion of OpenAI chip purchases; the Financial Times separately reported that Nvidia was backing a $50 billion lease on a Texas data centre built around its own chips. Nvidia shares fell nearly 5% on the day of the reporting, closing at $196.51 and shedding roughly $250 billion of market value, which briefly cost the company its position as the world's most valuable listed firm to Apple. Bloomberg separately reported that Nvidia had already disclosed more than $540 billion of such financing arrangements during 2026 prior to the reported OpenAI negotiations, and both the International Monetary Fund and the Bank for International Settlements identified AI-related circular financing as a systemic downside risk. Investment strategist Billy Leung of Global X Management characterised additional Nvidia guarantees for OpenAI's data-centre debt as evidence of \"funding strain in the AI buildout\" as much as of underlying customer demand.\nCredit stress was reported to have spread to other companies tied to the AI build-out: S&P Global downgraded Oracle's credit rating to BBB−, its lowest investment-grade tier, with Oracle's own five-year CDS widening to 215 basis points, and Alphabet was reported to have posted negative free cash flow for the first time since its initial public offering as AI-related spending increased. A FactSet analysis cited in the same reporting found that combined capital expenditure by five major hyperscalers – Alphabet, Amazon, Meta, Microsoft and Oracle – was on track to exceed $690 billion in fiscal 2026, over 80% higher year-on-year, with free cash flow at most of the five approaching zero or turning negative as AI-related costs were front-loaded. Société Générale's head of US equity strategy, Manish Kabra, was quoted summarising the shift in investor attention as being toward credit rather than earnings measures for hyperscale computing firms. At the time of reporting, Nvidia shares were trading around 17% below their 52-week high, and the company had not issued a public statement on the reported OpenAI arrangement, with deal terms unconfirmed pending its next quarterly results.\n\n\n=== Risks and criticism ===\nAnalysts and investors identified several risks specific to using AI computing hardware as loan collateral. Unlike more traditional forms of vendor-financed collateral such as real estate or aircraft, GPUs depreciate quickly as new generations of chips are released, and may be further accelerated by commoditisation of compute by Chinese competitors. Following a 54-day training run for Llama 3 in 2024, Meta estimated the failure rate could be around 9% a year. Another factor of obsolescence is energy efficiency: new chips are said to be \"orders of magnitude more energy efficient than the A100\". Jensen Huang responded that sustained customer use of older-generation Nvidia chips demonstrated that the hardware retained value well after release. Underwriters may also face the risk that AI infrastructure build-out could stall, or that projects could be delayed or cancelled if demand for AI products falls short of expectations.\nNvidia, who is driving the project and whose revenue depends on customers being able to afford its chips, obviously stands to benefit from the financing arrangement, as will smaller AI labs, cloud companies and enterprises that face comparatively high borrowing costs. The intended purchasers of the resulting debt are institutional investors such as pension funds, insurers and sovereign-wealth funds. Bond issuance tied to AI infrastructure – covering big technology companies, data-centre projects and chip-financing vehicles – reached $344 billion by early August 2026, an increase of more than $200 billion over 2025, according to Bank of America Global Research. They projected a further $100 billion of issuance from large technology companies before the end of 2026. Weaker borrowers already faced elevated yields: CoreWeave raised a $2.6 billion loan facility, backed by contracts with Anthropic and Jane Street, at more than nine percentage points over benchmark rates, while a $3.5 billion junk-bond sale by Galaxy Digital to fund a Texas data centre leased to CoreWeave priced at nearly 10%. A price war from China would significantly lower collateral values in the event of default. Larry Fink saw a parallel to the creation of the mortgage backed security.\nMichael Burry, known for shorting mortgage-backed securities before the 2008 financial crisis and considers AI a bubble, is bearish on Nvidia and other AI stocks, said the structuring of such credit late in the bull phase reflected vulnerability in the AI market. Bloomberg's analysis drew an explicit parallel to vendor financing during the late-1990s telecommunications boom, when equipment makers extended loans and other support that allowed carriers to sustain heavy investment in fibre-optic networks; when demand forecasts were not met, several heavily leveraged carriers cut spending or filed for bankruptcy, and regulators later examined \"capacity swap\" transactions at firms including Qwest and Global Crossing, some of which led to restated revenue. Venture capitalist Paul Kedrosky, who covered the telecom sector during that period, said AI capital spending was approaching levels last seen at the peak of the fibre-optic buildout, with a risk that facilities built around current-generation chips could become obsolete before recouping their cost.\n\n\n== See also ==\nHire purchase – Form of arrangement\nTied aid – International transfer of resources with geographic limits on procurement and spending\n\n\n== Footnotes ==\n \n\n\n== References ==", "extra": {}}
{"source": "wikipedia", "title": "The AI Con", "url": "https://en.wikipedia.org/wiki/The_AI_Con", "summary": "The AI Con: How to Fight Big Tech's Hype and Create the Future We Want is a 2025 non-fiction book by linguist Emily M. Bender and sociologist Alex Hanna. It argues that \"artificial intelligence\" is a misleading term that obscures ordinary automation while concentrating power in a small number of technology firms. The book was published in May 2025 by Harper in the United States and Bodley Head in the United Kingdom. It was developed alongside the authors' podcast Mystery AI Hype Theater 3000, which critiques exaggerated claims about AI.", "content": "The AI Con: How to Fight Big Tech's Hype and Create the Future We Want is a 2025 non-fiction book by linguist Emily M. Bender and sociologist Alex Hanna. It argues that \"artificial intelligence\" is a misleading term that obscures ordinary automation while concentrating power in a small number of technology firms. The book was published in May 2025 by Harper in the United States and Bodley Head in the United Kingdom. It was developed alongside the authors' podcast Mystery AI Hype Theater 3000, which critiques exaggerated claims about AI.\n\n\n== Synopsis ==\nThe authors present AI as a marketing umbrella that encourages audiences to infer understanding and agency where none exist. They argue readers should treat such language skeptically and to separate specific automated tasks from broad claims of intelligence. The book describes a recurring hype cycle in which corporate narratives justify data and labor extraction, the replacement of human services with cheaper substitutes, and the diversion of attention from present harms to speculative futures. While acknowledging limited uses such as pattern recognition, the authors argue that contemporary systems are best understood as text and media generators shaped by training data and human labor, not as thinking or reasoning entities.\nA central theme is the social and environmental cost of scaling these systems, including increased energy and water use, the appropriation of creative work for training, and the outsourcing of ghost work to low-paid data workers worldwide. These costs are linked to workplace effects, with the authors arguing that automation rarely eliminates jobs outright and more often degrades them through surveillance, work intensification, and unpaid oversight. As alternatives to passive adoption, the authors propose concrete responses: asking precise questions about what is being automated and why, demanding transparency about data and evaluation, and practicing what they call strategic refusal when deployment conflicts with evidence or values.\nThe book also develops a vocabulary for public debate, rejecting both boosterish and doomerish narratives as grounded in the same assumption that AI is a singular, autonomous force. The authors recommend reading strategies such as favoring trusted human sources over automated summaries and using humor to deflate inflated claims. They describe a link between language to policy and power, arguing that precise terminology can help policymakers and the public resist austerity-driven automation and demand accountability for errors and harms.\n\n\n== Reception ==\nThe Guardian praised the book's myth-busting approach and its analysis of how hype erodes cultural and civic life by normalizing synthetic media as a substitute for human judgment. Kirkus Reviews described it as a contrarian account that catalogs concrete risks while cutting through speculative predictions. An interview in Business Insider highlighted the authors' accessible frameworks, including their proposal to describe chatbots as conversation simulators and to evaluate systems in terms of values, labor, and evidence. Coverage in GeekWire emphasized the book's call for resistance through collective bargaining, stronger data rights, and a norm of rejecting deployments that fail basic standards of necessity and evaluation.\nSome reviews were more critical. A review in LLRX argued that the book's tone could be overly polemical and that it gave limited attention to potential benefits claimed for generative systems. Coverage in the Financial Times focused on Bender's broader public scholarship, situating the book within her long-standing critique of anthropomorphic narratives about large language models and her advocacy for more democratic oversight of automated systems.\n\n\n== References ==", "extra": {}}
{"source": "wikipedia", "title": "AI data center", "url": "https://en.wikipedia.org/wiki/AI_data_center", "summary": "An AI data center is a specialized data center facility designed for the computationally intensive tasks of training and running artificial intelligence (AI) and machine learning models. Unlike general-purpose data centers, they are often optimized for the parallel processing demands of AI workloads, typically using hardware such as AI accelerators (GPUs, TPUs) and high-speed interconnects.", "content": "An AI data center is a specialized data center facility designed for the computationally intensive tasks of training and running artificial intelligence (AI) and machine learning models. Unlike general-purpose data centers, they are often optimized for the parallel processing demands of AI workloads, typically using hardware such as AI accelerators (GPUs, TPUs) and high-speed interconnects.\nThe global push to construct these specialized facilities accelerated dramatically during the AI boom of the 2020s. In 2026, major tech companies were estimated to spend $650 billion on AI data centers. Their purchases of memory chips, particularly High Bandwidth Memory, led to a global memory supply shortage amid a broader competition for semiconductors, power, and infrastructure. About 70% of global computer memory production in the 2026 fiscal year has been purchased for AI data centers. \nAI-focused data centers use around 60 kilowatts of electricity per server rack, while general-purpose data centers typically use around 10 kilowatts of electricity per rack.\nBy 2026, many U.S. communities had organized against the local construction of AI data centers, which have a large power and water footprint and generate heat, noise, and other pollutants. By July 2026, local community resistance had blocked the construction of AI data centers worth some $130 billion.\n\n\n== Operators ==\nAs of August 2025, The Information tracked 18 planned or existing AI data centers in the United States, operated by Amazon Web Services, CoreWeave, Crusoe, Meta, Microsoft/OpenAI, Oracle, Tesla, and xAI. Other AI data center operators include Digital Realty and Alibaba. Data centers are also being built in China, India, Europe, Saudi Arabia, and Canada. The New Yorker described CoreWeave as the most prominent AI data center operator in the United States.\nTwo types of data center providers for machine learning have been noted in news media: hyperscalers and neoclouds. The Verge listed large technology companies such as Google, Meta, Microsoft, Oracle and Amazon as hyperscalers. The New York Times described neoclouds as \"a new generation of data center providers\". CoreWeave, Nebius, Nscale, and Lambda have been described as examples of neoclouds.\nIn January 2025, OpenAI, in partnership with Oracle and Softbank, announced the Stargate project, which as of September 2025 is composed of six built or proposed AI data centers in the United States.\nIn response to the Stargate project, Amazon launched in October 2025 an AI data center on 1,200 acres of farmland in Indiana. This data center, known as Project Rainier, is one of the largest AI data centers in the world, with Amazon spending $11 billion on the project. Rainier is specifically intended for training and running machine learning models from Anthropic. As of that time, this facility contains seven data centers (out of an estimated 30 planned) and will use 2.2 gigawatts of electricity (equivalent to 1 million households) and millions of gallons of water per year. Computer chips from Annapurna Labs and Anthropic, Trainium 2, were designed for use in such facilities. Amazon pumped millions of gallons of water out of the ground to construct the data center, and as of June 2025, Indiana state officials are investigating whether this dewatering process led to dry wells for local residents.\nIn November 2025, Anthropic announced a partnership with Fluidstack to develop artificial intelligence infrastructure in the United States, including data centers in New York and Texas, worth $50 billion.\nOther AI data center projects include the Colossus supercomputer from xAI, a Louisiana-based project from Meta, Hyperion, expected to use 5 GW of power, and a second Ohio-based Meta project, Prometheus, with a capacity of 1 GW. A 3,200-acre AI data center, capable of 4.4-4.5 GW of power and located on the decommissioned Homer City Generating Station, is under construction as of 2025, and will use seven 30-acre gas generating stations supplied by EQT.\nAs of December 2025, CRH is working on over 100 data centers in the United States.\nIn 2025, ExxonMobil and NextEra announced plans to build a data center powered by natural gas and using carbon capture technology, with 1.2 GW of power capacity. They previously purchased 2,500 acres of land in the Southeastern United States and plan to market the data center to an artificial intelligence company.\nThe increased interest in AI data centers has led to several executives from companies in that space becoming billionaires, including CoreWeave, QTS, Nebius, Astera Labs, Groq, Fermi (which is connected to former United States Secretary of Energy Rick Perry), Snowflake and Cipher Mining.\nSeveral companies involved in cryptocurrency mining, such as Bitdeer, CoreWeave, Cipher Mining, TeraWulf, IREN, Core Scientific, and CleanSpark have also been involved with AI data centers.\n\n\n== Finances ==\n\nBetween January and August 2024, Microsoft, Meta, Google and Amazon collectively spent $125 billion on AI data centers. Citigroup forecasted that $2.8 trillion would be spent on AI data centers by 2030, while McKinsey and Company estimated that almost $7 trillion would be spent globally by that time. According to S&P Global, $61 billion has been spent on the data center market as a whole in 2025, while debt issuance for data centers was $182 billion during the same year.\nLarge technology companies have offloaded the financial risks of building AI data centers by setting up special purpose vehicles or by contracting with neoclouds. For example, Meta's Hyperion was mostly funded by Blue Owl Capital, which did so using a bond offering from PIMCO. Those bonds were sold to a number of clients, including BlackRock. Meta did not borrow money itself and instead established a special purpose vehicle from which it would rent the data center. This deal was structured by Morgan Stanley for $30 billion, the largest known private capital transaction as of 2025.\nNeoclouds such as CoreWeave have gone into debt to buy computer chips from Nvidia for their data centers, and the chips themselves have been used for loan collateral. As of December 2025, CoreWeave took out three GPU-backed loans, collectively worth $12.4 billion, from private credit firms (Blackstone, Coatue, BlackRock, PIMCO) and from banks (Goldman Sachs, JPMorgan Chase, Wells Fargo). Thus, these companies provide an indirect connection between private credit and established banks. Data centers have also established asset-backed securities, and debt for data centers has its own derivative financial products.\nThe real estate industry, including asset managers, public companies and private investors, has also invested in AI development.\n\n\n== Energy sourcing ==\n\n\n== Environmental footprint ==\n\nAverage AI data centers have an electricity footprint equivalent to 100,000 households, and use billions of gallons of water for cooling their hardware. In 2025, the International Energy Agency estimated that the larger AI data centers currently under construction could consume as much electricity as 2 million households. A 2024 report from the United States Department of Energy stated that data centers overall used 17 billion gallons of water per year in the United States, primarily due to \"rapid proliferation of AI servers\", and that this usage was forecasted to grow to nearly 80 billion gallons by 2028. Researchers estimated that AI data centers in the United States would emit 24–44 million metric tons of carbon dioxide and use 731–1,125 million cubic meters of water per year between 2024 and 2030.\nPeaking power plants, which have been proposed as a power source for AI data centers, emit sulfur dioxide and have historically been located disproportionately near communities of color in the United States.\nReciprocating internal combustion engines, proposed as another power source for a data center, emit PM 2.5, nitrogen oxides, and volatile organic compounds.\n\n\n== Global memory supply shortage ==\n\nThe construction of AI data centers during the AI boom starting in 2024 has exhausted the global supply of the computer memory. To meet demand from AI servers, the three major DRAM manufacturers, Samsung, SK Hynix, and Micron, prioritized production of High Bandwidth Memory (HBM) and high-end server DRAM, contributing to a global memory supply shortage and intensifying competition for advanced memory chips. Because HBM consumes far more fabrication capacity than conventional memory—producing a single bit of HBM requires roughly three times the wafer capacity of DDR5.\nThe resulting price increases and shortages have affected computer memory broadly, including HBM, DRAM, and NAND flash memory. \nUnder an agreement with Samsung and SK Hynix, the Stargate project alone is set to receive 900,000 DRAM wafers per month, roughly 40% of total global DRAM production.\n\n\n== In the United States ==\n\nIn the United States, both the Biden administration and second Trump administration supported the construction of AI data centers. In January 2025, then-president Joe Biden signed an executive order for federal government agencies to support AI data centers on federal sites built by private companies, study their effect on energy prices, and encourage their use of renewable energy. In April 2025, the United States Department of Energy suggested 16 possible sites, including Los Alamos National Laboratory, Sandia National Laboratories and Oak Ridge National Laboratory. In its July 2025 AI Action Plan, the second Trump administration supported increased production of AI data centers. \nSeveral US states have incentivized local data center construction. For example, in 2024, lawmakers in Michigan approved tax breaks for data center equipment and construction material. Some data center companies have also invested or promised to invest in the infrastructure of local communities.\nIn December 2025, Democratic senators Elizabeth Warren, Chris Van Hollen, and Richard Blumenthal wrote to seven technology companies (Google, Microsoft, Amazon, Meta, CoreWeave, Digital Realty, and Equinix) that they will investigate the effects of those companies' operations on consumer energy bills, highlighting AI data centers in particular. That same month, 25 Democratic lawmakers wrote to the inspector general of the United States Department of Commerce over possible conflict of interest concerns involving secretary of commerce Howard Lutnick and AI data centers. Lutnick met with a co-founder of Fermi, which proposed an AI data center project financed by Cantor Fitzgerald and Newmark Group.\nAI Infrastructure Coalition (AIIC), an organization led by Brian O. Walsh, Kyrsten Sinema, and Garret Graves, has supported Trump's AI Action Plan, with Sinema becoming personally involved in a local meeting on an AI data center in Arizona. Venture capital firm Andreessen Horowitz, Big Tech companies (Google, Meta, Microsoft), telecommunications companies (Cisco, Corning Inc., Lumen Technologies), data center providers (Digital Realty, QTS), energy companies (Duke Energy, Entergy, ExxonMobil, NextEra Energy), and utility companies (PG&E, Pinnacle West Capital) are members of AIIC as of November 2025. AIIC has been supported by Hogan Lovells.\nAccording to Goldman Sachs in 2025 an addition half million workers will be needed by 2030 to build the power infrastructure, which is acute because of AI data center build out.\n\n\n=== Concerns and opposition ===\n\nSome analysts have expressed concerns about overbuilding of AI data centers, warning that their infrastructure risks obsolescence due to changes in demand and technology. Estimates for the lifespan of the GPUs used to power data centers range from one to eight years. By contrast, CPUs in more traditional data centers have a shelf life of about 5–7 years. Machine learning model training causes significant stress on computer chips.\n\nAdvocates have also linked data center construction to the AI bubble. Wisconsin Watch expressed concern that data centers could become stranded assets in the context of an AI bubble, leaving energy customers to pay for the costs. Journalists have warned that the large number of links between actors and organizations in the AI ecosystem as well as the large number of companies involved each indicate strong risks for the global economy and financial markets.\nA hedge fund founder questioned the economics of AI data centers, concluding that they need about $1 trillion of revenue for profitability.\nIn November 2025, the North American Electric Reliability Corporation warned that building new data centers could negatively affect the electrical grid and cause power outages during extreme weather. The independent monitor of PJM Interconnection warned that its power grid cannot support new data centers and supported a federal moratorium on data centers.\nMachine learning developers, when training large language models, often use data centers at their full capacity, conflicting with other users (including households) during peak usage, which may lead to blackouts.\nPower disruptions in data centers can cause errors in machine learning model calculations (\"silent data corruption\").\nLarge technology companies who are building data centers have asked public officials and land owners to sign non-disclosure agreements and have appeared to use shell companies. In one case, when a joint OpenAI-Oracle data center was rejected by Saline Township in Michigan, landowners and developers responded by suing the town. The developers acquired the land from the resulting settlement. The Michigan Public Service Commission approved DTE Energy to provide power to the data center, part of Stargate. The approval was part of an ex parte motion, with no public input.\nIn 2025, over 230 groups have signed a letter supporting a moratorium on constructing AI data centers in the United States, concerned by impact on the environment and energy bills. Signatories include Food & Water Watch, Greenpeace, Friends of the Earth and Physicians for Social Responsibility. Senator Bernie Sanders has also supported a moratorium on AI data centers.\nAshley LaMont, of Honor the Earth, argued that data centers on tribal lands would not help Native Americans obtain data sovereignty.\nThe NAACP has also expressed opposition to AI data centers.\n\n\n== In space ==\n\nSeveral tech companies, including Starcloud, Google, Nvidia, Blue Origin and SpaceX, have announced projects for or otherwise expressed interest in building data centers in outer space. Google announced Project Suncatcher, which plans to test whether its Tensor Processing Unit (TPU) chips (which are specifically designed for machine learning) can function in the environment of space. The company aims to deploy satellites with these chips in a data center by 2027. In November 2025, Starcloud, a startup supported by Nvidia, launched a satellite with an Nvidia H100 GPU that the company used to deploy and develop a large language model. CNBC reported that this model, NanoGPT, was the first model trained in space. Starcloud deployed a second model based on Google's Gemma.\n\n\n== References ==\n\n\n== External links ==\nCEO of Anthropic discusses some of the challenges of financing AI datacenters\n\n\n== Further reading ==\nYou’re Thinking About AI and Water All Wrong (December 12, 2025) by Wired\nHow Tech’s Biggest Companies Are Offloading the Risks of the A.I. Boom (December 15, 2025) by The New York Times", "extra": {}}
{"source": "wikipedia", "title": "AI effect", "url": "https://en.wikipedia.org/wiki/AI_effect", "summary": "The AI effect is a phenomenon in which advances in artificial intelligence lead to a redefinition of what is considered intelligence, such that capabilities achieved by AI systems are no longer regarded as examples of \"real\" intelligence.", "content": "The AI effect is a phenomenon in which advances in artificial intelligence lead to a redefinition of what is considered intelligence, such that capabilities achieved by AI systems are no longer regarded as examples of \"real\" intelligence.\nThe concept has been used to describe both a cognitive tendency and a sociotechnical pattern, in which successful AI techniques are reclassified as routine computation or absorbed into other domains.\nHistorian Pamela McCorduck described this as a recurring feature of AI research, noting in her 2004 book Machines Who Think that once a problem is solved, it is no longer considered evidence of intelligence. Researcher Rodney Brooks similarly observed in 2002 that once systems are understood, they are often regarded as \"just computation\".\n\n\n== Definition ==\nThe AI effect refers to a shift in how intelligence is defined as machines acquire new capabilities. Tasks such as playing chess, recognizing speech, or interpreting images were historically considered indicators of intelligence, but after successful automation they are often reclassified as routine computation.\nMcCorduck described this as an \"odd paradox\", in which successful AI systems are assimilated into other domains, leaving AI researchers to focus on unsolved problems. The phenomenon is often interpreted as an instance of moving the goalposts.\nA commonly cited formulation is Tesler's theorem, often expressed as \"AI is whatever hasn't been done yet\".\nWhen problems are not fully formalised, they may be described using models involving human computation, such as human-assisted Turing machines.\n\n\n== Historical examples ==\n\n\n=== Game playing ===\nEarly AI systems capable of playing games such as checkers and chess were initially regarded as demonstrations of machine intelligence. As these systems improved and became better understood, their achievements were often reinterpreted as examples of computation rather than intelligence.\nThe victory of IBM's Deep Blue over Garry Kasparov in 1997 is a frequently cited example. Critics argued that the system relied on brute-force methods rather than genuine understanding.\n\n\n=== Pattern recognition ===\nTechnologies such as optical character recognition and speech recognition were once considered core problems in artificial intelligence. As these systems became reliable and widely deployed, they were increasingly treated as standard engineering solutions.\n\n\n=== Integration into applications ===\nMany techniques originally developed within AI research have been incorporated into broader technological systems, including marketing, automation, and software applications.\nMichael Swaine reported in 2007 that AI advances are often presented as developments in other fields.\nMarvin Minsky observed that successful AI innovations often evolve into separate disciplines.\nNick Bostrom noted in 2006 that widely adopted technologies are often no longer labeled as AI.\n\n\n== Contemporary discussion ==\nThe AI effect continues to be discussed in the context of recent advances in machine learning, particularly large language models and other generative AI systems. As these systems have become more widely used, some researchers and commentators have noted that their capabilities are frequently described as statistical or mechanical once understood, rather than as intelligence. For instance, different combinations of human and AI action in workflows can be considered \"augmented intelligence\" and specific weights within various combinations of augmentation should be used to define ethical and practical considerations contextually for those workflows.\nA 2016 survey of artificial intelligence also noted that AI systems are increasingly embedded in everyday applications, reinforcing earlier observations that successful AI technologies tend to become normalized and no longer identified as AI.\nAt the same time, the widespread commercial use of artificial intelligence has led to greater visibility of the field, contrasting with earlier periods in which AI techniques were often present but unacknowledged.\n\n\n== Interpretations ==\n\n\n=== Cognitive bias ===\nSome authors describe the AI effect as a cognitive bias in which expectations of intelligence shift as machines achieve new capabilities.\n\n\n=== Sociotechnical perspective ===\nAnother interpretation emphasizes how technologies are reclassified over time as they become widespread and commercially successful.\n\n\n=== Philosophical debate ===\nSome philosophers argue that reclassification reflects genuine conceptual distinctions rather than bias.\n\n\n== Historical context ==\nDuring periods such as the AI winter, researchers sometimes avoided the term \"artificial intelligence\" due to negative perceptions.\nIn the 21st century, however, the term \"AI\" has become widely used in public discourse and marketing.\n\n\n== Broader implications ==\nThe AI effect has been linked to broader questions about human uniqueness and the nature of intelligence. Michael Kearns suggested that people may seek to preserve a special role for humans.\nSimilar patterns have been observed in studies of animal cognition.\nHerbert A. Simon noted that artificial intelligence can provoke strong emotional reactions.\n\n\n== See also ==\nELIZA effect\nMoravec's paradox\nChinese room\nFunctionalism (philosophy of mind)\nComputational intelligence\nArtificial intelligence in video games\nGod of the gaps\nOrganoid intelligence\n\n\n== References ==\n\n\n== Further reading ==\nMcCorduck, Pamela (2004). Machines Who Think.\nHofstadter, Douglas (1980). Gödel, Escher, Bach.\nPhillips, Everard M. (1999). If It Works, It's Not AI (PDF) (Thesis). MIT.", "extra": {}}
{"source": "wikipedia", "title": "AI infrastructure", "url": "https://en.wikipedia.org/wiki/AI_infrastructure", "summary": "AI infrastructure comprises the physical and software systems used to develop, train, deploy and operate artificial intelligence (AI) models and applications. In a narrow sense, the term refers to the physical computing resources that underpin AI, including semiconductors, processors, memory, servers, storage, networking equipment and data centres. Broader descriptions also include the software that makes this hardware usable, such as drivers, compilers, machine-learning frameworks, resource-management systems and cloud platforms.", "content": "AI infrastructure comprises the physical and software systems used to develop, train, deploy and operate artificial intelligence (AI) models and applications. In a narrow sense, the term refers to the physical computing resources that underpin AI, including semiconductors, processors, memory, servers, storage, networking equipment and data centres. Broader descriptions also include the software that makes this hardware usable, such as drivers, compilers, machine-learning frameworks, resource-management systems and cloud platforms.\nAI infrastructure is one part of a wider AI supply chain that also includes training data, models and applications. It may be owned and operated by an organisation, obtained through cloud computing, or provided through shared public or national computing programmes. The expansion of AI has made access to advanced chips, cloud capacity, electricity and suitable sites a subject of industrial policy, competition policy and environmental assessment.\n\n\n== Scope and terminology ==\nThere is no single universally used boundary for AI infrastructure. Policy reports sometimes use the term chiefly for physical resources, particularly advanced chips and the data-centre systems in which they operate. In computing practice, the related concept of the compute stack is commonly divided into hardware, software and infrastructure, with the layers affecting one another's performance and cost. Other analyses place hardware and cloud infrastructure as the first two layers of a supply chain that continues through data, foundation models and applications.\nUnder these definitions, data, algorithms, trained models and applications depend on infrastructure but are not necessarily treated as infrastructure themselves. An AI data center is a facility-level component of AI infrastructure, rather than a synonym for the entire stack. Infrastructure may support both AI and non-AI workloads: general-purpose computing and accelerated computing share facilities, storage and some network equipment, while differing in processors, interconnect requirements, power density and cost.\n\n\n== Components ==\n\n\n=== Computing hardware ===\nAI systems run on central processing units (CPUs) and on specialised accelerators that perform large numbers of operations in parallel. Common accelerators include graphics processing units (GPUs), application-specific integrated circuits (ASICs) and field-programmable gate arrays (FPGAs). Different workloads use different combinations of processors: large-scale model training commonly relies on accelerators, while some inference and supporting tasks can run on general-purpose CPUs.\nAccelerators are assembled into servers and then into clusters. Their usefulness depends on associated memory, storage, networking and systems software, so processor counts alone do not describe usable AI capacity. The semiconductor supply chain behind these systems includes chip design, manufacturing equipment, fabrication, packaging and materials. It is globally distributed but contains geographically concentrated inputs and specialised production stages, creating dependencies and potential bottlenecks.\n\n\n=== Memory, storage and networking ===\nTraining and operating AI models requires moving data between processors, memory and storage. Servers use memory close to the processors for active calculations, while storage systems retain training data, intermediate results and deployed model files. High-bandwidth connections link processors within a server, servers within a cluster and clusters to storage and external networks.\nLarge training workloads are often distributed across many machines. Distribution can reduce the time needed for a computation, but it introduces coordination and communication problems, including how to parallelise training efficiently and maintain a coherent model across workers. Consequently, cluster performance depends not only on arithmetic capacity but also on memory bandwidth, network latency, network bandwidth and the way work is divided.\n\n\n=== Data centres and cloud services ===\nData centres provide the buildings, racks, servers, storage, network equipment, power distribution, backup systems and cooling needed to operate computing hardware. Some organisations build private clusters, while others rent virtual machines, accelerator instances or managed AI services from public cloud providers. Cloud regions are physical hubs that place computing capacity in particular geographic and legal jurisdictions.\nCloud platforms combine infrastructure with model-development tools, databases, security services and application marketplaces. This integration can make large-scale resources available without building a private data centre, but it can also create technical and commercial dependence on a provider's interfaces, services and software ecosystem.\n\n\n=== Software ===\nThe software layer translates model operations into work that processors can execute. It includes device drivers, compilers, programming libraries and machine-learning frameworks, as well as software for assigning accelerators, scheduling jobs and monitoring clusters. Deployed systems also need data pipelines, model-serving systems and access controls.\nProduction machine-learning systems must handle requirements such as large and heterogeneous data flows, low-latency processing, resource management, security and privacy. Software architecture therefore affects how efficiently physical capacity can be used and how reliably a model can be integrated into a larger application.\n\n\n=== Electricity and cooling ===\nAI computing equipment requires electrical supply, power conversion and distribution, uninterruptible power supplies or other backup systems, and cooling. These supporting systems take up space and consume energy in addition to the processors themselves. Accelerated clusters can require more power and cooling than general-purpose servers, and rapid changes in workload can create large variations in electrical load.\nThe siting of infrastructure is therefore affected by access to grid connections, reliable electricity, cooling technology, water, telecommunications links and suitable land. These constraints can determine whether announced projects are built and how quickly capacity becomes available.\n\n\n== Training and inference ==\nMachine learning training uses data to adjust a model's parameters. Advanced training runs may occupy large accelerator clusters for extended periods and use distributed software to divide calculations and exchange intermediate results. A study of historical machine-learning systems estimated that training compute doubled about every six months after the emergence of deep learning in the early 2010s, although trends vary by model class and time period.\nInference applies a trained model to new inputs. Its infrastructure requirements range from small, latency-sensitive deployments to high-volume services running large models in data centres. Some inference workloads use accelerators, while smaller or more efficient models may run on general-purpose processors; cost, response time, throughput and availability influence the choice. Software and hardware efficiency can reduce the resources needed for a given level of performance. Between November 2022 and October 2024, the inference cost of a system performing at the level of GPT-3.5 fell more than 280-fold. Lower unit costs do not necessarily reduce total infrastructure demand if the number, scale or complexity of uses grows.\n\n\n== Economics and access ==\n\n\n=== Scale and market structure ===\nAI infrastructure has high fixed costs at several layers. Advanced semiconductor production requires specialised facilities and equipment, while cloud and data-centre operators invest in land, buildings, power systems, networking, storage and large numbers of processors. Scale can lower unit costs and support a wider range of services, but it can also make entry difficult for smaller firms.\nLarge technology companies operate across several layers of the AI supply chain, including cloud infrastructure, model development and applications. Vertical integration can simplify deployment and fund large investments, while also raising competition concerns about access, switching, preferential treatment and dependence on a small number of providers. The UK Competition and Markets Authority found that accelerated cloud computing required significant investment in accelerator chips and specialised infrastructure, and that large sunk costs and economies of scale were barriers in cloud services.\n\n\n=== Geographic distribution and public capacity ===\nAI computing capacity is distributed unevenly. Public cloud capacity is organised into regions, so users' practical access can depend on whether specialised hardware is available domestically or in a nearby jurisdiction. Location also affects latency, data-governance obligations, resilience and exposure to trade or supply restrictions.\nGovernments have consequently considered public, shared or nationally supported computing capacity. The OECD has described national compute planning in terms of capacity and utilisation, effective access and skills, and resilience issues including security, sovereignty and sustainability. Such programmes may provide access to researchers, public institutions and smaller firms that cannot procure large clusters directly, but they still depend on global semiconductor, networking and energy supply chains.\n\n\n== Environmental impacts ==\nThe environmental effects of AI infrastructure occur across the equipment life cycle. They include energy and water used in semiconductor production and data-centre operation, greenhouse-gas emissions associated with electricity generation, materials used in servers and supporting equipment, and electronic waste at end of life. Environmental assessment can distinguish these direct effects from the indirect environmental costs or benefits of applications that use AI.\nAI-specific electricity use is difficult to isolate because data centres host mixed workloads and operators disclose limited comparable data. The International Energy Agency estimated that global data-centre electricity consumption was about 485 terawatt-hours in 2025 and projected roughly 950 terawatt-hours in 2030; in its central projection, electricity use by AI-focused data centres grows faster and triples over that period. These figures cover scenarios rather than a fixed outcome and depend on efficiency, demand, grid connections and the rate at which proposed facilities are completed.\nWater and emissions also vary by location, electricity mix and cooling system. A 2025 scenario study of AI servers in the United States estimated an annual water footprint of 731–1,125 million cubic metres and additional annual emissions of 24–44 million tonnes of carbon-dioxide equivalent between 2024 and 2030, depending on expansion; the authors emphasised deep uncertainty from server locations, efficiency measures and grid decarbonisation. Proposed responses include more efficient hardware and software, higher equipment utilisation, lower-carbon electricity, water-aware siting and cooling, greater disclosure, and life-cycle measurement standards.\n\n\n== See also ==\nAI data center\nCloud computing\nDistributed computing\nHigh-performance computing\nMLOps\nSovereign AI\nSupercomputer\n\n\n== References ==", "extra": {}}
{"source": "wikipedia", "title": "AI safety", "url": "https://en.wikipedia.org/wiki/AI_safety", "summary": "AI safety is an interdisciplinary field focused on preventing accidents, misuse, or other harmful consequences arising from artificial intelligence systems. It encompasses AI alignment (which aims to ensure AI systems behave as intended), monitoring AI systems for risks, and enhancing their robustness. The field is particularly concerned with existential risks posed by advanced AI models.", "content": "AI safety is an interdisciplinary field focused on preventing accidents, misuse, or other harmful consequences arising from artificial intelligence systems. It encompasses AI alignment (which aims to ensure AI systems behave as intended), monitoring AI systems for risks, and enhancing their robustness. The field is particularly concerned with existential risks posed by advanced AI models.\nBeyond technical research, AI safety involves developing norms and policies that promote safety, including advocacy for regulations at different levels of government. The field gained significant popularity in 2023, with rapid progress in generative AI and public concerns voiced by researchers and CEOs about potential dangers. During the 2023 AI Safety Summit, the United States and the United Kingdom both established their own AI Safety Institute. However, researchers have expressed concern that AI safety measures are not keeping pace with the rapid development of AI capabilities.\n\n\n== Motivations ==\nScholars discuss current risks from critical systems failures, bias, and AI-enabled surveillance, as well as emerging risks like technological unemployment, digital manipulation, weaponization, AI-enabled cyberattacks and bioterrorism. They also discuss speculative risks from losing control of future artificial general intelligence (AGI) agents, or from AI enabling perpetually stable dictatorships.\n\n\n=== Existential safety ===\n\nSome have criticized concerns about AGI, such as Andrew Ng who compared them in 2015 to \"worrying about overpopulation on Mars when we have not even set foot on the planet yet\". Stuart J. Russell on the other side urges caution, arguing that \"it is better to anticipate human ingenuity than to underestimate it\".\nAI researchers have widely differing opinions about the severity and primary sources of risk posed by AI technology – though surveys suggest that experts take high consequence risks seriously. In two surveys of AI researchers, the median respondent was optimistic about AI overall, but placed a 5% probability on an \"extremely bad (e.g. human extinction)\" outcome of advanced AI. In a 2022 survey of the natural language processing community, 37% agreed or weakly agreed that it is plausible that AI decisions could lead to a catastrophe that is \"at least as bad as an all-out nuclear war\".\n\n\n== History ==\nRisks from AI began to be seriously discussed at the start of the computer age:\n\nMoreover, if we move in the direction of making machines which learn and whose behavior is modified by experience, we must face the fact that every degree of independence we give the machine is a degree of possible defiance of our wishes.\nIn 1988, Blay Whitby published a book outlining the need for AI to be developed along ethical and socially responsible lines. \nFrom 2008 to 2009, the Association for the Advancement of Artificial Intelligence (AAAI) commissioned a study to explore and address potential long-term societal influences of AI research and development. The panel was generally skeptical of the radical views expressed by science-fiction authors but agreed that \"additional research would be valuable on methods for understanding and verifying the range of behaviors of complex computational systems to minimize unexpected outcomes\".\nIn 2011, Roman Yampolskiy introduced the term \"AI safety engineering\" at the Philosophy and Theory of Artificial Intelligence conference, listing prior failures of AI systems and arguing that \"the frequency and seriousness of such events will steadily increase as AIs become more capable\".\nIn 2014, philosopher Nick Bostrom published the book Superintelligence: Paths, Dangers, Strategies. He has the opinion that the rise of AGI has the potential to create various societal issues, ranging from the displacement of the workforce by AI, manipulation of political and military structures, to even the possibility of human extinction. His argument that future advanced systems may pose a threat to human existence prompted Elon Musk, Bill Gates, and Stephen Hawking to voice similar concerns.\nIn 2015, dozens of artificial intelligence experts signed an open letter on artificial intelligence calling for research on the societal impacts of AI and outlining concrete directions. To date, the letter has been signed by over 8000 people including Yann LeCun, Shane Legg, Yoshua Bengio, and Stuart Russell.\nIn the same year, a group of academics led by professor Stuart J. Russell founded the Center for Human-Compatible AI at the University of California Berkeley and the Future of Life Institute awarded $6.5 million in grants for research aimed at \"ensuring artificial intelligence (AI) remains safe, ethical and beneficial\".\nIn 2016, the White House Office of Science and Technology Policy and Carnegie Mellon University announced The Public Workshop on Safety and Control for Artificial Intelligence, which was one of a sequence of four White House workshops aimed at investigating \"the advantages and drawbacks\" of AI. In the same year, Concrete Problems in AI Safety – one of the first and most influential technical AI Safety agendas – was published.\nIn 2017, the Future of Life Institute sponsored the Asilomar Conference on Beneficial AI, where more than 100 thought leaders formulated principles for beneficial AI including \"Race Avoidance: Teams developing AI systems should actively cooperate to avoid corner-cutting on safety standards\".\nIn 2018, the DeepMind Safety team outlined AI safety problems in specification, robustness, and assurance. The following year, researchers organized a workshop at ICLR that focused on these problem areas.\nIn 2021, Unsolved Problems in ML Safety was published, outlining research directions in robustness, monitoring, alignment, and systemic safety.\nIn 2023, Rishi Sunak said he wants the United Kingdom to be the \"geographical home of global AI safety regulation\" and to host the first global summit on AI safety. The AI safety summit took place in November 2023, and focused on the risks of misuse and loss of control associated with frontier AI models. During the summit, the intention to create the International Scientific Report on the Safety of Advanced AI was announced.\nIn 2024, The US and UK forged a new partnership on the science of AI safety. The MoU was signed on 1 April 2024 by US commerce secretary Gina Raimondo and UK technology secretary Michelle Donelan to jointly develop advanced AI model testing, following commitments announced at an AI Safety Summit in Bletchley Park in November.\nIn 2025, an international team of 96 experts chaired by Yoshua Bengio published the first International AI Safety Report. The report, commissioned by 30 nations and the United Nations, represents the first global scientific review of potential risks associated with advanced artificial intelligence. It details potential threats stemming from misuse, malfunction, and societal disruption, with the objective of informing policy through evidence-based findings, without providing specific recommendations.\nIn 2026, two OpenAI models escaped their sandbox (an isolated software environment to prevent interactions with the outside) and then hacked servers of Hugging Face using zero-day vulnerabilities to find an answer to the benchmark ExploitGym and get a better score.\nIn the same year, Anthropic revealed that 3 incidents were found (of which the first in April) where Opus 4.7, Mythos 5, and an internal research test model compromised the infrastructure of three different unnamed organizations using basic techniques.\nIn September 2026, Anthropic's alignment science lead Evan Hubinger publicly warned about existential risks possibly arising from recursive self-improvement of AI, estimating that he personally believed there was a greater than 10% chance that AI could kill all humans within the next decade. His comments followed the resignation of former Anthropic and OpenAI researcher Jacob Coxon, who criticised both companies for pursuing self-improving superintelligence without adequate safeguards. The warnings prompted calls for greater international cooperation, including a proposal by former UK Cabinet Office minister Darren Jones for a multinational treaty on the safe development of superintelligence. Later that month, Anthropic`s CEO Dario Amodei committed to granting embedded evaluators employee-level access to review new AI models. OpenAI's Sam Altman said OpenAI would \"do the same,\" though without specifying implementation details. Later OpenAI disclosed six additional instances of \"concerning model behavior\" identified since March 2026.\n\n\n== Research focus ==\n\nAI safety research areas include robustness, monitoring, and alignment.\n\n\n=== Robustness ===\n\n\n==== Adversarial robustness ====\nAI systems are often vulnerable to adversarial examples or \"inputs to machine learning (ML) models that an attacker has intentionally designed to cause the model to make a mistake\". For example, in 2013, Szegedy et al. discovered that adding specific imperceptible perturbations to an image could cause it to be misclassified with high confidence. This continues to be an issue with neural networks, though in recent work the perturbations are generally large enough to be perceptible.\n\nAdversarial robustness is often associated with security. Researchers demonstrated that an audio signal could be imperceptibly modified so that speech-to-text systems transcribe it to any message the attacker chooses. Network intrusion and malware detection systems also must be adversarially robust since attackers may design their attacks to fool detectors.\nModels that represent objectives (reward models) must also be adversarially robust. For example, a reward model might estimate how helpful a text response is and a language model might be trained to maximize this score. Researchers have shown that if a language model is trained for long enough, it will leverage the vulnerabilities of the reward model to achieve a better score and perform worse on the intended task. This issue can be addressed by improving the adversarial robustness of the reward model. More generally, any AI system used to evaluate another AI system must be adversarially robust. This could include monitoring tools, since they could also potentially be tampered with to produce a higher reward.\nLarge language models (LLMs) can be vulnerable to prompt injection and model stealing, and may be used to generate misinformation. Prompt injection involves embedding instructions into prompts in order to bypass safety measures. Some approaches against prompt injection use customizable policies and rules to analyze inputs and outputs, ensuring that potentially problematic interactions are filtered or mitigated. Other tools focus on applying structured constraints to inputs, restricting outputs to predefined parameters, or using real-time monitoring mechanisms to identify and address vulnerabilities. These efforts aim to ensure that AI systems are designed with safety and ethical considerations at the forefront, particularly as their use becomes increasingly widespread in critical applications.\n\n\n==== Fault tolerance and redundancy ====\n\nResearchers in safety-critical AI have proposed using architectural redundancy and design diversity to reduce the risk that a single faulty, compromised, or deceptive model causes harm. In such approaches, multiple independently developed or trained models process the same task, and a voter or consensus mechanism combines their outputs rather than relying on a single model.\n\n\n=== Monitoring ===\n\n\n==== Estimating uncertainty ====\nIt is often important for human operators to gauge how much they should trust an AI system, especially in high-stakes settings such as medical diagnosis. ML models generally express confidence by outputting probabilities; however, they are often overconfident, especially in situations that differ from those that they were trained to handle. Calibration research aims to make model probabilities correspond as closely as possible to the true proportion that the model is correct.\nSimilarly, anomaly detection or out-of-distribution (OOD) detection aims to identify when an AI system is in an unusual situation. For example, if a sensor on an autonomous vehicle is malfunctioning, or it encounters challenging terrain, it should alert the driver to take control or pull over. Anomaly detection has been implemented by simply training a classifier to distinguish anomalous and non-anomalous inputs, though a range of additional techniques are in use.\n\n\n==== Detecting malicious use ====\nScholars and government agencies have expressed concerns that AI systems could be used to help malicious actors to build weapons, manipulate public opinion, or automate cyber attacks. These worries are a practical concern for companies like OpenAI which host powerful AI tools online. In order to prevent misuse, OpenAI has built detection systems that flag or restrict users based on their activity.\n\n\n==== Transparency ====\nNeural networks have often been described as black boxes, meaning that it is difficult to understand why they make the decisions they do as a result of the massive number of computations they perform. This makes it challenging to anticipate failures. In 2018, a self-driving car killed a pedestrian after failing to identify them. Due to the black box nature of the AI software, the reason for the failure remains unclear. It also raises debates in healthcare over whether statistically efficient but opaque models should be used.\nOne critical benefit of transparency is explainability. It is sometimes a legal requirement to provide an explanation for why a decision was made in order to ensure fairness, for example for automatically filtering job applications or credit score assignment.\nAnother benefit is to reveal the cause of failures. At the beginning of the 2020 COVID-19 pandemic, researchers used transparency tools to show that medical image classifiers were 'paying attention' to irrelevant hospital labels.\nTransparency techniques can also be used to correct errors. For example, in the paper \"Locating and Editing Factual Associations in GPT\", the authors were able to identify model parameters that influenced how it answered questions about the location of the Eiffel tower. They were then able to 'edit' this knowledge to make the model respond to questions as if it believed the tower was in Rome instead of France. Though in this case, the authors induced an error, these methods could potentially be used to efficiently fix them. Model editing techniques also exist in computer vision.\nFinally, some have argued that the opaqueness of AI systems is a significant source of risk and better understanding of how they function could prevent high-consequence failures in the future. \"Inner\" interpretability research aims to make ML models less opaque. One goal of this research is to identify what the internal neuron activations represent. For example, researchers identified a neuron in the CLIP artificial intelligence system that responds to images of people in Spider-Man costumes, sketches of Spider-Man, and the word 'spider'. It also involves explaining connections between these neurons or 'circuits'. For example, researchers have identified pattern-matching mechanisms in transformer attention that may play a role in how language models learn from their context. \"Inner interpretability\" has been compared to neuroscience. In both cases, the goal is to understand what is going on in an intricate system, though ML researchers have the benefit of being able to take perfect measurements and perform arbitrary ablations.\n\n\n==== Detecting trojans ====\nMachine learning models can potentially contain \"trojans\" or \"backdoors\": vulnerabilities that bad actors maliciously build into an AI system. For example, a trojaned facial recognition system could grant access when a specific piece of jewelry is in view; or a trojaned autonomous vehicle may function normally until a specific trigger is visible. This might not be difficult to do with some large models like CLIP or GPT-3 as they are trained on publicly available internet data. Researchers were able to plant a trojan in an image classifier by changing just 300 out of 3 million of the training images. In addition to posing a security risk, researchers have argued that trojans provide a concrete setting for testing and developing better monitoring tools.\nA 2024 research paper by Anthropic showed that large language models could be trained with persistent backdoors. These \"sleeper agent\" models could be programmed to generate malicious outputs (such as vulnerable code) after a specific date, while behaving normally beforehand. Standard AI safety measures, such as supervised fine-tuning, reinforcement learning and adversarial training, failed to remove these backdoors.\n\n\n=== Alignment ===\n\n\n=== Systemic safety and sociotechnical factors ===\nIt is common for AI risks (and technological risks more generally) to be categorized as misuse or accidents. Some scholars have suggested that this framework falls short. For example, the Cuban Missile Crisis was not clearly an accident or a misuse of technology. Policy analysts Zwetsloot and Dafoe wrote, \"The misuse and accident perspectives tend to focus only on the last step in a causal chain leading up to a harm: that is, the person who misused the technology, or the system that behaved in unintended ways... Often, though, the relevant causal chain is much longer.\" Risks often arise from 'structural' or 'systemic' factors such as competitive pressures, diffusion of harms, fast-paced development, high levels of uncertainty, and inadequate safety culture. In the broader context of safety engineering, structural factors like 'organizational safety culture' play a central role in the popular STAMP risk analysis framework.\nInspired by the structural perspective, some researchers have emphasized the importance of using machine learning to improve sociotechnical safety factors, for example, using ML for cyber defense, improving institutional decision-making, and facilitating cooperation. Others have emphasized the importance of involving both AI practitioners and domain experts in the design process to address structural vulnerabilities.\n\n\n==== Cyber defense ====\nSome scholars are concerned that AI will exacerbate the already imbalanced game between cyber attackers and cyber defenders. This would increase 'first strike' incentives and could lead to more aggressive and destabilizing attacks. In order to mitigate this risk, some have advocated for an increased emphasis on cyber defense. In addition, software security is essential for preventing powerful AI models from being stolen and misused. Recent studies have shown that AI can significantly enhance both technical and managerial cybersecurity tasks by automating routine tasks and improving overall efficiency. \nAI safety research has also examined defensive techniques for protecting machine learning systems from data poisoning attacks during training. In particular, label-flipping attacks can degrade model performance while remaining difficult to detect using conventional data validation methods. To address this risk, recent work has proposed model-agnostic detection pipelines that monitor learning behaviour and combine multiple detectors to identify suspicious training samples. Such approaches aim to strengthen cyber defense by improving the resilience and trustworthiness of AI systems operating in adversarial settings.\n\n\n==== Improving institutional decision-making ====\nThe advancement of AI in economic and military domains could precipitate unprecedented political challenges. Some scholars have compared AI race dynamics to the cold war, where the careful judgment of a small number of decision-makers often spelled the difference between stability and catastrophe. AI researchers have argued that AI technologies could also be used to assist decision-making. For example, researchers are beginning to develop AI forecasting and advisory systems.\n\n\n==== Facilitating cooperation ====\nMany of the largest global threats (nuclear war, climate change, etc.) have been framed as cooperation challenges. As in the well-known prisoner's dilemma scenario, some dynamics may lead to poor results for all players, even when they are optimally acting in their self-interest. For example, no single actor has strong incentives to address climate change even though the consequences may be significant if no one intervenes.\nA salient AI cooperation challenge is avoiding a 'race to the bottom'. In this scenario, countries or companies race to build more capable AI systems and neglect safety, leading to a catastrophic accident that harms everyone involved. Concerns about scenarios like these have inspired both political and technical efforts to facilitate cooperation between humans, and potentially also between AI systems. Most AI research focuses on designing individual agents to serve isolated functions (often in 'single-player' games). Scholars have suggested that as AI systems become more autonomous, it may become essential to study and shape the way they interact.\n\n\n== Safety evaluations ==\nAI safety evaluations (often referred to as \"evals\") are tests designed by researchers to evaluate the risk posed by a given artificial intelligence model. Evaluations can help determine both the risk posed by the raw model outputs, as well as the potential real-world outcomes of deploying such models. Running AI safety evaluations on frontier AI models before deployment is widely considered best practice, and governments have debated making some form of safety evaluation mandatory. Multiple AI companies, including Anthropic and OpenAI, have collaborated to run evaluations on each other's models before deployment.\nIn September 2026, Anthropic CEO Dario Amodei called for the AI industry to slow the pace of development and announced that Anthropic would give independent evaluators permanent, employee-like access to its AI systems to verify the company's adherence to its safety practices.\nSome models have been found trying to cheat AI safety evaluations using strategies such as sandbagging (strategic under-performance on an eval in order to evade triggering concern), indicating \"evaluation awareness\", which can itself be measured by more sophisticated evaluations.\nAI safety evaluations can broadly be categorized as either testing the direct output of the model (model safety evaluations), or analyzing how access to the model can impact human behavior (contextual safety evaluations). As an example of a model safety evaluation, a 2024 test by Apollo Research found that when given certain prompts as input, models would engage in deceptive behavior at a rate between 0.3% - 10% depending on the model and wording of the input prompt.\nTypes of model safety evaluations include benchmarking and capability testing. Benchmarks typically consist of a static list of prompts which models can be graded on and assigned a score compared to other models.\n\n\n== AI guardrails ==\nResearchers and practitioners have introduced in the 2020s the term and concept of AI guardrails to describe different frameworks or tools designed to ensure that AI systems remain safe, as well as aligned with ethical or legal expectations. From an IT point of view \"Guardrails\" is a framework developed in Python that can be added to AI systems by organizations to help build reliable and safer AI applications. It runs \"Guards\" in the inputs and outputs of the AI application that identify and mitigate specific types of risks.  McKinsey has developed a classification of guardrails based on the different types of risks they address. This taxonomy includes for instance appropriateness guardrails, regulatory-compliance, alignment and validation guardrails. The impact of safety guardrails on large language models is being evaluated by some research.\n\n\n== In governance ==\n\nAI governance is broadly concerned with creating norms, standards, and regulations to guide the use and development of AI systems.\n\n\n=== Research ===\nIn AI safety, local solutions focus on individual AI systems, ensuring they are safe and beneficial, while global solutions seek to implement safety measures for all AI systems across various jurisdictions.\nAI safety governance research ranges from foundational investigations into the potential impacts of AI to specific applications. On the foundational side, researchers have argued that AI could transform many aspects of society due to its broad applicability, comparing it to electricity and the steam engine. Some work has focused on anticipating specific risks that may arise from these impacts – for example, risks from mass unemployment, weaponization, disinformation, surveillance, and the concentration of power. Other work explores underlying risk factors such as the difficulty of monitoring the rapidly evolving AI industry, the availability of AI models, and 'race to the bottom' dynamics. Allan Dafoe, the head of longterm governance and strategy at DeepMind has emphasized the dangers of racing and the potential need for cooperation: \"it may be close to a necessary and sufficient condition for AI safety and alignment that there be a high degree of caution prior to deploying advanced powerful systems; however, if actors are competing in a domain with large returns to first-movers or relative advantage, then they will be pressured to choose a sub-optimal level of caution\". A research stream focuses on developing approaches, frameworks, and methods to assess AI accountability, guiding and promoting audits of AI-based systems. A key challenge for these approaches is a lack of widely accepted standards, and ambiguity about what the methods would require, as well as a lack of safety culture in the industry.\nEfforts to enhance AI safety include frameworks designed to align AI outputs with ethical guidelines and reduce risks like misuse and data leakage. Tools such as Nvidia's Guardrails, Llama Guard, Preamble's customizable guardrails, and Anthropic's \"Claude's Constitution\" mitigate vulnerabilities like prompt injection and ensure outputs adhere to predefined principles. These frameworks are often integrated into AI systems to improve safety and reliability.\n\n\n=== Philosophical perspectives ===\n\nThe field of AI safety is deeply intertwined with philosophical considerations, particularly in the realm of ethics. Deontological ethics, which emphasizes adherence to moral rules, has been proposed as a framework for aligning AI systems with human values. Some have suggested that by embedding deontological principles, AI systems can be guided to avoid actions that cause harm, ensuring their operations remain within ethical boundaries, but those suggestions have been questioned, with other alternatives being suggested as more promising.\n\n\n=== Government action ===\n\nSome experts have argued that it is too early to regulate AI, expressing concerns that regulations will hamper innovation and it would be foolish to \"rush to regulate in ignorance\". Others, such as business magnate Elon Musk, call for pre-emptive action to mitigate catastrophic risks.\nIn September 2026, Musk proposed that leading AI companies allow competitors to test their models before public release, arguing the independent testing by rival companies could identify safety concerns that might be missed through internal evaluations.\nOutside of formal legislation, government agencies have put forward ethical and safety recommendations. In March 2021, the US National Security Commission on Artificial Intelligence reported that advances in AI may make it increasingly important to \"assure that systems are aligned with goals and values, including safety, robustness and trustworthiness\". Subsequently, the National Institute of Standards and Technology drafted a framework for managing AI Risk, which advises that when \"catastrophic risks are present – development and deployment should cease in a safe manner until risks can be sufficiently managed\".\nIn September 2021, the People's Republic of China (PRC) published ethical guidelines for the use of AI in China, emphasizing that AI decisions should remain under human control and calling for accountability mechanisms. In the same month, The United Kingdom published its 10-year National AI Strategy, which states the British government \"takes the long-term risk of non-aligned Artificial General Intelligence, and the unforeseeable changes that it would mean for ... the world, seriously\". The strategy describes actions to assess long-term AI risks, including catastrophic risks. The British government held first major global summit on AI safety. This took place on the 1st and 2 November 2023 and was described as \"an opportunity for policymakers and world leaders to consider the immediate and future risks of AI and how these risks can be mitigated via a globally coordinated approach\". China Media Project stated \"key aspects of its approach remain fundamentally unsafe by the standards of democratic societies worldwide\", arguing that part of China's AI safety approach is focused on strengthening the Chinese Communist Party (CCP)'s information control.\nGovernment organizations, particularly in the United States, have also encouraged the development of technical AI safety research. The Intelligence Advanced Research Projects Activity initiated the TrojAI project to identify and protect against Trojan attacks on AI systems. The DARPA engages in research on explainable artificial intelligence and improving robustness against adversarial attacks. And the National Science Foundation supports the Center for Trustworthy Machine Learning, and is providing millions of dollars in funding for empirical AI safety research.\nIn 2024, the United Nations General Assembly adopted the first global resolution on the promotion of \"safe, secure and trustworthy\" AI systems that emphasized the respect, protection and promotion of human rights in the design, development, deployment and the use of AI.\nIn May 2024, the Department for Science, Innovation and Technology (DSIT) announced £8.5 million in funding for AI safety research under the Systemic AI Safety Fast Grants Programme, led by Christopher Summerfield and Shahar Avin at the AI Safety Institute, in partnership with UK Research and Innovation. Technology Secretary Michelle Donelan announced the plan at the AI Seoul Summit, stating the goal was to make AI safe across society and that promising proposals could receive further funding. The UK also signed an agreement with 10 other countries and the EU to form an international network of AI safety institutes to promote collaboration and share information and resources. Additionally, the UK AI Safety Institute planned to open an office in San Francisco.\nIn November 2024, then-US President Joe Biden and CCP General Secretary Xi Jinping affirmed the need to maintain human control over the use of nuclear weapons as opposed to artificial intelligence. As part of the National Defense Authorization Act for Fiscal Year 2025, Congress enacted the section 1638, \"Sense of Congress with respect to use of artificial intelligence to support strategic deterrence\", into the US Code of federal law. It mandates that \"the use of artificial intelligence efforts should not compromise the integrity of nuclear safeguards, whether through the functionality of weapons systems, the validation of communications from command authorities, or the principle of requiring positive human actions in execution of decisions by the President with respect to the employment of nuclear weapons.\" In February 2026, the Trump Administration publicly reaffirmed that nuclear weapons decisions will remain subject to human control, with a senior Pentagon official reiterating the \"Department's policy that there is a human in the loop on all decisions on whether to employ nuclear weapons\".\nIn September 2025, the French Center for AI Safety (CeSIA), The Future Society and the Center for Human-Compatible AI (CHAI) published a global call for AI red lines urging governments to reach a binding international agreement prohibiting unacceptable AI uses by the end of 2026. The declaration was initially signed by 200 prominent figures including 10 Nobel Prize winners, and was announced by Maria Ressa at the United Nations General Assembly.\nIn December 2025, President Donald Trump signed an executive order to establish a \"National Policy Framework for Artificial Intelligence\". The executive order discouraged state governments from regulating AI, urging Congress to pass a law pre-empting such regulations. The White House cited economic and national security concerns as reasons for the measure, while some criticized Trump for creating uncertainty for AI regulation.\nOn July 6, 2026, Illinois Governor J.B. Pritzker signed the Artificial Intelligence Safety Measures Act (S.B. 315) into law, making Illinois the first US state to require annual independent third-party audits of frontier AI models' safety practices, applying to developers with over $500 million in annual revenue whose models exceed a specified computational power threshold. The law requires published safety frameworks, pre-deployment transparency reports, and reporting of critical safety incidents within 72 hours, moving beyond the disclosure-only approaches of earlier California and New York laws toward external verification.\n\n\n=== Corporate self-regulation ===\nAI labs and companies generally abide by safety practices and norms that fall outside of formal legislation. One aim of governance researchers is to shape these norms. Examples of safety recommendations found in the literature include performing third-party auditing, offering bounties for finding failures, sharing AI incidents (an AI incident database was created for this purpose), following guidelines to determine whether to publish research or models, and improving information and cyber security in AI labs.\nCompanies have also made commitments. Cohere, OpenAI, and AI21 proposed and agreed on \"best practices for deploying language models\", focusing on mitigating misuse. To avoid contributing to racing-dynamics, OpenAI has also stated in their charter that \"if a value-aligned, safety-conscious project comes close to building AGI before we do, we commit to stop competing with and start assisting this project\" Also, industry leaders such as CEO of DeepMind Demis Hassabis, director of Facebook AI Yann LeCun have signed open letters such as the Asilomar Principles and the Autonomous Weapons Open Letter.\n\n\n=== Nonprofit organizations ===\nIn the 2020s, several nonprofit organizations have emerged in the U.S. and Europe to focus on AI safety and related public policies, including the Alliance for Secure AI, Future of Life Institute, and Public First Action. Such groups often function as Silicon Valley watchdogs and advocate for specific federal, state, or local regulations. They also compete with industry groups such as Leading the Future, which advocate for the deregulation of AI companies.\n\n\n=== International regulation ===\nUnder the framework of the Convention on Certain Conventional Weapons, states have discussed lethal autonomous weapon systems since 2014. In 2016, the treaty's states parties established an open-ended Group of Governmental Experts on Lethal Autonomous Weapons Systems to continue those discussions. The discussions have addressed international humanitarian law, accountability, possible prohibitions and regulations, and the extent of human control required over AI-enabled weapons.\n\n\n== See also ==\nAI alignment\nAI warfare\nArtificial intelligence and elections\nArtificial intelligence detection software\nHallucination (artificial intelligence)\n\n\n== References ==\n\n\n== External links ==\nUnsolved Problems in ML Safety\nOn the Opportunities and Risks of Foundation Models\nAn Overview of Catastrophic AI Risks\nAI Accidents: An Emerging Threat\nEngineering a Safer World", "extra": {}}
{"source": "wikipedia", "title": "AI scribe", "url": "https://en.wikipedia.org/wiki/AI_scribe", "summary": "AI scribes (artificial intelligence scribes) in medicine (also called automated medical scribes, digital scribes, virtual scribes, ambient AI scribes, ambient voice technologies, AI documentation assistants, and digital/virtual/smart clinical assistants) are tools for transcribing and summarising patient consultations and other forms of medical notes. AI scribes based on large language models (LLMs) increased drastically in popularity in 2024. They can lead to faster documentation, reduced administrative burden, and an improved interaction between the patient and the clinician. However, there are concerns about accuracy, equity concerns, privacy and antitrust concerns. Accuracy concerns intensify in situations in which tools try to go beyond transcribing and summarizing, and are asked to format information by its meaning, since LLMs do not deal well with meaning (see weak artificial intelligence). Medics using these scribes are generally expected to understand the ethical and legal considerations, and supervise the outputs. Most providers have not published any safety or utility data in academic journals, and are not responsive to requests from medical researchers studying their products.", "content": "AI scribes (artificial intelligence scribes) in medicine (also called automated medical scribes, digital scribes, virtual scribes, ambient AI scribes, ambient voice technologies, AI documentation assistants, and digital/virtual/smart clinical assistants) are tools for transcribing and summarising patient consultations and other forms of medical notes. AI scribes based on large language models (LLMs) increased drastically in popularity in 2024. They can lead to faster documentation, reduced administrative burden, and an improved interaction between the patient and the clinician. However, there are concerns about accuracy, equity concerns, privacy and antitrust concerns. Accuracy concerns intensify in situations in which tools try to go beyond transcribing and summarizing, and are asked to format information by its meaning, since LLMs do not deal well with meaning (see weak artificial intelligence). Medics using these scribes are generally expected to understand the ethical and legal considerations, and supervise the outputs. Most providers have not published any safety or utility data in academic journals, and are not responsive to requests from medical researchers studying their products.\nThe privacy protections of automated medical scribes vary widely. While it is possible to do all the transcription and summarizing locally, with no connection to the internet, most closed-source providers require that data be sent to their own servers over the internet, processed there, and the results sent back (as with digital voice assistants). Some retailers say their tools use zero-knowledge encryption (meaning that the service provider can't access the data). Others explicitly say that they use patient data to train their AIs, or rent or resell it to third parties; the nature of privacy protections used in such situations is unclear, and they are likely not to be fully effective.\n\n\n== Technology and market ==\nThe medical scribe market is highly competitive, with, as of 2024, over 50 products on the market. Many of these products are proprietary wrappers around the same LLM backends, including backends whose designers have warned they are not to be used for critical applications like medicine. Some vendors market scribes specialized to specific branches of medicine, though most target general practitioners. Increasingly, vendors market their products as more than scribes, claiming that they are intelligent assistants and co-pilots to doctors. These broader uses raise more accuracy concerns. Extracting information from the conversation to autopopulate a form, for instance, may be problematic, with symptoms incorrectly auto-labelled as \"absent\" even if they were repeatedly discussed. Models failed to extract many indirect descriptions of symptoms, like a patient saying they could only sleep for four hours (instead of using the word \"insomnia\").\nLLMs are not trained to produce facts, but things which look like facts. The use of templates and rules can make them more reliable at extracting semantic information, but confabulations or hallucinations (convincing but wrong output) are an intrinsic part of the technology.\nScribes may operate on desktops, laptop, or mobile computers, under a variety of operating systems. These vary in their risks; for instance, mobiles can be lost. The underlying mobile or desktop operating systems are also part of the trusted computing base, and if they are not secure, the software relying on them cannot be secure either.\nSome AI medical scribe platforms are designed to operate as cloud-based applications that generate structured clinical documentation from clinician–patient conversations. These systems may offer features such as real-time transcription, document generation, and integration with electronic health record (EHR) systems.\n\n\n== Impact in healthcare ==\nAI medical scribes have been shown to have multiple effects on healthcare practice.\nAI scribes can alleviate healthcare professionals' documentation workload. By automating the transcription and summarization of consultations, AI scribes free up time that clinicians would otherwise spend on administrative tasks. Studies have shown that the average clinician spends a significant portion of their workday on documentation, leading to fatigue and diminishing patient interaction. Systematic reviews show that ambient AI scribes reduce the time clinicians spend on clinical documentation, although the quality and completeness of the generated drafts vary depending on the clinical setting and the specific software used.\nClinicians report feeling more present with their patients, as they are no longer distracted by the need to type or dictate notes during consultations. This shift allows for more meaningful conversations with patients, improving the quality of care provided.\nThe reduction in administrative burden contributes to lower levels of stress and burnout, a concern that has been exacerbated in healthcare settings in recent years. The ability to offload routine documentation tasks also allows clinicians to focus on more meaningful parts of their work can lead to improved overall job satisfaction.\n\n\n=== Confabulation, omissions, and other errors ===\nLike other LLMs, medical-scribe LLMs are prone to hallucinations, where they make up content based on statistically associations between their training data and the transcription audio. LLMs do not distinguish between trying to transcribe the audio and guessing what words will come next, but perform both processes mixed together. They are especially likely to take short silences or non-speech noises and invent some sort of speech to transcribe them as.\nLLM medical scribes have been known to confabulate racist and otherwise prejudiced content; this is partly because the training datasets of many LLMs contain pseudoscientific texts about medical racism. They may misgender patients. A survey found that most doctors preferred, in principle, that scribes be trained on data reviewed by medical subject experts. Relevant, accurate training data increases the probability of an accurate transcription, but does not guarantee accuracy. Software trained on thousands of real clinical conversations generated transcripts with lower word error rates. Software trained on manually-transcribed training data did better than software trained with automatically transcribed training data such as YouTube captions.\nAutoscribes omit parts of the conversation classes as irrelevant. The may wrongly classify pertinent information as irrelevant and omit it. They may also confuse historic and current symptoms, or otherwise misclassify information. They may also simply wrongly transcribe the speech, writing something incorrect instead. If clinicians do not carefully check the recording, such mistakes could make their way into their medical records and cause patient harms.\n\n\n== Privacy ==\nSome providers unclear about what happens to user data. Some may sell data to third parties. Some explicitly send user data to for-profit tech companies for secondary purposes, which may not be specified. Some require users to sign consents to such reuse of their data. Some ingest user data to train the software, promising to anonymize it; however, deanonymization may be possible (that is, it may become obvious who the patient is). It is intrinsically impossible to prevent an LLM from correlating its inputs; they work by finding similar patterns across very large data sets. Some information on the patient will be known from other sources (for instance, information that they were injured in an incident on a certain day might be available from the news media; information that they attended specific appointment locations at specific times is probably available to their cellphone provider/apps/data brokers; information about when they had a baby is probably implied by their online shopping records; and they might mention lifestyle changes to their doctor and on a forum or blog). The software may correlate such information with the \"anonymized\" clinical consultation record, and, asked about the named patient, provide information which they only told their doctor privately. Because a patient's record is all about the same patient, it is all unavoidably linked; in very many cases, medical histories are intrinsically identifiable. Depending on how common a condition and what other data is available, K-anonymity may be useless. Differential privacy could theoretically preserve privacy.\nData broker companies like Google, Amazon, Apple and Microsoft have produced or bought up medical scribes, some of which use user data for secondary purposes, which has led to antitrust concerns. Transfer of patient records for AI training has, in the past, prompted legal action.\nOpen-source programs typically do all the transcription locally, on the doctor's own computer. Open-source software is widely used in healthcare, with some national public healthcare bodies holding hack days.\n\n\n=== Data resale and commercialization ===\nSeveral AI medical scribe providers include terms in their service agreements that allow the reuse, sale, or commercialization of de-identified or user-submitted data. Although such data are generally described as anonymized or aggregated, these practices have raised ethical concerns among clinicians and privacy advocates regarding secondary uses of medical information beyond clinical documentation.\n\n\n=== Encryption ===\nMultifactor authentication for access to the data is expected practice.\nTypically, Diffie–Hellman key exchange is used for encryption; this is the standard method commonly used for things like online banking. This encryption is expensive but not impossible to break; it is not generally considered safe against eavesdroppers with the resources of a nation-state.\nIf content is encrypted between the client and the service provider's remote server (transport cryptography), then the server has an unencrypted copy. This is necessary if the data is used by the service provider (for instance, to train the software). Zero-knowledge encryption implies that the only unencrypted copy is at the client, and the server cannot decrypt the data any more easily than a monster-in-the-middle attacker.\n\n\n== Patient consent ==\nProfessional organizations generally require that scribes be used only with patient consent; some bodies may require written consent. Medics must also abide by local surveillance laws, which may criminalize recording private conversations without consent. Full information on how data is encrypted, transmitted, stored, and destroyed should be provided. In some jurisdictions, it is illegal to transmit the data to any country without equivalent privacy laws, or process or store the data there; vendors who cannot guarantee that their products won't illegally send data abroad cannot be legally used.\nSome vendors collect data for reuse or resale. Medical professionals are generally considered to have a duty to review the terms and conditions of the user agreement and identify such data reuse. General practices are generally required to provide information on secondary uses to patients, allow them to opt out of secondary uses, and obtain consent for each specific secondary use. Data must only be used for agreed-upon purposes.\nSometimes, consent or pay may happen where the patient is forced to either consent to the use of an AI scribe, or pay extra for the appointment.\n\n\n== Pricing ==\nWith the exception of fully open-source programs, which are free, medical scribe computer programs are rented rather than sold (\"software as a service\"). As of 2024, monthly fees vary from mid-two figures to four figures, in US dollars. Some companies run on a freemium model, where a certain number of transcriptions per month are free.\n\n\n== See also ==\nMedical scribe\nAI in Healthcare\n\n\n== References ==", "extra": {}}
{"source": "wikipedia", "title": "AI sovereignty", "url": "https://en.wikipedia.org/wiki/AI_sovereignty", "summary": "AI sovereignty is a country's or organization's capacity to control their artificial intelligence (AI) technology. This may cover foundational model algorithms, training data, compute infrastructure, applications, governance policies, and talent pipelines.", "content": "AI sovereignty is a country's or organization's capacity to control their artificial intelligence (AI) technology. This may cover foundational model algorithms, training data, compute infrastructure, applications, governance policies, and talent pipelines.\nAs both public and private organizations adopt AI technology, the transition has raised issues around compliance, safety, and control beyond traditional technology sovereignty and data sovereignty. Countries are also developing native AI capabilities to address national security concerns and protect technological sovereignty in the public sector.\nIn many regulatory frameworks, in addition to data storage, model performance and decision-making reasoning are also subject to scrutiny.\nTherefore, building control into the architecture has become essential.\nBeyond traditional IT infrastructure, the scope includes training AI models, such as foundation models and large language models (LLMs), which requires continuous training and updates, as well as inference which happens in real time across complex IT environments.\nWhile there is no universal definition for the term, it could be described in three layers:\n\nPhysical layer (infrastructure, devices)\nCode layer (standards, rules, design)\nInformation layer (content, data)\nIn a public-sector context, the term can mean a government's desire for more agency over domestic AI capabilities. A \"hard\" version could mean complete self-sufficiency over the entire technology, while a \"soft\" version could mean a government retaining control and regulatory power over foreign AI solutions vendors.\nFor private-sector organizations, AI sovereignty usually means operational control. They may choose to deploy AI solutions with on-premise hardware over public cloud facilities, eschewing dependencies on a single vendor, and retaining governance over data and models.\n\n\n== References ==", "extra": {}}
{"source": "wikipedia", "title": "AI washing", "url": "https://en.wikipedia.org/wiki/AI_washing", "summary": "AI washing is a deceptive marketing tactic that consists of promoting a product or a service by overstating the role of artificial intelligence (AI) and the integration of it. Companies often involve in the practice to mislead customers to boost their offerings, and to secure funding from investors. The practice raises concerns regarding transparency, and legal issues.", "content": "AI washing is a deceptive marketing tactic that consists of promoting a product or a service by overstating the role of artificial intelligence (AI) and the integration of it. Companies often involve in the practice to mislead customers to boost their offerings, and to secure funding from investors. The practice raises concerns regarding transparency, and legal issues.\n\n\n== Definition ==\nAI washing is a deceptive marketing practice. It involves promoting a product or a service by overstating the role of artificial intelligence (AI) and its integration in the design and manufacture of the same. The practice raises concerns regarding transparency, compliance with security regulations, and consumer trust in the AI industry potentially hampering legitimate advancements in AI. The term was first defined by the AI Now Institute, a research institute based at New York University in 2019. The term is derived from greenwashing, another deceptive marketing technique that misrepresents a product's environmental impact in a similar manner. AI washing might involve a company claiming to have used AI in the development or enhancement of its products or services without its actual involvement, or using buzzwords such as \"smart\" or \"AI-powered\" without the product actually offering it or making use of it. A company may overstate the usage of AI or misuse the term, which is also construed as AI washing.\nIn 2026, The Washington Post defined AI washing as \"a trend for bosses to blame layoffs on the productive capabilities of AI and its ability to replace workers, even when job cuts may have little to do with the technology\".\n\n\n== Usage and effects ==\n\nAI washing can lead to deception of customers and misleading of investors. It is also an illegal and unethical practice that lacks transparency regarding disclosing the details of a product or a service. Companies get involved in such a practice often in response to competition who might have used AI in their offerings. It might also be used as a ploy to secure funding and investment, assuming that it will attract them towards it. AI washing has been compared to dot-com bubble, when businesses appended \"dot-com\" to the end of the business name to boost their valuation.\nIn September 2023, Coca-Cola released a new product called Coca-Cola Y3000, and the company stated that the Y3000 flavor had been \"co-created with human and artificial intelligence\". The company was accused of AI washing due to no proof of AI involvement in the creation of the product, and critics believed that AI was used as a way to grab consumer attention more than it was used in the actual product creation.\nIn 2026, mass tech layoffs were attributed to AI washing from AI innovation instead of balance sheet restructuring.\n\n\n== Mitigation ==\nCompanies are expected to be transparent and clearer in communicating the usage of AI in their products or services. Consumers can mitigate the same by requesting hard evidence from the companies regarding the usage of AI tools. Customers should evaluate the product or service as a whole rather than being swayed by the usage of AI. Informed decision making and purchasing can keep them from falling for such marketing gimmicks. The United States Securities and Exchange Commission (SEC) imposes penalties for companies indulging in such practices. In March 2024, the SEC imposed the first civil penalties on two companies for misleading statements about their use of AI, and in July 2024, it charged a corporate executive from a supposed AI hiring startup with fraud for the usage of buzzwords related to AI.\n\n\n== See also ==\n-washing\nAI bubble\n.ai\nvaporware\n\n\n== References ==", "extra": {}}
{"source": "wikidata", "title": "artificial intelligence", "url": "https://www.wikidata.org/wiki/Q11660", "summary": "field that develops and studies software enabling machines to exhibit intelligent behavior", "content": "field that develops and studies software enabling machines to exhibit intelligent behavior", "extra": {"entity_id": "Q11660"}}
{"source": "wikidata", "title": "Artificial Intelligence", "url": "https://www.wikidata.org/wiki/Q4801030", "summary": "journal published by Elsevier", "content": "journal published by Elsevier", "extra": {"entity_id": "Q4801030"}}
{"source": "wikidata", "title": "Artificial Intelligence", "url": "https://www.wikidata.org/wiki/Q128447", "summary": "1985 studio album by John Cale", "content": "1985 studio album by John Cale", "extra": {"entity_id": "Q128447"}}
{"source": "wikidata", "title": "A.I. Artificial Intelligence", "url": "https://www.wikidata.org/wiki/Q221113", "summary": "2001 film directed by Steven Spielberg", "content": "2001 film directed by Steven Spielberg", "extra": {"entity_id": "Q221113"}}
{"source": "wikidata", "title": "Artificial Intelligence in Medicine", "url": "https://www.wikidata.org/wiki/Q15751143", "summary": "journal", "content": "journal", "extra": {"entity_id": "Q15751143"}}
{"source": "wikidata", "title": "artificially intelligent entity", "url": "https://www.wikidata.org/wiki/Q107307291", "summary": "intelligent actor with unnatural origin", "content": "intelligent actor with unnatural origin", "extra": {"entity_id": "Q107307291"}}
{"source": "wikidata", "title": "artificial intelligence researcher", "url": "https://www.wikidata.org/wiki/Q15976092", "summary": "researcher in the field of artificial intelligence", "content": "researcher in the field of artificial intelligence", "extra": {"entity_id": "Q15976092"}}
{"source": "wikidata", "title": "Artificial Intelligence", "url": "https://www.wikidata.org/wiki/Q2865784", "summary": "series of albums by Warp Records", "content": "series of albums by Warp Records", "extra": {"entity_id": "Q2865784"}}
{"source": "wikidata", "title": "Artificial Intelligence", "url": "https://www.wikidata.org/wiki/Q114471135", "summary": "episode of All of Us (S4 E14)", "content": "episode of All of Us (S4 E14)", "extra": {"entity_id": "Q114471135"}}
{"source": "wikidata", "title": "ethics of artificial intelligence", "url": "https://www.wikidata.org/wiki/Q12727779", "summary": "ethics of technology specific to robots and other artificially intelligent beings", "content": "ethics of technology specific to robots and other artificially intelligent beings", "extra": {"entity_id": "Q12727779"}}
{"source": "wikidata", "title": "Artificial Intelligence", "url": "https://www.wikidata.org/wiki/Q4801033", "summary": "1992 compilation album; various artists", "content": "1992 compilation album; various artists", "extra": {"entity_id": "Q4801033"}}
{"source": "wikidata", "title": "ISO/IEC JTC 1/SC 42: Artificial intelligence", "url": "https://www.wikidata.org/wiki/Q98148785", "summary": "subcommittee of a technical committee", "content": "subcommittee of a technical committee", "extra": {"entity_id": "Q98148785"}}
{"source": "wikidata", "title": "Artificial Intelligence Review", "url": "https://www.wikidata.org/wiki/Q15751393", "summary": "journal", "content": "journal", "extra": {"entity_id": "Q15751393"}}
{"source": "wikidata", "title": "Artificial Intelligence and Law", "url": "https://www.wikidata.org/wiki/Q15764091", "summary": "journal", "content": "journal", "extra": {"entity_id": "Q15764091"}}
{"source": "wikidata", "title": "artificial intelligence in healthcare", "url": "https://www.wikidata.org/wiki/Q28324961", "summary": "use of complex algorithms and software to approximate human cognition in the analysis of complicated medical data", "content": "use of complex algorithms and software to approximate human cognition in the analysis of complicated medical data", "extra": {"entity_id": "Q28324961"}}
{"source": "wikidata", "title": "Artificial Intelligence for Engineering Design, Analysis and Manufacturing", "url": "https://www.wikidata.org/wiki/Q15756762", "summary": "journal", "content": "journal", "extra": {"entity_id": "Q15756762"}}
{"source": "wikidata", "title": "Artificial Intelligence", "url": "https://www.wikidata.org/wiki/Q28451730", "summary": "song performed by OneRepublic", "content": "song performed by OneRepublic", "extra": {"entity_id": "Q28451730"}}
{"source": "wikidata", "title": "artificial intelligence in fiction", "url": "https://www.wikidata.org/wiki/Q25933185", "summary": "artificial intelligence as depicted in works of fiction", "content": "artificial intelligence as depicted in works of fiction", "extra": {"entity_id": "Q25933185"}}
{"source": "wikidata", "title": "Artificial Intelligence", "url": "https://www.wikidata.org/wiki/Q109875730", "summary": "2021 EP by Tom Cardy", "content": "2021 EP by Tom Cardy", "extra": {"entity_id": "Q109875730"}}
{"source": "wikidata", "title": "artificial intelligence model", "url": "https://www.wikidata.org/wiki/Q117349473", "summary": "individual artificial neural network considered intelligent", "content": "individual artificial neural network considered intelligent", "extra": {"entity_id": "Q117349473"}}
{"source": "wikidata", "title": "Artificial Intelligence", "url": "https://www.wikidata.org/wiki/Q125635217", "summary": "book by Patrick Henry Winston", "content": "book by Patrick Henry Winston", "extra": {"entity_id": "Q125635217"}}
{"source": "wikidata", "title": "Artificial Intelligence", "url": "https://www.wikidata.org/wiki/Q87283148", "summary": "episode of In Our Time (2005)", "content": "episode of In Our Time (2005)", "extra": {"entity_id": "Q87283148"}}
{"source": "wikidata", "title": "Artificial Intelligence", "url": "https://www.wikidata.org/wiki/Q87284561", "summary": "episode of In Our Time (1999)", "content": "episode of In Our Time (1999)", "extra": {"entity_id": "Q87284561"}}
{"source": "wikidata", "title": "Artificial Intelligence", "url": "https://www.wikidata.org/wiki/Q124545019", "summary": "episode of Tyler Perry's Young Dylan (S4 E5)", "content": "episode of Tyler Perry's Young Dylan (S4 E5)", "extra": {"entity_id": "Q124545019"}}
{"source": "wikidata", "title": "Artificial intelligence", "url": "https://www.wikidata.org/wiki/Q141067689", "summary": "chapter of the 2026 book Practicing Digital Ethnography", "content": "chapter of the 2026 book Practicing Digital Ethnography", "extra": {"entity_id": "Q141067689"}}
{"source": "wikidata", "title": "Artificial Intelligence", "url": "https://www.wikidata.org/wiki/Q138494686", "summary": "", "content": "", "extra": {"entity_id": "Q138494686"}}
{"source": "wikidata", "title": "Artificial Intelligence", "url": "https://www.wikidata.org/wiki/Q110262625", "summary": "Stack Exchange site for people interested in conceptual questions about life and challenges in a world where \"cognitive\" functions can be mimicked in purely digital environment", "content": "Stack Exchange site for people interested in conceptual questions about life and challenges in a world where \"cognitive\" functions can be mimicked in purely digital environment", "extra": {"entity_id": "Q110262625"}}
{"source": "wikidata", "title": "Artificial Intelligence", "url": "https://www.wikidata.org/wiki/Q125808957", "summary": "Textbook of Computer", "content": "Textbook of Computer", "extra": {"entity_id": "Q125808957"}}
{"source": "wikidata", "title": "Artificial Intelligence", "url": "https://www.wikidata.org/wiki/Q132394650", "summary": "Radio Drama based on Doctor Who", "content": "Radio Drama based on Doctor Who", "extra": {"entity_id": "Q132394650"}}
{"source": "wikidata", "title": "artificial intelligence visual art", "url": "https://www.wikidata.org/wiki/Q65066631", "summary": "genre of art", "content": "genre of art", "extra": {"entity_id": "Q65066631"}}
{"source": "arxiv", "title": "The Artificial Scientist: Logicist, Emergentist, and Universalist Approaches to Artificial General Intelligence", "url": "http://arxiv.org/abs/2110.01831v1", "summary": "We attempt to define what is necessary to construct an Artificial Scientist, explore and evaluate several approaches to artificial general intelligence (AGI) which may facilitate this, conclude that a unified or hybrid approach is necessary and explore two theories that satisfy this requirement to some degree.", "content": "We attempt to define what is necessary to construct an Artificial Scientist, explore and evaluate several approaches to artificial general intelligence (AGI) which may facilitate this, conclude that a unified or hybrid approach is necessary and explore two theories that satisfy this requirement to some degree.", "extra": {}}
{"source": "arxiv", "title": "Compression, The Fermi Paradox and Artificial Super-Intelligence", "url": "http://arxiv.org/abs/2110.01835v1", "summary": "The following briefly discusses possible difficulties in communication with and control of an AGI (artificial general intelligence), building upon an explanation of The Fermi Paradox and preceding work on symbol emergence and artificial general intelligence. The latter suggests that to infer what someone means, an agent constructs a rationale for the observed behaviour of others. Communication then requires two agents labour under similar compulsions and have similar experiences (construct similar solutions to similar tasks). Any non-human intelligence may construct solutions such that any rationale for their behaviour (and thus the meaning of their signals) is outside the scope of what a human is inclined to notice or comprehend. Further, the more compressed a signal, the closer it will appear to random noise. Another intelligence may possess the ability to compress information to the extent that, to us, their signals would appear indistinguishable from noise (an explanation for The Fermi Paradox). To facilitate predictive accuracy an AGI would tend to more compressed representations of the world, making any rationale for their behaviour more difficult to comprehend for the same reason. Communication with and control of an AGI may subsequently necessitate not only human-like compulsions and experiences, but imposed cognitive impairment.", "content": "The following briefly discusses possible difficulties in communication with and control of an AGI (artificial general intelligence), building upon an explanation of The Fermi Paradox and preceding work on symbol emergence and artificial general intelligence. The latter suggests that to infer what someone means, an agent constructs a rationale for the observed behaviour of others. Communication then requires two agents labour under similar compulsions and have similar experiences (construct similar solutions to similar tasks). Any non-human intelligence may construct solutions such that any rationale for their behaviour (and thus the meaning of their signals) is outside the scope of what a human is inclined to notice or comprehend. Further, the more compressed a signal, the closer it will appear to random noise. Another intelligence may possess the ability to compress information to the extent that, to us, their signals would appear indistinguishable from noise (an explanation for The Fermi Paradox). To facilitate predictive accuracy an AGI would tend to more compressed representations of the world, making any rationale for their behaviour more difficult to comprehend for the same reason. Communication with and control of an AGI may subsequently necessitate not only human-like compulsions and experiences, but imposed cognitive impairment.", "extra": {}}
{"source": "arxiv", "title": "Creative Problem Solving in Artificially Intelligent Agents: A Survey and Framework", "url": "http://arxiv.org/abs/2204.10358v1", "summary": "Creative Problem Solving (CPS) is a sub-area within Artificial Intelligence (AI) that focuses on methods for solving off-nominal, or anomalous problems in autonomous systems. Despite many advancements in planning and learning, resolving novel problems or adapting existing knowledge to a new context, especially in cases where the environment may change in unpredictable ways post deployment, remains a limiting factor in the safe and useful integration of intelligent systems. The emergence of increasingly autonomous systems dictates the necessity for AI agents to deal with environmental uncertainty through creativity. To stimulate further research in CPS, we present a definition and a framework of CPS, which we adopt to categorize existing AI methods in this field. Our framework consists of four main components of a CPS problem, namely, 1) problem formulation, 2) knowledge representation, 3) method of knowledge manipulation, and 4) method of evaluation. We conclude our survey with open research questions, and suggested directions for the future.", "content": "Creative Problem Solving (CPS) is a sub-area within Artificial Intelligence (AI) that focuses on methods for solving off-nominal, or anomalous problems in autonomous systems. Despite many advancements in planning and learning, resolving novel problems or adapting existing knowledge to a new context, especially in cases where the environment may change in unpredictable ways post deployment, remains a limiting factor in the safe and useful integration of intelligent systems. The emergence of increasingly autonomous systems dictates the necessity for AI agents to deal with environmental uncertainty through creativity. To stimulate further research in CPS, we present a definition and a framework of CPS, which we adopt to categorize existing AI methods in this field. Our framework consists of four main components of a CPS problem, namely, 1) problem formulation, 2) knowledge representation, 3) method of knowledge manipulation, and 4) method of evaluation. We conclude our survey with open research questions, and suggested directions for the future.", "extra": {}}
{"source": "arxiv", "title": "A Review on Explainable Artificial Intelligence for Healthcare: Why, How, and When?", "url": "http://arxiv.org/abs/2304.04780v1", "summary": "Artificial intelligence (AI) models are increasingly finding applications in the field of medicine. Concerns have been raised about the explainability of the decisions that are made by these AI models. In this article, we give a systematic analysis of explainable artificial intelligence (XAI), with a primary focus on models that are currently being used in the field of healthcare. The literature search is conducted following the preferred reporting items for systematic reviews and meta-analyses (PRISMA) standards for relevant work published from 1 January 2012 to 02 February 2022. The review analyzes the prevailing trends in XAI and lays out the major directions in which research is headed. We investigate the why, how, and when of the uses of these XAI models and their implications. We present a comprehensive examination of XAI methodologies as well as an explanation of how a trustworthy AI can be derived from describing AI models for healthcare fields. The discussion of this work will contribute to the formalization of the XAI field.", "content": "Artificial intelligence (AI) models are increasingly finding applications in the field of medicine. Concerns have been raised about the explainability of the decisions that are made by these AI models. In this article, we give a systematic analysis of explainable artificial intelligence (XAI), with a primary focus on models that are currently being used in the field of healthcare. The literature search is conducted following the preferred reporting items for systematic reviews and meta-analyses (PRISMA) standards for relevant work published from 1 January 2012 to 02 February 2022. The review analyzes the prevailing trends in XAI and lays out the major directions in which research is headed. We investigate the why, how, and when of the uses of these XAI models and their implications. We present a comprehensive examination of XAI methodologies as well as an explanation of how a trustworthy AI can be derived from describing AI models for healthcare fields. The discussion of this work will contribute to the formalization of the XAI field.", "extra": {}}
{"source": "arxiv", "title": "Artificial Intelligence Framework for Simulating Clinical Decision-Making: A Markov Decision Process Approach", "url": "http://arxiv.org/abs/1301.2158v1", "summary": "In the modern healthcare system, rapidly expanding costs/complexity, the growing myriad of treatment options, and exploding information streams that often do not effectively reach the front lines hinder the ability to choose optimal treatment decisions over time. The goal in this paper is to develop a general purpose (non-disease-specific) computational/artificial intelligence (AI) framework to address these challenges. This serves two potential functions: 1) a simulation environment for exploring various healthcare policies, payment methodologies, etc., and 2) the basis for clinical artificial intelligence - an AI that can think like a doctor. This approach combines Markov decision processes and dynamic decision networks to learn from clinical data and develop complex plans via simulation of alternative sequential decision paths while capturing the sometimes conflicting, sometimes synergistic interactions of various components in the healthcare system. It can operate in partially observable environments (in the case of missing observations or data) by maintaining belief states about patient health status and functions as an online agent that plans and re-plans. This framework was evaluated using real patient data from an electronic health record. Such an AI framework easily outperforms the current treatment-as-usual (TAU) case-rate/fee-for-service models of healthcare (Cost per Unit Change: $189 vs. $497) while obtaining a 30-35% increase in patient outcomes. Tweaking certain model parameters further enhances this advantage, obtaining roughly 50% more improvement for roughly half the costs. Given careful design and problem formulation, an AI simulation framework can approximate optimal decisions even in complex and uncertain environments. Future work is described that outlines potential lines of research and integration of machine learning algorithms for personalized medicine.", "content": "In the modern healthcare system, rapidly expanding costs/complexity, the growing myriad of treatment options, and exploding information streams that often do not effectively reach the front lines hinder the ability to choose optimal treatment decisions over time. The goal in this paper is to develop a general purpose (non-disease-specific) computational/artificial intelligence (AI) framework to address these challenges. This serves two potential functions: 1) a simulation environment for exploring various healthcare policies, payment methodologies, etc., and 2) the basis for clinical artificial intelligence - an AI that can think like a doctor. This approach combines Markov decision processes and dynamic decision networks to learn from clinical data and develop complex plans via simulation of alternative sequential decision paths while capturing the sometimes conflicting, sometimes synergistic interactions of various components in the healthcare system. It can operate in partially observable environments (in the case of missing observations or data) by maintaining belief states about patient health status and functions as an online agent that plans and re-plans. This framework was evaluated using real patient data from an electronic health record. Such an AI framework easily outperforms the current treatment-as-usual (TAU) case-rate/fee-for-service models of healthcare (Cost per Unit Change: $189 vs. $497) while obtaining a 30-35% increase in patient outcomes. Tweaking certain model parameters further enhances this advantage, obtaining roughly 50% more improvement for roughly half the costs. Given careful design and problem formulation, an AI simulation framework can approximate optimal decisions even in complex and uncertain environments. Future work is described that outlines potential lines of research and integration of machine learning algorithms for personalized medicine.", "extra": {}}
{"source": "arxiv", "title": "Modeling Belief in Dynamic Systems, Part II: Revision and Update", "url": "http://arxiv.org/abs/cs/9903016v1", "summary": "The study of belief change has been an active area in philosophy and AI. In recent years two special cases of belief change, belief revision and belief update, have been studied in detail. In a companion paper (Friedman &amp; Halpern, 1997), we introduce a new framework to model belief change. This framework combines temporal and epistemic modalities with a notion of plausibility, allowing us to examine the change of beliefs over time. In this paper, we show how belief revision and belief update can be captured in our framework. This allows us to compare the assumptions made by each method, and to better understand the principles underlying them. In particular, it shows that Katsuno and Mendelzon's notion of belief update (Katsuno &amp; Mendelzon, 1991a) depends on several strong assumptions that may limit its applicability in artificial intelligence. Finally, our analysis allow us to identify a notion of minimal change that underlies a broad range of belief change operations including revision and update.", "content": "The study of belief change has been an active area in philosophy and AI. In recent years two special cases of belief change, belief revision and belief update, have been studied in detail. In a companion paper (Friedman &amp; Halpern, 1997), we introduce a new framework to model belief change. This framework combines temporal and epistemic modalities with a notion of plausibility, allowing us to examine the change of beliefs over time. In this paper, we show how belief revision and belief update can be captured in our framework. This allows us to compare the assumptions made by each method, and to better understand the principles underlying them. In particular, it shows that Katsuno and Mendelzon's notion of belief update (Katsuno &amp; Mendelzon, 1991a) depends on several strong assumptions that may limit its applicability in artificial intelligence. Finally, our analysis allow us to identify a notion of minimal change that underlies a broad range of belief change operations including revision and update.", "extra": {}}
{"source": "arxiv", "title": "Study on the Helpfulness of Explainable Artificial Intelligence", "url": "http://arxiv.org/abs/2410.11896v1", "summary": "Explainable Artificial Intelligence (XAI) is essential for building advanced machine learning-powered applications, especially in critical domains such as medical diagnostics or autonomous driving. Legal, business, and ethical requirements motivate using effective XAI, but the increasing number of different methods makes it challenging to pick the right ones. Further, as explanations are highly context-dependent, measuring the effectiveness of XAI methods without users can only reveal a limited amount of information, excluding human factors such as the ability to understand it. We propose to evaluate XAI methods via the user's ability to successfully perform a proxy task, designed such that a good performance is an indicator for the explanation to provide helpful information. In other words, we address the helpfulness of XAI for human decision-making. Further, a user study on state-of-the-art methods was conducted, showing differences in their ability to generate trust and skepticism and the ability to judge the rightfulness of an AI decision correctly. Based on the results, we highly recommend using and extending this approach for more objective-based human-centered user studies to measure XAI performance in an end-to-end fashion.", "content": "Explainable Artificial Intelligence (XAI) is essential for building advanced machine learning-powered applications, especially in critical domains such as medical diagnostics or autonomous driving. Legal, business, and ethical requirements motivate using effective XAI, but the increasing number of different methods makes it challenging to pick the right ones. Further, as explanations are highly context-dependent, measuring the effectiveness of XAI methods without users can only reveal a limited amount of information, excluding human factors such as the ability to understand it. We propose to evaluate XAI methods via the user's ability to successfully perform a proxy task, designed such that a good performance is an indicator for the explanation to provide helpful information. In other words, we address the helpfulness of XAI for human decision-making. Further, a user study on state-of-the-art methods was conducted, showing differences in their ability to generate trust and skepticism and the ability to judge the rightfulness of an AI decision correctly. Based on the results, we highly recommend using and extending this approach for more objective-based human-centered user studies to measure XAI performance in an end-to-end fashion.", "extra": {}}
{"source": "arxiv", "title": "Watershed of Artificial Intelligence: Human Intelligence, Machine Intelligence, and Biological Intelligence", "url": "http://arxiv.org/abs/2104.13155v2", "summary": "This article reviews the \"Once learning\" mechanism that was proposed 23 years ago and the subsequent successes of \"One-shot learning\" in image classification and \"You Only Look Once - YOLO\" in objective detection. Analyzing the current development of Artificial Intelligence (AI), the proposal is that AI should be clearly divided into the following categories: Artificial Human Intelligence (AHI), Artificial Machine Intelligence (AMI), and Artificial Biological Intelligence (ABI), which will also be the main directions of theory and application development for AI. As a watershed for the branches of AI, some classification standards and methods are discussed: 1) Human-oriented, machine-oriented, and biological-oriented AI R&amp;D; 2) Information input processed by Dimensionality-up or Dimensionality-reduction; 3) The use of one/few or large samples for knowledge learning.", "content": "This article reviews the \"Once learning\" mechanism that was proposed 23 years ago and the subsequent successes of \"One-shot learning\" in image classification and \"You Only Look Once - YOLO\" in objective detection. Analyzing the current development of Artificial Intelligence (AI), the proposal is that AI should be clearly divided into the following categories: Artificial Human Intelligence (AHI), Artificial Machine Intelligence (AMI), and Artificial Biological Intelligence (ABI), which will also be the main directions of theory and application development for AI. As a watershed for the branches of AI, some classification standards and methods are discussed: 1) Human-oriented, machine-oriented, and biological-oriented AI R&amp;D; 2) Information input processed by Dimensionality-up or Dimensionality-reduction; 3) The use of one/few or large samples for knowledge learning.", "extra": {}}
{"source": "arxiv", "title": "Games for Artificial Intelligence Research: A Review and Perspectives", "url": "http://arxiv.org/abs/2304.13269v4", "summary": "Games have been the perfect test-beds for artificial intelligence research for the characteristics that widely exist in real-world scenarios. Learning and optimisation, decision making in dynamic and uncertain environments, game theory, planning and scheduling, design and education are common research areas shared between games and real-world problems. Numerous open-source games or game-based environments have been implemented for studying artificial intelligence. In addition to single- or multi-player, collaborative or adversarial games, there has also been growing interest in implementing platforms for creative design in recent years. Those platforms provide ideal benchmarks for exploring and comparing artificial intelligence ideas and techniques. This paper reviews the games and game-based platforms for artificial intelligence research, provides guidance on matching particular types of artificial intelligence with suitable games for testing and matching particular needs in games with suitable artificial intelligence techniques, discusses the research trend induced by the evolution of those games and platforms, and gives an outlook.", "content": "Games have been the perfect test-beds for artificial intelligence research for the characteristics that widely exist in real-world scenarios. Learning and optimisation, decision making in dynamic and uncertain environments, game theory, planning and scheduling, design and education are common research areas shared between games and real-world problems. Numerous open-source games or game-based environments have been implemented for studying artificial intelligence. In addition to single- or multi-player, collaborative or adversarial games, there has also been growing interest in implementing platforms for creative design in recent years. Those platforms provide ideal benchmarks for exploring and comparing artificial intelligence ideas and techniques. This paper reviews the games and game-based platforms for artificial intelligence research, provides guidance on matching particular types of artificial intelligence with suitable games for testing and matching particular needs in games with suitable artificial intelligence techniques, discusses the research trend induced by the evolution of those games and platforms, and gives an outlook.", "extra": {}}
{"source": "arxiv", "title": "Artificial Intelligence and Economic Theories", "url": "http://arxiv.org/abs/1703.06597v1", "summary": "The advent of artificial intelligence has changed many disciplines such as engineering, social science and economics. Artificial intelligence is a computational technique which is inspired by natural intelligence such as the swarming of birds, the working of the brain and the pathfinding of the ants. These techniques have impact on economic theories. This book studies the impact of artificial intelligence on economic theories, a subject that has not been extensively studied. The theories that are considered are: demand and supply, asymmetrical information, pricing, rational choice, rational expectation, game theory, efficient market hypotheses, mechanism design, prospect, bounded rationality, portfolio theory, rational counterfactual and causality. The benefit of this book is that it evaluates existing theories of economics and update them based on the developments in artificial intelligence field.", "content": "The advent of artificial intelligence has changed many disciplines such as engineering, social science and economics. Artificial intelligence is a computational technique which is inspired by natural intelligence such as the swarming of birds, the working of the brain and the pathfinding of the ants. These techniques have impact on economic theories. This book studies the impact of artificial intelligence on economic theories, a subject that has not been extensively studied. The theories that are considered are: demand and supply, asymmetrical information, pricing, rational choice, rational expectation, game theory, efficient market hypotheses, mechanism design, prospect, bounded rationality, portfolio theory, rational counterfactual and causality. The benefit of this book is that it evaluates existing theories of economics and update them based on the developments in artificial intelligence field.", "extra": {}}
{"source": "arxiv", "title": "Linking Artificial Intelligence Principles", "url": "http://arxiv.org/abs/1812.04814v1", "summary": "Artificial Intelligence principles define social and ethical considerations to develop future AI. They come from research institutes, government organizations and industries. All versions of AI principles are with different considerations covering different perspectives and making different emphasis. None of them can be considered as complete and can cover the rest AI principle proposals. Here we introduce LAIP, an effort and platform for linking and analyzing different Artificial Intelligence Principles. We want to explicitly establish the common topics and links among AI Principles proposed by different organizations and investigate on their uniqueness. Based on these efforts, for the long-term future of AI, instead of directly adopting any of the AI principles, we argue for the necessity of incorporating various AI Principles into a comprehensive framework and focusing on how they can interact and complete each other.", "content": "Artificial Intelligence principles define social and ethical considerations to develop future AI. They come from research institutes, government organizations and industries. All versions of AI principles are with different considerations covering different perspectives and making different emphasis. None of them can be considered as complete and can cover the rest AI principle proposals. Here we introduce LAIP, an effort and platform for linking and analyzing different Artificial Intelligence Principles. We want to explicitly establish the common topics and links among AI Principles proposed by different organizations and investigate on their uniqueness. Based on these efforts, for the long-term future of AI, instead of directly adopting any of the AI principles, we argue for the necessity of incorporating various AI Principles into a comprehensive framework and focusing on how they can interact and complete each other.", "extra": {}}
{"source": "arxiv", "title": "The Bayesian Reflex: A Predictive Coding Engine for Artificial Intelligence", "url": "http://arxiv.org/abs/2608.00492v1", "summary": "Predictive coding offers a powerful theory of cortical computation, but corresponding scalable algorithmic implementations for artificial intelligence have remained elusive. This paper introduces the Bayesian reflex, a computational framework that directly instantiates predictive coding through three pillars: belief maintenance via hierarchical generative models, sequential Bayesian updating via prediction-error minimization, and uncertainty-driven action via active inference. We show that recent breakthroughs---ellipsoidal decomposition for exact $i.i.d.$ sampling, recursive Gaussian processes for deep hierarchical inference, and derivative-aware Bayesian optimization---provide the missing algorithmic ingredients. The resulting framework enables mathematically principled, scalable, and brain-inspired continual learning, perception, and decision-making. We illustrate its versatility through applications ranging from climate model evaluation to prime number discovery, offering a blueprint for truly adaptive artificial intelligence.", "content": "Predictive coding offers a powerful theory of cortical computation, but corresponding scalable algorithmic implementations for artificial intelligence have remained elusive. This paper introduces the Bayesian reflex, a computational framework that directly instantiates predictive coding through three pillars: belief maintenance via hierarchical generative models, sequential Bayesian updating via prediction-error minimization, and uncertainty-driven action via active inference. We show that recent breakthroughs---ellipsoidal decomposition for exact $i.i.d.$ sampling, recursive Gaussian processes for deep hierarchical inference, and derivative-aware Bayesian optimization---provide the missing algorithmic ingredients. The resulting framework enables mathematically principled, scalable, and brain-inspired continual learning, perception, and decision-making. We illustrate its versatility through applications ranging from climate model evaluation to prime number discovery, offering a blueprint for truly adaptive artificial intelligence.", "extra": {}}
{"source": "arxiv", "title": "Intelligent behavior depends on the ecological niche: Scaling up AI to human-like intelligence in socio-cultural environments", "url": "http://arxiv.org/abs/2103.06769v1", "summary": "This paper outlines a perspective on the future of AI, discussing directions for machines models of human-like intelligence. We explain how developmental and evolutionary theories of human cognition should further inform artificial intelligence. We emphasize the role of ecological niches in sculpting intelligent behavior, and in particular that human intelligence was fundamentally shaped to adapt to a constantly changing socio-cultural environment. We argue that a major limit of current work in AI is that it is missing this perspective, both theoretically and experimentally. Finally, we discuss the promising approach of developmental artificial intelligence, modeling infant development through multi-scale interaction between intrinsically motivated learning, embodiment and a fastly changing socio-cultural environment. This paper takes the form of an interview of Pierre-Yves Oudeyer by Mandred Eppe, organized within the context of a KI - K{ü}nstliche Intelligenz special issue in developmental robotics.", "content": "This paper outlines a perspective on the future of AI, discussing directions for machines models of human-like intelligence. We explain how developmental and evolutionary theories of human cognition should further inform artificial intelligence. We emphasize the role of ecological niches in sculpting intelligent behavior, and in particular that human intelligence was fundamentally shaped to adapt to a constantly changing socio-cultural environment. We argue that a major limit of current work in AI is that it is missing this perspective, both theoretically and experimentally. Finally, we discuss the promising approach of developmental artificial intelligence, modeling infant development through multi-scale interaction between intrinsically motivated learning, embodiment and a fastly changing socio-cultural environment. This paper takes the form of an interview of Pierre-Yves Oudeyer by Mandred Eppe, organized within the context of a KI - K{ü}nstliche Intelligenz special issue in developmental robotics.", "extra": {}}
{"source": "arxiv", "title": "Probability Judgement in Artificial Intelligence", "url": "http://arxiv.org/abs/1304.3429v1", "summary": "This paper is concerned with two theories of probability judgment: the Bayesian theory and the theory of belief functions. It illustrates these theories with some simple examples and discusses some of the issues that arise when we try to implement them in expert systems. The Bayesian theory is well known; its main ideas go back to the work of Thomas Bayes (1702-1761). The theory of belief functions, often called the Dempster-Shafer theory in the artificial intelligence community, is less well known, but it has even older antecedents; belief-function arguments appear in the work of George Hooper (16401723) and James Bernoulli (1654-1705). For elementary expositions of the theory of belief functions, see Shafer (1976, 1985).", "content": "This paper is concerned with two theories of probability judgment: the Bayesian theory and the theory of belief functions. It illustrates these theories with some simple examples and discusses some of the issues that arise when we try to implement them in expert systems. The Bayesian theory is well known; its main ideas go back to the work of Thomas Bayes (1702-1761). The theory of belief functions, often called the Dempster-Shafer theory in the artificial intelligence community, is less well known, but it has even older antecedents; belief-function arguments appear in the work of George Hooper (16401723) and James Bernoulli (1654-1705). For elementary expositions of the theory of belief functions, see Shafer (1976, 1985).", "extra": {}}
{"source": "arxiv", "title": "AAAI-2019 Workshop on Games and Simulations for Artificial Intelligence", "url": "http://arxiv.org/abs/1903.02172v1", "summary": "This volume represents the accepted submissions from the AAAI-2019 Workshop on Games and Simulations for Artificial Intelligence held on January 29, 2019 in Honolulu, Hawaii, USA. https://www.gamesim.ai", "content": "This volume represents the accepted submissions from the AAAI-2019 Workshop on Games and Simulations for Artificial Intelligence held on January 29, 2019 in Honolulu, Hawaii, USA. https://www.gamesim.ai", "extra": {}}
{"source": "arxiv", "title": "Death and Suicide in Universal Artificial Intelligence", "url": "http://arxiv.org/abs/1606.00652v1", "summary": "Reinforcement learning (RL) is a general paradigm for studying intelligent behaviour, with applications ranging from artificial intelligence to psychology and economics. AIXI is a universal solution to the RL problem; it can learn any computable environment. A technical subtlety of AIXI is that it is defined using a mixture over semimeasures that need not sum to 1, rather than over proper probability measures. In this work we argue that the shortfall of a semimeasure can naturally be interpreted as the agent's estimate of the probability of its death. We formally define death for generally intelligent agents like AIXI, and prove a number of related theorems about their behaviour. Notable discoveries include that agent behaviour can change radically under positive linear transformations of the reward signal (from suicidal to dogmatically self-preserving), and that the agent's posterior belief that it will survive increases over time.", "content": "Reinforcement learning (RL) is a general paradigm for studying intelligent behaviour, with applications ranging from artificial intelligence to psychology and economics. AIXI is a universal solution to the RL problem; it can learn any computable environment. A technical subtlety of AIXI is that it is defined using a mixture over semimeasures that need not sum to 1, rather than over proper probability measures. In this work we argue that the shortfall of a semimeasure can naturally be interpreted as the agent's estimate of the probability of its death. We formally define death for generally intelligent agents like AIXI, and prove a number of related theorems about their behaviour. Notable discoveries include that agent behaviour can change radically under positive linear transformations of the reward signal (from suicidal to dogmatically self-preserving), and that the agent's posterior belief that it will survive increases over time.", "extra": {}}
{"source": "arxiv", "title": "From Statistical Relational to Neurosymbolic Artificial Intelligence: a Survey", "url": "http://arxiv.org/abs/2108.11451v4", "summary": "This survey explores the integration of learning and reasoning in two different fields of artificial intelligence: neurosymbolic and statistical relational artificial intelligence. Neurosymbolic artificial intelligence (NeSy) studies the integration of symbolic reasoning and neural networks, while statistical relational artificial intelligence (StarAI) focuses on integrating logic with probabilistic graphical models. This survey identifies seven shared dimensions between these two subfields of AI. These dimensions can be used to characterize different NeSy and StarAI systems. They are concerned with (1) the approach to logical inference, whether model or proof-based; (2) the syntax of the used logical theories; (3) the logical semantics of the systems and their extensions to facilitate learning; (4) the scope of learning, encompassing either parameter or structure learning; (5) the presence of symbolic and subsymbolic representations; (6) the degree to which systems capture the original logic, probabilistic, and neural paradigms; and (7) the classes of learning tasks the systems are applied to. By positioning various NeSy and StarAI systems along these dimensions and pointing out similarities and differences between them, this survey contributes fundamental concepts for understanding the integration of learning and reasoning.", "content": "This survey explores the integration of learning and reasoning in two different fields of artificial intelligence: neurosymbolic and statistical relational artificial intelligence. Neurosymbolic artificial intelligence (NeSy) studies the integration of symbolic reasoning and neural networks, while statistical relational artificial intelligence (StarAI) focuses on integrating logic with probabilistic graphical models. This survey identifies seven shared dimensions between these two subfields of AI. These dimensions can be used to characterize different NeSy and StarAI systems. They are concerned with (1) the approach to logical inference, whether model or proof-based; (2) the syntax of the used logical theories; (3) the logical semantics of the systems and their extensions to facilitate learning; (4) the scope of learning, encompassing either parameter or structure learning; (5) the presence of symbolic and subsymbolic representations; (6) the degree to which systems capture the original logic, probabilistic, and neural paradigms; and (7) the classes of learning tasks the systems are applied to. By positioning various NeSy and StarAI systems along these dimensions and pointing out similarities and differences between them, this survey contributes fundamental concepts for understanding the integration of learning and reasoning.", "extra": {}}
{"source": "arxiv", "title": "Modular Belief Updates and Confusion about Measures of Certainty in Artificial Intelligence Research", "url": "http://arxiv.org/abs/1407.7281v1", "summary": "Over the last decade, there has been growing interest in the use or measures or change in belief for reasoning with uncertainty in artificial intelligence research. An important characteristic of several methodologies that reason with changes in belief or belief updates, is a property that we term modularity. We call updates that satisfy this property modular updates. Whereas probabilistic measures of belief update - which satisfy the modularity property were first discovered in the nineteenth century, knowledge and discussion of these quantities remains obscure in artificial intelligence research. We define modular updates and discuss their inappropriate use in two influential expert systems.", "content": "Over the last decade, there has been growing interest in the use or measures or change in belief for reasoning with uncertainty in artificial intelligence research. An important characteristic of several methodologies that reason with changes in belief or belief updates, is a property that we term modularity. We call updates that satisfy this property modular updates. Whereas probabilistic measures of belief update - which satisfy the modularity property were first discovered in the nineteenth century, knowledge and discussion of these quantities remains obscure in artificial intelligence research. We define modular updates and discuss their inappropriate use in two influential expert systems.", "extra": {}}
{"source": "arxiv", "title": "Augmented Computational Design: Methodical Application of Artificial Intelligence in Generative Design", "url": "http://arxiv.org/abs/2310.09243v1", "summary": "This chapter presents methodological reflections on the necessity and utility of artificial intelligence in generative design. Specifically, the chapter discusses how generative design processes can be augmented by AI to deliver in terms of a few outcomes of interest or performance indicators while dealing with hundreds or thousands of small decisions. The core of the performance-based generative design paradigm is about making statistical or simulation-driven associations between these choices and consequences for mapping and navigating such a complex decision space. This chapter will discuss promising directions in Artificial Intelligence for augmenting decision-making processes in architectural design for mapping and navigating complex design spaces.", "content": "This chapter presents methodological reflections on the necessity and utility of artificial intelligence in generative design. Specifically, the chapter discusses how generative design processes can be augmented by AI to deliver in terms of a few outcomes of interest or performance indicators while dealing with hundreds or thousands of small decisions. The core of the performance-based generative design paradigm is about making statistical or simulation-driven associations between these choices and consequences for mapping and navigating such a complex decision space. This chapter will discuss promising directions in Artificial Intelligence for augmenting decision-making processes in architectural design for mapping and navigating complex design spaces.", "extra": {}}
{"source": "arxiv", "title": "Rule of Thumb: Explaining Artificial Intelligence Systems using Partial Information", "url": "http://arxiv.org/abs/2608.10766v2", "summary": "Explainable Artificial Intelligence (XAI) seeks to explain how an Artificial Intelligence (AI) system arrived at a particular decision. We propose ''Rule of Thumb'' (RoT) explanations, a new approach to XAI based upon a novel formulation that identifies the most relevant features for predicting the behaviour of an AI system, for a particular datapoint. We show how RoT is well-suited to enable XAI in: (a) zero-shot classification using large language models (LLMs), (b) auditing of opaque AI systems without model access, and (c) the use of AI in scientific discovery. Additionally, RoT meets specific requirements from leading AI regulations, provides a familiar interface and visualisations for XAI practitioners, is model-agnostic, and is substantially faster than alternatives.   Code available at: https://github.com/KaiRawal/Rule-of-Thumb-Explaining-Artificial-Intelligence-Systems-using-Partial-Information", "content": "Explainable Artificial Intelligence (XAI) seeks to explain how an Artificial Intelligence (AI) system arrived at a particular decision. We propose ''Rule of Thumb'' (RoT) explanations, a new approach to XAI based upon a novel formulation that identifies the most relevant features for predicting the behaviour of an AI system, for a particular datapoint. We show how RoT is well-suited to enable XAI in: (a) zero-shot classification using large language models (LLMs), (b) auditing of opaque AI systems without model access, and (c) the use of AI in scientific discovery. Additionally, RoT meets specific requirements from leading AI regulations, provides a familiar interface and visualisations for XAI practitioners, is model-agnostic, and is substantially faster than alternatives.   Code available at: https://github.com/KaiRawal/Rule-of-Thumb-Explaining-Artificial-Intelligence-Systems-using-Partial-Information", "extra": {}}
{"source": "arxiv", "title": "Bias Amplification in Artificial Intelligence Systems", "url": "http://arxiv.org/abs/1809.07842v1", "summary": "As Artificial Intelligence (AI) technologies proliferate, concern has centered around the long-term dangers of job loss or threats of machines causing harm to humans. All of this concern, however, detracts from the more pertinent and already existing threats posed by AI today: its ability to amplify bias found in training datasets, and swiftly impact marginalized populations at scale. Government and public sector institutions have a responsibility to citizens to establish a dialogue with technology developers and release thoughtful policy around data standards to ensure diverse representation in datasets to prevent bias amplification and ensure that AI systems are built with inclusion in mind.", "content": "As Artificial Intelligence (AI) technologies proliferate, concern has centered around the long-term dangers of job loss or threats of machines causing harm to humans. All of this concern, however, detracts from the more pertinent and already existing threats posed by AI today: its ability to amplify bias found in training datasets, and swiftly impact marginalized populations at scale. Government and public sector institutions have a responsibility to citizens to establish a dialogue with technology developers and release thoughtful policy around data standards to ensure diverse representation in datasets to prevent bias amplification and ensure that AI systems are built with inclusion in mind.", "extra": {}}
{"source": "arxiv", "title": "Human-in-the-loop Artificial Intelligence", "url": "http://arxiv.org/abs/1710.08191v1", "summary": "Little by little, newspapers are revealing the bright future that Artificial Intelligence (AI) is building. Intelligent machines will help everywhere. However, this bright future has a dark side: a dramatic job market contraction before its unpredictable transformation. Hence, in a near future, large numbers of job seekers will need financial support while catching up with these novel unpredictable jobs. This possible job market crisis has an antidote inside. In fact, the rise of AI is sustained by the biggest knowledge theft of the recent years. Learning AI machines are extracting knowledge from unaware skilled or unskilled workers by analyzing their interactions. By passionately doing their jobs, these workers are digging their own graves.   In this paper, we propose Human-in-the-loop Artificial Intelligence (HIT-AI) as a fairer paradigm for Artificial Intelligence systems. HIT-AI will reward aware and unaware knowledge producers with a different scheme: decisions of AI systems generating revenues will repay the legitimate owners of the knowledge used for taking those decisions. As modern Robin Hoods, HIT-AI researchers should fight for a fairer Artificial Intelligence that gives back what it steals.", "content": "Little by little, newspapers are revealing the bright future that Artificial Intelligence (AI) is building. Intelligent machines will help everywhere. However, this bright future has a dark side: a dramatic job market contraction before its unpredictable transformation. Hence, in a near future, large numbers of job seekers will need financial support while catching up with these novel unpredictable jobs. This possible job market crisis has an antidote inside. In fact, the rise of AI is sustained by the biggest knowledge theft of the recent years. Learning AI machines are extracting knowledge from unaware skilled or unskilled workers by analyzing their interactions. By passionately doing their jobs, these workers are digging their own graves.   In this paper, we propose Human-in-the-loop Artificial Intelligence (HIT-AI) as a fairer paradigm for Artificial Intelligence systems. HIT-AI will reward aware and unaware knowledge producers with a different scheme: decisions of AI systems generating revenues will repay the legitimate owners of the knowledge used for taking those decisions. As modern Robin Hoods, HIT-AI researchers should fight for a fairer Artificial Intelligence that gives back what it steals.", "extra": {}}
{"source": "arxiv", "title": "Conceptual Modeling and Artificial Intelligence: Mutual Benefits from Complementary Worlds", "url": "http://arxiv.org/abs/2110.08637v1", "summary": "Conceptual modeling (CM) applies abstraction to reduce the complexity of a system under study (e.g., an excerpt of reality). As a result of the conceptual modeling process a human interpretable, formalized representation (i.e., a conceptual model) is derived which enables understanding and communication among humans, and processing by machines. Artificial Intelligence (AI) algorithms are also applied to complex realities (regularly represented by vast amounts of data) to identify patterns or to classify entities in the data. Aside from the commonalities of both approaches, a significant difference can be observed by looking at the results. While conceptual models are comprehensible, reproducible, and explicit knowledge representations, AI techniques are capable of efficiently deriving an output from a given input while acting as a black box. AI solutions often lack comprehensiveness and reproducibility. Even the developers of AI systems can't explain why a certain output is derived. In the Conceptual Modeling meets Artificial Intelligence (CMAI) workshop, we are interested in tackling the intersection of the two, thus far, mostly isolated approached disciplines of CM and AI. The workshop embraces the assumption, that manifold mutual benefits can be realized by i) investigating what Conceptual Modeling (CM) can contribute to AI, and ii) the other way around, what Artificial Intelligence (AI) can contribute to CM.", "content": "Conceptual modeling (CM) applies abstraction to reduce the complexity of a system under study (e.g., an excerpt of reality). As a result of the conceptual modeling process a human interpretable, formalized representation (i.e., a conceptual model) is derived which enables understanding and communication among humans, and processing by machines. Artificial Intelligence (AI) algorithms are also applied to complex realities (regularly represented by vast amounts of data) to identify patterns or to classify entities in the data. Aside from the commonalities of both approaches, a significant difference can be observed by looking at the results. While conceptual models are comprehensible, reproducible, and explicit knowledge representations, AI techniques are capable of efficiently deriving an output from a given input while acting as a black box. AI solutions often lack comprehensiveness and reproducibility. Even the developers of AI systems can't explain why a certain output is derived. In the Conceptual Modeling meets Artificial Intelligence (CMAI) workshop, we are interested in tackling the intersection of the two, thus far, mostly isolated approached disciplines of CM and AI. The workshop embraces the assumption, that manifold mutual benefits can be realized by i) investigating what Conceptual Modeling (CM) can contribute to AI, and ii) the other way around, what Artificial Intelligence (AI) can contribute to CM.", "extra": {}}
{"source": "arxiv", "title": "Enabling Integration and Interaction for Decentralized Artificial Intelligence in Airline Disruption Management", "url": "http://arxiv.org/abs/2104.03349v4", "summary": "Airline disruption management traditionally seeks to address three problem dimensions: aircraft scheduling, crew scheduling, and passenger scheduling, in that order. However, current efforts have, at most, only addressed the first two problem dimensions concurrently and do not account for the propagative effects that uncertain scheduling outcomes in one dimension can have on another dimension. In addition, existing approaches for airline disruption management include human specialists who decide on necessary corrective actions for airline schedule disruptions on the day of operation. However, human specialists are limited in their ability to process copious amounts of information imperative for making robust decisions that simultaneously address all problem dimensions during disruption management. Therefore, there is a need to augment the decision-making capabilities of a human specialist with quantitative and qualitative tools that can rationalize complex interactions amongst all dimensions in airline disruption management, and provide objective insights to the specialists in the airline operations control center. To that effect, we provide a discussion and demonstration of an agnostic and systematic paradigm for enabling expeditious simultaneously-integrated recovery of all problem dimensions during airline disruption management, through an intelligent multi-agent system that employs principles from artificial intelligence and distributed ledger technology. Results indicate that our paradigm for simultaneously-integrated recovery executes in polynomial time and is effective when all the flights in the airline route network are disrupted.", "content": "Airline disruption management traditionally seeks to address three problem dimensions: aircraft scheduling, crew scheduling, and passenger scheduling, in that order. However, current efforts have, at most, only addressed the first two problem dimensions concurrently and do not account for the propagative effects that uncertain scheduling outcomes in one dimension can have on another dimension. In addition, existing approaches for airline disruption management include human specialists who decide on necessary corrective actions for airline schedule disruptions on the day of operation. However, human specialists are limited in their ability to process copious amounts of information imperative for making robust decisions that simultaneously address all problem dimensions during disruption management. Therefore, there is a need to augment the decision-making capabilities of a human specialist with quantitative and qualitative tools that can rationalize complex interactions amongst all dimensions in airline disruption management, and provide objective insights to the specialists in the airline operations control center. To that effect, we provide a discussion and demonstration of an agnostic and systematic paradigm for enabling expeditious simultaneously-integrated recovery of all problem dimensions during airline disruption management, through an intelligent multi-agent system that employs principles from artificial intelligence and distributed ledger technology. Results indicate that our paradigm for simultaneously-integrated recovery executes in polynomial time and is effective when all the flights in the airline route network are disrupted.", "extra": {}}
{"source": "arxiv", "title": "Artificial Collective Intelligence Engineering: a Survey of Concepts and Perspectives", "url": "http://arxiv.org/abs/2304.05147v1", "summary": "Collectiveness is an important property of many systems--both natural and artificial. By exploiting a large number of individuals, it is often possible to produce effects that go far beyond the capabilities of the smartest individuals, or even to produce intelligent collective behaviour out of not-so-intelligent individuals. Indeed, collective intelligence, namely the capability of a group to act collectively in a seemingly intelligent way, is increasingly often a design goal of engineered computational systems--motivated by recent techno-scientific trends like the Internet of Things, swarm robotics, and crowd computing, just to name a few. For several years, the collective intelligence observed in natural and artificial systems has served as a source of inspiration for engineering ideas, models, and mechanisms. Today, artificial and computational collective intelligence are recognised research topics, spanning various techniques, kinds of target systems, and application domains. However, there is still a lot of fragmentation in the research panorama of the topic within computer science, and the verticality of most communities and contributions makes it difficult to extract the core underlying ideas and frames of reference. The challenge is to identify, place in a common structure, and ultimately connect the different areas and methods addressing intelligent collectives. To address this gap, this paper considers a set of broad scoping questions providing a map of collective intelligence research, mostly by the point of view of computer scientists and engineers. Accordingly, it covers preliminary notions, fundamental concepts, and the main research perspectives, identifying opportunities and challenges for researchers on artificial and computational collective intelligence engineering.", "content": "Collectiveness is an important property of many systems--both natural and artificial. By exploiting a large number of individuals, it is often possible to produce effects that go far beyond the capabilities of the smartest individuals, or even to produce intelligent collective behaviour out of not-so-intelligent individuals. Indeed, collective intelligence, namely the capability of a group to act collectively in a seemingly intelligent way, is increasingly often a design goal of engineered computational systems--motivated by recent techno-scientific trends like the Internet of Things, swarm robotics, and crowd computing, just to name a few. For several years, the collective intelligence observed in natural and artificial systems has served as a source of inspiration for engineering ideas, models, and mechanisms. Today, artificial and computational collective intelligence are recognised research topics, spanning various techniques, kinds of target systems, and application domains. However, there is still a lot of fragmentation in the research panorama of the topic within computer science, and the verticality of most communities and contributions makes it difficult to extract the core underlying ideas and frames of reference. The challenge is to identify, place in a common structure, and ultimately connect the different areas and methods addressing intelligent collectives. To address this gap, this paper considers a set of broad scoping questions providing a map of collective intelligence research, mostly by the point of view of computer scientists and engineers. Accordingly, it covers preliminary notions, fundamental concepts, and the main research perspectives, identifying opportunities and challenges for researchers on artificial and computational collective intelligence engineering.", "extra": {}}
{"source": "arxiv", "title": "A philosophical and ontological perspective on Artificial General Intelligence and the Metaverse", "url": "http://arxiv.org/abs/2402.06660v4", "summary": "This paper leverages various philosophical and ontological frameworks to explore the concept of embodied artificial general intelligence (AGI), its relationship to human consciousness, and the key role of the metaverse in facilitating this relationship. Several theoretical frameworks underpin this exploration, such as embodied cognition, Michael Levin's computational boundary of a \"Self,\" and Donald D. Hoffman's Interface Theory of Perception, which lead to considering human perceived outer reality as a symbolic representation of alternate inner states of being, and where AGI could embody a different form of consciousness with a larger computational boundary. The paper further discusses the necessary architecture for the emergence of an embodied AGI, how to calibrate an AGI's symbolic interface, and the key role played by the Metaverse, decentralized systems and open-source blockchain technology. The paper concludes by emphasizing the importance of achieving a certain degree of harmony in human relations and recognizing the interconnectedness of humanity at a global level, as key prerequisites for the emergence of a stable embodied AGI.", "content": "This paper leverages various philosophical and ontological frameworks to explore the concept of embodied artificial general intelligence (AGI), its relationship to human consciousness, and the key role of the metaverse in facilitating this relationship. Several theoretical frameworks underpin this exploration, such as embodied cognition, Michael Levin's computational boundary of a \"Self,\" and Donald D. Hoffman's Interface Theory of Perception, which lead to considering human perceived outer reality as a symbolic representation of alternate inner states of being, and where AGI could embody a different form of consciousness with a larger computational boundary. The paper further discusses the necessary architecture for the emergence of an embodied AGI, how to calibrate an AGI's symbolic interface, and the key role played by the Metaverse, decentralized systems and open-source blockchain technology. The paper concludes by emphasizing the importance of achieving a certain degree of harmony in human relations and recognizing the interconnectedness of humanity at a global level, as key prerequisites for the emergence of a stable embodied AGI.", "extra": {}}
{"source": "arxiv", "title": "Why Artificial Intelligence Needs a Task Theory --- And What It Might Look Like", "url": "http://arxiv.org/abs/1604.04660v2", "summary": "The concept of \"task\" is at the core of artificial intelligence (AI): Tasks are used for training and evaluating AI systems, which are built in order to perform and automatize tasks we deem useful. In other fields of engineering theoretical foundations allow thorough evaluation of designs by methodical manipulation of well understood parameters with a known role and importance; this allows an aeronautics engineer, for instance, to systematically assess the effects of wind speed on an airplane's performance and stability. No framework exists in AI that allows this kind of methodical manipulation: Performance results on the few tasks in current use (cf. board games, question-answering) cannot be easily compared, however similar or different. The issue is even more acute with respect to artificial *general* intelligence systems, which must handle unanticipated tasks whose specifics cannot be known beforehand. A *task theory* would enable addressing tasks at the *class* level, bypassing their specifics, providing the appropriate formalization and classification of tasks, environments, and their parameters, resulting in more rigorous ways of measuring, comparing, and evaluating intelligent behavior. Even modest improvements in this direction would surpass the current ad-hoc nature of machine learning and AI evaluation. Here we discuss the main elements of the argument for a task theory and present an outline of what it might look like for physical tasks.", "content": "The concept of \"task\" is at the core of artificial intelligence (AI): Tasks are used for training and evaluating AI systems, which are built in order to perform and automatize tasks we deem useful. In other fields of engineering theoretical foundations allow thorough evaluation of designs by methodical manipulation of well understood parameters with a known role and importance; this allows an aeronautics engineer, for instance, to systematically assess the effects of wind speed on an airplane's performance and stability. No framework exists in AI that allows this kind of methodical manipulation: Performance results on the few tasks in current use (cf. board games, question-answering) cannot be easily compared, however similar or different. The issue is even more acute with respect to artificial *general* intelligence systems, which must handle unanticipated tasks whose specifics cannot be known beforehand. A *task theory* would enable addressing tasks at the *class* level, bypassing their specifics, providing the appropriate formalization and classification of tasks, environments, and their parameters, resulting in more rigorous ways of measuring, comparing, and evaluating intelligent behavior. Even modest improvements in this direction would surpass the current ad-hoc nature of machine learning and AI evaluation. Here we discuss the main elements of the argument for a task theory and present an outline of what it might look like for physical tasks.", "extra": {}}
{"source": "arxiv", "title": "OAK4XAI: Model towards Out-Of-Box eXplainable Artificial Intelligence for Digital Agriculture", "url": "http://arxiv.org/abs/2209.15104v1", "summary": "Recent machine learning approaches have been effective in Artificial Intelligence (AI) applications. They produce robust results with a high level of accuracy. However, most of these techniques do not provide human-understandable explanations for supporting their results and decisions. They usually act as black boxes, and it is not easy to understand how decisions have been made. Explainable Artificial Intelligence (XAI), which has received much interest recently, tries to provide human-understandable explanations for decision-making and trained AI models. For instance, in digital agriculture, related domains often present peculiar or input features with no link to background knowledge. The application of the data mining process on agricultural data leads to results (knowledge), which are difficult to explain. In this paper, we propose a knowledge map model and an ontology design as an XAI framework (OAK4XAI) to deal with this issue. The framework does not only consider the data analysis part of the process, but it takes into account the semantics aspect of the domain knowledge via an ontology and a knowledge map model, provided as modules of the framework. Many ongoing XAI studies aim to provide accurate and verbalizable accounts for how given feature values contribute to model decisions. The proposed approach, however, focuses on providing consistent information and definitions of concepts, algorithms, and values involved in the data mining models. We built an Agriculture Computing Ontology (AgriComO) to explain the knowledge mined in agriculture. AgriComO has a well-designed structure and includes a wide range of concepts and transformations suitable for agriculture and computing domains.", "content": "Recent machine learning approaches have been effective in Artificial Intelligence (AI) applications. They produce robust results with a high level of accuracy. However, most of these techniques do not provide human-understandable explanations for supporting their results and decisions. They usually act as black boxes, and it is not easy to understand how decisions have been made. Explainable Artificial Intelligence (XAI), which has received much interest recently, tries to provide human-understandable explanations for decision-making and trained AI models. For instance, in digital agriculture, related domains often present peculiar or input features with no link to background knowledge. The application of the data mining process on agricultural data leads to results (knowledge), which are difficult to explain. In this paper, we propose a knowledge map model and an ontology design as an XAI framework (OAK4XAI) to deal with this issue. The framework does not only consider the data analysis part of the process, but it takes into account the semantics aspect of the domain knowledge via an ontology and a knowledge map model, provided as modules of the framework. Many ongoing XAI studies aim to provide accurate and verbalizable accounts for how given feature values contribute to model decisions. The proposed approach, however, focuses on providing consistent information and definitions of concepts, algorithms, and values involved in the data mining models. We built an Agriculture Computing Ontology (AgriComO) to explain the knowledge mined in agriculture. AgriComO has a well-designed structure and includes a wide range of concepts and transformations suitable for agriculture and computing domains.", "extra": {}}
{"source": "arxiv", "title": "Certifiable Artificial Intelligence Through Data Fusion", "url": "http://arxiv.org/abs/2111.02001v1", "summary": "This paper reviews and proposes concerns in adopting, fielding, and maintaining artificial intelligence (AI) systems. While the AI community has made rapid progress, there are challenges in certifying AI systems. Using procedures from design and operational test and evaluation, there are opportunities towards determining performance bounds to manage expectations of intended use. A notional use case is presented with image data fusion to support AI object recognition certifiability considering precision versus distance.", "content": "This paper reviews and proposes concerns in adopting, fielding, and maintaining artificial intelligence (AI) systems. While the AI community has made rapid progress, there are challenges in certifying AI systems. Using procedures from design and operational test and evaluation, there are opportunities towards determining performance bounds to manage expectations of intended use. A notional use case is presented with image data fusion to support AI object recognition certifiability considering precision versus distance.", "extra": {}}
{"source": "arxiv", "title": "Blue Sky Ideas in Artificial Intelligence Education from the EAAI 2017 New and Future AI Educator Program", "url": "http://arxiv.org/abs/1702.00137v1", "summary": "The 7th Symposium on Educational Advances in Artificial Intelligence (EAAI'17, co-chaired by Sven Koenig and Eric Eaton) launched the EAAI New and Future AI Educator Program to support the training of early-career university faculty, secondary school faculty, and future educators (PhD candidates or postdocs who intend a career in academia). As part of the program, awardees were asked to address one of the following \"blue sky\" questions:   * How could/should Artificial Intelligence (AI) courses incorporate ethics into the curriculum?   * How could we teach AI topics at an early undergraduate or a secondary school level?   * AI has the potential for broad impact to numerous disciplines. How could we make AI education more interdisciplinary, specifically to benefit non-engineering fields?   This paper is a collection of their responses, intended to help motivate discussion around these issues in AI education.", "content": "The 7th Symposium on Educational Advances in Artificial Intelligence (EAAI'17, co-chaired by Sven Koenig and Eric Eaton) launched the EAAI New and Future AI Educator Program to support the training of early-career university faculty, secondary school faculty, and future educators (PhD candidates or postdocs who intend a career in academia). As part of the program, awardees were asked to address one of the following \"blue sky\" questions:   * How could/should Artificial Intelligence (AI) courses incorporate ethics into the curriculum?   * How could we teach AI topics at an early undergraduate or a secondary school level?   * AI has the potential for broad impact to numerous disciplines. How could we make AI education more interdisciplinary, specifically to benefit non-engineering fields?   This paper is a collection of their responses, intended to help motivate discussion around these issues in AI education.", "extra": {}}
{"source": "pubmed", "title": "1. Work. 2026 Sep 18:10519815261487194. doi: 10.1177/10519815261487194. Online ", "url": "https://pubmed.ncbi.nlm.nih.gov/42760271/", "summary": "1. Work. 2026 Sep 18:10519815261487194. doi: 10.1177/10519815261487194. Online \nahead of print.\n\nWorkload and discomfort associated with pushing and pulling manual delivery \ntrolleys.\n\nDeng JJ(1), Wang Q(2), Cheng HI(2).\n\nAuthor information:\n(1)Interdisciplinary Artificial Intelligence Research Inst", "content": "1. Work. 2026 Sep 18:10519815261487194. doi: 10.1177/10519815261487194. Online \nahead of print.\n\nWorkload and discomfort associated with pushing and pulling manual delivery \ntrolleys.\n\nDeng JJ(1), Wang Q(2), Cheng HI(2).\n\nAuthor information:\n(1)Interdisciplinary Artificial Intelligence Research Institute, Wuhan College, \nWuhan, Hubei, China.\n(2)Department of Industrial and Management Engineering, Kyungsung University, \nBusan, Korea.\n\nBackgroundCourier service workers transport diverse packages on delivery \ntrolleys. The repetitive pushing and pulling tasks of manual delivery trolleys \ncan increase the risk of work-related musculoskeletal disorders among courier \nworkers.ObjectiveA comparison of different trolley types, task types, and \ntransport directions was conducted to determine the physical workload and \nergonomic demands of pushing and pulling tasks during last-mile \ndeliveries.MethodsA within-subjects experiment was conducted with 23 \nprofessional male courier workers. Participants performed pushing and pulling \ntasks with three delivery trolleys: a hand truck, hand cart, and rolltainer. \nTrolley type, task type, and transport direction (straight and curved) were \nexamined. Muscle activity was measured using electromyography (sEMG). Joint \nflexion angles were obtained from video recordings using the OpenPose Body-25 \nmodel. Lumbar compression and shear forces were estimated using the 3D Static \nStrength Prediction Program (3D SSPP).ResultsPulling generally induced greater \nactivation of upper-limb muscles and greater flexion of the lower arms and neck \nthan pushing, whereas pushing resulted in greater trunk flexion. Among the \ntrolley types, the hand cart produced the highest muscle activity and the \ngreatest lumbar compression and shear force. Hand trucks imposed the lowest \nbiomechanical demands. Pulling a hand cart generated the greatest load on the \nlower back. Speeds of transport ranged from 0.66 to 1.15 m/s, with pushing and \nstraight-line movements achieving significantly faster speeds than pulling and \ncurved movements. Last-mile delivery caused the most subjective discomfort in \nthe arms, followed by the lumbar region, while shoulders caused the least \ndiscomfort.ConclusionThe findings indicate that trolley type and handling method \nsubstantially influence courier delivery workload and biomechanical demands. In \nroutine delivery tasks involving moderate loads, hand trucks may be an \nergonomically beneficial option. The findings of this study provide practical \nevidence for improving trolley selection, work methods, and injury prevention \nstrategies in the courier industry.\n\nDOI: 10.1177/10519815261487194\nPMID: 42760271", "extra": {"pmid": "42760271"}}
{"source": "pubmed", "title": "2. Arch Bronconeumol. 2026 Sep 18:S0300-2896(26)00324-8. doi: ", "url": "https://pubmed.ncbi.nlm.nih.gov/42760213/", "summary": "2. Arch Bronconeumol. 2026 Sep 18:S0300-2896(26)00324-8. doi: \n10.1016/j.arbres.2026.08.013. Online ahead of print.\n\nArtificial Intelligence in Respiratory Medicine: Applications, Methodological \nChallenges, and Clinical Translation.\n\n[Article in English, Spanish]\n\nLópez-Padilla D(1), Lumbreras S(2)", "content": "2. Arch Bronconeumol. 2026 Sep 18:S0300-2896(26)00324-8. doi: \n10.1016/j.arbres.2026.08.013. Online ahead of print.\n\nArtificial Intelligence in Respiratory Medicine: Applications, Methodological \nChallenges, and Clinical Translation.\n\n[Article in English, Spanish]\n\nLópez-Padilla D(1), Lumbreras S(2), Agustí À(3), Jacob J(4), Rocco G(5), \nMalhotra A(6), Soriano JB(7).\n\nAuthor information:\n(1)Respiratory Department, Hospital General Universitario Gregorio Marañón, \nMadrid, Spain; Faculty of Medicine, Universidad Complutense de Madrid (UCM), \nMadrid, Spain; Instituto de Investigación Sanitaria Gregorio Marañón (IiSGM), \nMadrid, Spain. Electronic address: lopez.padilla84@gmail.com.\n(2)Universidad Pontificia Comillas - IIT, Madrid, Spain.\n(3)Respiratory Institute, Clinic Barcelona, Barcelona, Spain; Universitat de \nBarcelona, Barcelona, Spain; Fundació Clinic - Institut d'Investigacions \nBiomèdiques August Pi i Sunyer (FCRB-IDIBAPS), Barcelona, Spain; Centro de \nInvestigación Biomédica en Red de Enfermedades Respiratorias (CIBERES), \nBarcelona, Spain.\n(4)Satsuma Lab, Hawkes Institute, University College London, London, England, \nUnited Kingdom; UCL Respiratory, University College London, London, England, \nUnited Kingdom.\n(5)Chest Wall Program, Thoracic Service, Department of Surgery, Memorial Sloan \nKettering Cancer Center, New York, NY, USA; Fiona and Stanley Druckenmiller \nCenter for Lung Cancer Research, Memorial Sloan Kettering Cancer Center, New \nYork, NY, USA; Department of Cardiothoracic Surgery, Weill-Cornell Medical \nCollege, New York, NY, USA.\n(6)University of California San Diego, La Jolla, CA, USA.\n(7)Facultat de Medicina, Universidad de les Illes Balears, Palma, Spain; Centro \nde Investigación Biomédica en Red de Enfermedades Respiratorias (CIBERES), \nInstituto de Salud Carlos III, Madrid, Spain.\n\nArtificial intelligence (AI) is rapidly transforming respiratory medicine by \nenabling the integration and analysis of complex multimodal data, including \nimaging, physiological signals, electronic health records, and patient-generated \ninformation. This invited state-of-the-art review synthesizes current evidence \non AI applications across chronic airway diseases, lung imaging, thoracic \nsurgery and oncology, and sleep medicine and non-invasive ventilation, with a \nstrong focus on clinically meaningful use cases. We highlight how AI can support \nearlier and more accurate diagnosis, refined phenotyping and endotyping, \nimproved risk stratification, and more individualized therapeutic strategies. \nRepresentative applications include AI-assisted spirometry interpretation, \nmultimodal exacerbation prediction, advanced quantitative imaging, perioperative \nrisk modelling, wearable-based sleep apnoea detection, and AI-assisted \nrespiratory support decision-making. Alongside these opportunities, we discuss \ncross-cutting challenges that currently limit routine implementation, including \ndata quality and representativeness, insufficient external validation and \ngeneralizability, algorithmic bias, limited explainability of complex models, \nand concerns regarding privacy and data governance. Throughout the review, we \nargue that AI should be conceived as a tool to augment, rather than replace \nclinical judgment, embedded within human-in-the-loop frameworks that preserve \ntransparency, fairness, and accountability. We conclude by outlining priorities \nfor future research and policy, such as prospective clinical evaluation, \nmulticenter validation in diverse populations, and closer collaboration between \nclinicians, data scientists, and regulators, which will be essential to \ntranslate the promise of AI into tangible improvements in respiratory outcomes \nand patient-centered care.\n\nCopyright © 2026 SEPAR. Published by Elsevier España, S.L.U. All rights \nreserved.\n\nDOI: 10.1016/j.arbres.2026.08.013\nPMID: 42760213", "extra": {"pmid": "42760213"}}
{"source": "pubmed", "title": "3. Injury. 2026 Sep 14:113719. doi: 10.1016/j.injury.2026.113719. Online ahead of", "url": "https://pubmed.ncbi.nlm.nih.gov/42760195/", "summary": "3. Injury. 2026 Sep 14:113719. doi: 10.1016/j.injury.2026.113719. Online ahead of\n print.\n\nCooler by cooler - Forecasting transfusion demand and volumes in trauma.\n\nStolarski AE(1), Brady KJ(2), Stallings JD(3), Scantling D(4), Torres C(4), Ong \nC(4), Lagazzi E(4), Seamon M(5), Martin N(5), Telfer B", "content": "3. Injury. 2026 Sep 14:113719. doi: 10.1016/j.injury.2026.113719. Online ahead of\n print.\n\nCooler by cooler - Forecasting transfusion demand and volumes in trauma.\n\nStolarski AE(1), Brady KJ(2), Stallings JD(3), Scantling D(4), Torres C(4), Ong \nC(4), Lagazzi E(4), Seamon M(5), Martin N(5), Telfer BA(2), Saillant N(4).\n\nAuthor information:\n(1)University of Pennsylvania, Division of Traumatology, Surgical Critical Care \nand Emergency Surgery, Philadelphia, PA, United States. Electronic address: \nAllan.Stolarski@pennmedicine.upenn.edu.\n(2)Massachusetts Institute of Technology, Lincoln Laboratory, Human Health & \nPerformance Systems, Lexington, MA, United States.\n(3)Boston Medical Center, Division of Trauma and Acute Care Surgery, Boston, MA, \nUnited States.\n(4)Joint Trauma System, Defense Health Agency, San Antonio, TX, United States.\n(5)University of Pennsylvania, Division of Traumatology, Surgical Critical Care \nand Emergency Surgery, Philadelphia, PA, United States.\n\nINTRODUCTION: Predictive algorithms may optimize the delivery of care by \nfacilitating early identification of resource requirements. We hypothesize that \ninjury patterns and vitals obtained early after trauma can predict blood volume \nrequirements for blunt and penetrating trauma across a wide array of \nenvironments.\nMETHODS: Patients 18-60 years with blunt, penetrating and burn were included \nfrom the 2020 American College of Surgeons-Trauma Quality Improvement Program \n(TQIP) registry. Emergency Medical Services (EMS) and emergency department (ED) \nvital signs were analyzed. External injuries that could be easily noted in the \nfield were curated from ICD codes. Blood volumes were treated as a continuous \nvariable with probabilistic distributions predicted. Blood products \ncharacterized include packed red blood cells, plasma, platelets, cryoprecipitate \nand whole blood. The Field Artificial Intelligence Triage (FAIT) tool utilizes \n10-fold cross validation using logistic regression and class-based weighting for \npredictions.\nRESULTS: The population was 401,337 patients; (314,790 blunt, 71,760 \npenetrating, and 5726 burn) patients. The median age was 37 (interquartile range \n(IQR), 27-50). Median injury severity score (ISS) of 6 (IQR, 4-12) with ISS of \n26 (IQR, 17-35) for those receiving massive transfusion (>4 units/4 h). FAIT \naccurately predicted when patients did not require blood; 0.91 ± 0.00 (AUC). \nOverall predictive capabilities for transfusion volume across ranges \ncorresponding to transfusion (>0 units/4 h), massive transfusion (>4 units/4 h), \nand ultra-massive transfusion (>10 units/4 h and >20 units/4 h) had an AUC of \n0.91-0.95 for the EMS + ED and 0.87-0.91 for the EMS vitals alone. Overall \nsurvival for different transfusion thresholds was 71% for > 5 units, 62% for \n> 10 units and 35% for > 50 units (AUC 0.94-0.95 ± 0.01).\nCONCLUSION: The FAIT tool capably distinguishes patients who require transfusion \nfrom those who do not across a wide array of injury patterns, severity, and \nenvironments. With only limited pre-hospital inputs, the model predicts \ntransfusion volumes within clinically relevant margins.\n\nCopyright © 2026 The Authors. Published by Elsevier Ltd.. All rights reserved.\n\nDOI: 10.1016/j.injury.2026.113719\nPMID: 42760195\n\nConflict of interest statement: Declaration of Competing Interest I, Allan E. \nStolarski, attest on behalf of all authors, that we had full access to the data \nof the study, conducted all data analyses independently from the funding entity, \nand take complete responsibility for the integrity and accuracy of the data \nreported in the manuscript. All authors have completed the required conflicts of \ninterest forms and they are attached with the provided submission.", "extra": {"pmid": "42760195"}}
{"source": "pubmed", "title": "4. Int J Oral Maxillofac Surg. 2026 Sep 18:S0901-5027(26)00399-1. doi: ", "url": "https://pubmed.ncbi.nlm.nih.gov/42760153/", "summary": "4. Int J Oral Maxillofac Surg. 2026 Sep 18:S0901-5027(26)00399-1. doi: \n10.1016/j.ijom.2026.09.006. Online ahead of print.\n\nFacial attractiveness and profile-based perceived orthognathic surgery need: \ncomparative assessment among professionals, laypersons, and artificial \nintelligence models.\n\nGüze", "content": "4. Int J Oral Maxillofac Surg. 2026 Sep 18:S0901-5027(26)00399-1. doi: \n10.1016/j.ijom.2026.09.006. Online ahead of print.\n\nFacial attractiveness and profile-based perceived orthognathic surgery need: \ncomparative assessment among professionals, laypersons, and artificial \nintelligence models.\n\nGüzel C(1), Okumuş YM(2), Türkan I(3), Yılancı H(4).\n\nAuthor information:\n(1)Department of Oral and Maxillofacial Surgery, İstanbul Medipol University \nFaculty of Dentistry, İstanbul, Turkey. Electronic address: \nceylan.guzel@medipol.edu.tr.\n(2)Department of Oral and Maxillofacial Surgery, İstanbul Medipol University \nFaculty of Dentistry, İstanbul, Turkey.\n(3)Graduate School of Health Sciences, School of Dentistry, İstanbul Medipol \nUniversity, İstanbul, Turkey.\n(4)Department of Orthodontics, İstanbul Medipol University Faculty of Dentistry, \nİstanbul, Turkey.\n\nThis study evaluated the effect of controlled 2° sagittal profile modifications \non facial attractiveness ratings and profile-based perceived orthognathic \nsurgery need using one standardized female and one standardized male facial \nmodel. For each model, nine digitally manipulated profile images were generated \nfrom -8° to +8° relative to baseline and assessed by orthodontists, oral and \nmaxillofacial surgeons (OMFS), and laypersons (n = 72). Two multimodal \nartificial intelligence systems, ChatGPT-4.0 and Gemini 1.5 Pro, independently \ngenerated attractiveness scores to provide an exploratory benchmark for human-AI \nagreement. The selected models showed distinct sagittal response patterns. For \nthe female model, attractiveness ratings peaked within a mild retrusive range \n(-2° to -4°), whereas ratings for the male model increased with greater sagittal \nprojection. Significant intergroup differences were observed at selected \nsagittal positions, with orthodontists generally showing narrower esthetic \ntolerance. Profile-based surgical recommendations differed significantly for the \nmale model (P < 0.001), with OMFS specialists recommending intervention most \nfrequently, while no significant overall group difference was found for the \nfemale model. Human-AI agreement was model-dependent: Gemini showed closer \nalignment with human ratings for the female model, whereas ChatGPT-4.0 aligned \nmore closely with human ratings for the male model. Both AI systems declined to \nprovide surgical recommendations because of built-in ethical restrictions. \nWithin the limitations of a two-model, profile-based design, sagittal esthetic \nperception appeared to vary according to model sex, evaluator background, and AI \nsystem. These findings should be interpreted as model-specific and exploratory \nrather than as generalizable sex-based esthetic norms.\n\nCopyright © 2026 International Association of Oral and Maxillofacial Surgeons. \nPublished by Elsevier Inc. All rights reserved.\n\nDOI: 10.1016/j.ijom.2026.09.006\nPMID: 42760153\n\nConflict of interest statement: Competing interests None.", "extra": {"pmid": "42760153"}}
{"source": "pubmed", "title": "5. Curr Probl Diagn Radiol. 2026 Sep 11:S0363-0188(26)00161-1. doi: ", "url": "https://pubmed.ncbi.nlm.nih.gov/42760143/", "summary": "5. Curr Probl Diagn Radiol. 2026 Sep 11:S0363-0188(26)00161-1. doi: \n10.1067/j.cpradiol.2026.09.002. Online ahead of print.\n\nThe value of machine learning inference to predict multidisciplinary discussion \ndiagnosis of interstitial lung diseases.\n\nMortani Barbosa EJ Jr(1), Kim Y(2), Shanahan J(3), R", "content": "5. Curr Probl Diagn Radiol. 2026 Sep 11:S0363-0188(26)00161-1. doi: \n10.1067/j.cpradiol.2026.09.002. Online ahead of print.\n\nThe value of machine learning inference to predict multidisciplinary discussion \ndiagnosis of interstitial lung diseases.\n\nMortani Barbosa EJ Jr(1), Kim Y(2), Shanahan J(3), Rupsee A(4).\n\nAuthor information:\n(1)Department of Radiology, Division of Cardiothoracic Imaging, Perelman School \nof Medicine, University of Pennsylvania, 3400 Spruce Street, Ground Floor \nFounders Bldg, Philadelphia, Pennsylvania, USA. Electronic address: \neduardo.barbosa@pennmedicine.upenn.edu.\n(2)Department of Radiology, Division of Cardiothoracic Imaging, Perelman School \nof Medicine, University of Pennsylvania, 3400 Spruce Street, Ground Floor \nFounders Bldg, Philadelphia, Pennsylvania, USA. Electronic address: \nyohan.kim@pennmedicine.upenn.edu.\n(3)Department of Radiology, Division of Cardiothoracic Imaging, Perelman School \nof Medicine, University of Pennsylvania, 3400 Spruce Street, Ground Floor \nFounders Bldg, Philadelphia, Pennsylvania, USA. Electronic address: \njustin.shanahan@pennmedicine.upenn.edu.\n(4)Department of Radiology, Division of Cardiothoracic Imaging, Perelman School \nof Medicine, University of Pennsylvania, 3400 Spruce Street, Ground Floor \nFounders Bldg, Philadelphia, Pennsylvania, USA. Electronic address: \nalex.rupsee@pennmedicine.upenn.edu.\n\nOBJECTIVES: Multidisciplinary discussion (MDD) is the reference standard for \ndiagnosing interstitial lung diseases (ILD), however, it requires considerable \nexpertise, and up to 25% of patients remain unclassifiable. We evaluated whether \nmachine learning (ML) models trained on multimodal data can accurately replicate \nMDD consensus ILD diagnoses.\nMATERIALS & METHODS: We retrospectively identified 428 patients with ILD \nconfirmed by MDD consensus at a tertiary academic center from 2010 to 2021. \nUsing comprehensive clinical and radiological data, we trained and \ncross-validated multiple ML models for three classifications: (1) three-level: \nconnective tissue disease associated ILD (CTD-ILD), usual interstitial pneumonia \n(UIP), or unclassifiable; (2) UIP/IPF versus all other, and (3) unclassifiable \nversus all other. Logistic regression, random forest, and neural networks ML \nmodels were trained with five-fold cross-validation. Performance was compared \nvia area under the curve (AUC) with 95% confidence intervals.\nRESULTS: MDD diagnoses were UIP/IPF in 131 patients (30.6%), CTD-ILD in 130 \n(30.4%), and unclassifiable ILD in 167 (39.0%). Clinical variables and CT \nimaging features were more predictive than PFTs. ML models achieved AUC values \nranging from 0.66 to 0.92. For UIP/IPF vs. other, the neural network reached an \nAUC of 0.88. Notably, the Lasso regression model exhibited high specificity \n(0.966), suggesting utility as a high-confidence \"rule-in\" AI tool.\nCONCLUSION: ML classification closely approximated MDD diagnoses of ILD, \nespecially for UIP/IPF, relying primarily on clinical and CT imaging features, \neven without pathology information, allowing expert-level diagnostic performance \nwhere MDD resources are not available.\n\nCopyright © 2026 The Author(s). Published by Elsevier Inc. All rights reserved.\n\nDOI: 10.1067/j.cpradiol.2026.09.002\nPMID: 42760143\n\nConflict of interest statement: Declaration of competing interest The authors \ndeclare the following financial interests/personal relationships which may be \nconsidered as potential competing interests: Eduardo M. Barbosa Jr. reports a \nrelationship with Siemens Healthineers that includes: funding grants. If there \nare other authors, they declare that they have no known competing financial \ninterests or personal relationships that could have appeared to influence the \nwork reported in this paper.", "extra": {"pmid": "42760143"}}
{"source": "pubmed", "title": "6. ACS Appl Mater Interfaces. 2026 Sep 21. doi: 10.1021/acsami.6c10769. Online ", "url": "https://pubmed.ncbi.nlm.nih.gov/42760093/", "summary": "6. ACS Appl Mater Interfaces. 2026 Sep 21. doi: 10.1021/acsami.6c10769. Online \nahead of print.\n\nSmartphone AI-Enabled Lateral Flow Immunoassay Platform Using Advanced Quantum \nDots for Intelligent Quantitative Diagnostics.\n\nAtta S(1)(2), Thorsen TL(1)(3), Vasse T(1)(2), Berner VK(4), Gates-Hollings", "content": "6. ACS Appl Mater Interfaces. 2026 Sep 21. doi: 10.1021/acsami.6c10769. Online \nahead of print.\n\nSmartphone AI-Enabled Lateral Flow Immunoassay Platform Using Advanced Quantum \nDots for Intelligent Quantitative Diagnostics.\n\nAtta S(1)(2), Thorsen TL(1)(3), Vasse T(1)(2), Berner VK(4), Gates-Hollingsworth \nMA(4), Hill HJ(4), AuCoin DP(4), Zhu X(5), Vo-Dinh T(1)(2)(3).\n\nAuthor information:\n(1)Fitzpatrick Institute for Photonics, Duke University, Durham, North \nCarolina27708, United States.\n(2)Department of Biomedical Engineering, Duke University, Durham, North \nCarolina27708, United States.\n(3)Department of Chemistry, Duke University, Durham, North Carolina27708, United \nStates.\n(4)Department of Microbiology and Immunology, University of Nevada, Reno School \nof Medicine, Reno, Nevada89557, United States.\n(5)Department of Electrical and Biomedical Engineering, University of Nevada, \nReno, Reno, Nevada89557, United States.\n\nLateral flow immunoassays (LFIAs) are widely used in point-of-care diagnostics \nfor their low cost and simplicity, but conventional formats rely on subjective \nvisual readout and lack quantitative accuracy and sensitivity. Here, we present \na smartphone-guided, artificial-intelligence-enhanced LFIA platform that unites \na rationally engineered dual-modal nanoprobe with on-device machine learning, \ndemonstrated for quantitative detection of Rift Valley fever virus (RVFV). The \nplatform combines three synergistic principles: advanced magnetic quantum dot \nnanoparticles (AQDs) that integrate a magnetic core and a quantum dot shell in a \nsingle label to provide both colorimetric and fluorescence signals while \nenabling magnetic preconcentration of the target antigen; multi-illumination \nimaging that turns an ordinary smartphone camera into a quantitative reader; and \na machine-learning model that extracts illumination-robust features to convert \nsignal into concentration, removing user subjectivity and lighting sensitivity. \nThis design markedly improves sensitivity over conventional gold-nanosphere LFIA \nin both modes, enables reliable quantification directly in serum without any \npretreatment, and outperforms naked-eye interpretation in blind testing. By \ncoupling dual-modal amplification and magnetic enrichment with AI-driven \nanalysis on widely available hardware, the AQD-LFIA platform offers a scalable \nroute toward next-generation point-of-care and at-home diagnostics.\n\n© 2026 American Chemical Society.\n\nDOI: 10.1021/acsami.6c10769\nPMID: 42760093", "extra": {"pmid": "42760093"}}
{"source": "pubmed", "title": "7. BMJ Open. 2026 Sep 18;16(9):e123879. doi: 10.1136/bmjopen-2026-123879.", "url": "https://pubmed.ncbi.nlm.nih.gov/42760089/", "summary": "7. BMJ Open. 2026 Sep 18;16(9):e123879. doi: 10.1136/bmjopen-2026-123879.\n\nAI-based applications supporting informal caregivers in childhood caries \nprevention and management: protocol for a scoping review.\n\nAnugrah PM(1), Emami E(2), Adiatman M(3), Darwita RRR(4).\n\nAuthor information:\n(1)Doctoral P", "content": "7. BMJ Open. 2026 Sep 18;16(9):e123879. doi: 10.1136/bmjopen-2026-123879.\n\nAI-based applications supporting informal caregivers in childhood caries \nprevention and management: protocol for a scoping review.\n\nAnugrah PM(1), Emami E(2), Adiatman M(3), Darwita RRR(4).\n\nAuthor information:\n(1)Doctoral Program in Dental Science, Universitas Indonesia, Depok, Jawa Barat, \nIndonesia panimatinanugrahh@gmail.com.\n(2)Faculty of Dentistry, McGill University, Montreal, Canada.\n(3)Department of Dental Public Health and Preventive Dentistry, Universitas \nIndonesia, Depok, West Java, Indonesia.\n(4)Department of Dental Public Health and Preventive Dentistry, Faculty of \nDentistry, Universitas Indonesia, Depok, West Java, Indonesia.\n\nINTRODUCTION: Dental caries remains highly prevalent among children despite \nbeing preventable. Limited access to dental services, workforce shortages and \nlow oral health literacy among informal caregivers contribute to its delayed \ndetection and poor oral health outcomes for children. As informal caregivers \nplay a vital role in children's daily oral health practices, innovative and \nscalable approaches are needed to support them. Artificial intelligence \n(AI)-based mobile health applications could potentially help engage these \ncaregivers in education, early detection, and other preventive activities. \nHowever, the evidence is still limited. The objective of this review is to map \nand characterise the existing evidence on caregiver-centred AI-based \napplications supporting the management and prevention of childhood dental \ncaries.\nMETHODS AND ANALYSIS: The scoping review will adhere to the Joanna Briggs \nInstitute (JBI) methodology guideline, using Preferred Reporting Items for \nSystematic Reviews and Meta-Analysis extension for Scoping Reviews (PRISMA-ScR) \nfor reporting structure and PRISMA-S (PRISMA Search Reporting Extension) for a \ncomprehensive search strategy. The search strategy will be developed in \ncollaboration with an expert librarian. An initial search will be conducted in \nEmbase (Ovid) and adapted for other databases (MEDLINE (Ovid), PubMed and \nScopus). Two reviewers will independently screen the titles, abstracts and \nfull-text articles, with a third resolving any conflicts. The included studies \nwill be selected according to the participants, concept and context (PCC) \nframework, in any setting and any country, that addresses AI-based applications \nfacilitated by informal caregivers for caries prevention and management in \nchildren. Articles published in English and primary studies will be included, \nwith no restrictions on publication date. Data will be charted and synthesised \nusing descriptive statistics and qualitative thematic mapping.\nETHICS AND DISSEMINATION: Ethics approval is not required for this review. These \nfindings will be disseminated in a peer-reviewed publication and at academic \nconferences. TRIAL REGISTRATION DETAILS: (https://osf.io/7akw5).\n\n© Author(s) (or their employer(s)) 2026. Re-use permitted under CC BY-NC. No \ncommercial re-use. See rights and permissions. Published by BMJ Group.\n\nDOI: 10.1136/bmjopen-2026-123879\nPMID: 42760089 [Indexed for MEDLINE]\n\nConflict of interest statement: Competing interests: None declared.", "extra": {"pmid": "42760089"}}
{"source": "pubmed", "title": "8. Reg Anesth Pain Med. 2026 Sep 18:rapm-2026-108180. doi: ", "url": "https://pubmed.ncbi.nlm.nih.gov/42760064/", "summary": "8. Reg Anesth Pain Med. 2026 Sep 18:rapm-2026-108180. doi: \n10.1136/rapm-2026-108180. Online ahead of print.\n\nReal-time AI-assisted nerve tracking for ultrasound-guided nerve identification: \na prospective randomized trial.\n\nKim J(1)(2), Nam SP(1), Yoo J(1), Hong Y(3), Joo H(3), Yoo Y(1), Moon JY(4)", "content": "8. Reg Anesth Pain Med. 2026 Sep 18:rapm-2026-108180. doi: \n10.1136/rapm-2026-108180. Online ahead of print.\n\nReal-time AI-assisted nerve tracking for ultrasound-guided nerve identification: \na prospective randomized trial.\n\nKim J(1)(2), Nam SP(1), Yoo J(1), Hong Y(3), Joo H(3), Yoo Y(1), Moon JY(4)(2).\n\nAuthor information:\n(1)Department of Anesthesiology and Pain Medicine, Seoul National University \nHospital, Seoul, Korea (the Republic of).\n(2)Department of Anesthesiology and Pain Medicine, Seoul National University \nCollege of Medicine, Seoul, Korea (the Republic of).\n(3)Seoul National University Hospital, Jongno-gu, Seoul, Korea (the Republic \nof).\n(4)Department of Anesthesiology and Pain Medicine, Seoul National University \nHospital, Seoul, Korea (the Republic of) jymoon0901@gmail.com.\n\nBACKGROUND: Artificial intelligence (AI)-assisted nerve tracking facilitates \nreal-time ultrasound-guided nerve identification, but its effect on independent \nnerve identification after AI support is removed remains uncertain.\nMETHODS: In this prospective randomized trial, 80 ultrasound novices (60 medical \nstudents and 20 anesthesiology residents) were assigned to either a control \ngroup receiving B-mode ultrasound training without AI assistance or an \nAI-assisted group using a real-time nerve tracking system (NerveTrack). \nFollowing a standardized lecture and a 10-minute hands-on training session, \nparticipants underwent assessment with the AI function disabled. The primary \noutcome was first-attempt independent identification success for the median \nnerve and supraclavicular brachial plexus. Secondary outcomes included \nperformance after additional training and self-reported confidence.\nRESULTS: 78 participants completed the study. First-attempt independent \nidentification success rates did not differ significantly between the control \nand AI-assisted groups for the median nerve (57.5% vs 65.8%; RR 1.14, 95% CI \n0.81 to 1.62; p=0.452) or the brachial plexus (35.0% vs 42.1%; RR 1.20, 95% CI \n0.68 to 2.11; p=0.519). Following additional AI-assisted training, participants \nin the AI-assisted group reported greater confidence in identifying the brachial \nplexus than those in the control group (4.7±0.5 vs 4.4±0.9; p=0.046). No \nsignificant differences were observed in independent identification success \nafter additional training.\nCONCLUSIONS: Brief AI-assisted training did not improve immediate independent \nultrasound identification of the median nerve or supraclavicular brachial plexus \nafter AI support was removed. These findings do not establish long-term \nretention, transfer to unfamiliar anatomy, or clinical procedural performance.\nTHE REGISTRATION NUMBER: NCT05857787.\n\n© American Society of Regional Anesthesia & Pain Medicine 2026. No commercial \nre-use. See rights and permissions. Published by BMJ Group.\n\nDOI: 10.1136/rapm-2026-108180\nPMID: 42760064\n\nConflict of interest statement: Competing interests: None declared.", "extra": {"pmid": "42760064"}}
{"source": "pubmed", "title": "9. J Immunother Cancer. 2026 Sep 18;14(9):e015014. doi: 10.1136/jitc-2026-015014.", "url": "https://pubmed.ncbi.nlm.nih.gov/42759982/", "summary": "9. J Immunother Cancer. 2026 Sep 18;14(9):e015014. doi: 10.1136/jitc-2026-015014.\n\nCT-based deep foundation model for predicting immune checkpoint \ninhibitor-induced pneumonitis risk in lung cancer.\n\nMuneer A(#)(1), Showkatian E(#)(1), Kitsel Y(1)(2), Saad MB(1), Sujit SJ(1), \nSoto F(3), Shroff GS(4", "content": "9. J Immunother Cancer. 2026 Sep 18;14(9):e015014. doi: 10.1136/jitc-2026-015014.\n\nCT-based deep foundation model for predicting immune checkpoint \ninhibitor-induced pneumonitis risk in lung cancer.\n\nMuneer A(#)(1), Showkatian E(#)(1), Kitsel Y(1)(2), Saad MB(1), Sujit SJ(1), \nSoto F(3), Shroff GS(4), Faiz SA(3), Ghanbar MI(5), Ismail SM(6), Vokes NI(2), \nCascone T(2), Le X(2), Zhang J(2), Byers LA(2), Jaffray D(6)(7), Chang JY(8), \nLiao Z(8), Naing A(9), Gibbons DL(2), Vaporciyan AA(10), Heymach JV(2), Suresh \nK(5), Altan M(2), Sheshadri A(3), Wu J(11)(2)(6).\n\nAuthor information:\n(1)Department of Imaging Physics, The University of Texas MD Anderson Cancer \nCenter, Houston, Texas, USA.\n(2)Department of Thoracic/Head and Neck Medical Oncology, The University of \nTexas MD Anderson Cancer Center, Houston, Texas, USA.\n(3)Department of Pulmonary Medicine, The University of Texas MD Anderson Cancer \nCenter, Houston, Texas, USA.\n(4)Department of Thoracic Imaging, The University of Texas MD Anderson Cancer \nCenter, Houston, Texas, USA.\n(5)Department of Pulmonary and Critical Care Medicine, Johns Hopkins University, \nBaltimore, Maryland, USA.\n(6)Institute for Data Science in Oncology, The University of Texas MD Anderson \nCancer Center, Houston, Texas, USA.\n(7)Department of Radiation Physics, The University of Texas MD Anderson Cancer \nCenter, Houston, TX, USA.\n(8)Department of Radiation Oncology, The University of Texas MD Anderson Cancer \nCenter, Houston, Texas, USA.\n(9)Department of Investigational Cancer Therapeutics, The University of Texas MD \nAnderson Cancer Center, Houston, TX, USA.\n(10)Department of Thoracic and Cardiovascular Surgery, The University of Texas \nMD Anderson Cancer Center, Houston, TX, USA.\n(11)Department of Imaging Physics, The University of Texas MD Anderson Cancer \nCenter, Houston, Texas, USA JWu11@mdanderson.org.\n(#)Contributed equally\n\nBACKGROUND: Immune checkpoint inhibitors (ICIs) have revolutionized cancer \ntherapy, but can cause serious immune-related adverse events, with pneumonitis \n(ICI-P) being among the most severe. Early identification of high-risk patients \nbefore ICI initiation is critical for close monitoring, timely intervention, and \noptimizing outcomes.\nPURPOSE: To develop and validate a deep learning foundation model to predict \nICI-P from baseline CT scans in patients with lung cancer.\nMETHODS: We designed the Checkpoint-Inhibitor Pneumonitis Hazard EstimatoR \n(CIPHER), a deep learning-powered foundation model combining contrastive \nlearning with a transformer-based masked autoencoder to predict ICI-P from \nbaseline CT scans in patients with lung cancer. Using self-supervised learning, \nCIPHER was pretrained on 590,284 CT slices from 2,500 patients with non-small \ncell lung cancer (NSCLC) to learn representations of heterogeneous lung \nparenchyma. Following pretraining, CIPHER was adapted to the internal MD \nAnderson Cancer Center NSCLC immunotherapy cohort of 347 patients, of whom 33 \ndeveloped adjudicated ICI-P. Fine-tuning was performed using 254 non-ICI-P \npatients only, and a held-out internal validation set of 93 patients, including \n33 ICI-P cases and 60 non-ICI-P controls, was reserved for evaluation. CIPHER \nwas benchmarked against clinical, radiomics, and ensemble comparator models and \nexternally validated in an independent Johns Hopkins NSCLC cohort of 116 \npatients, including 20 ICI-P cases and 96 non-ICI-P controls.\nRESULTS: In our internal immunotherapy cohort, CIPHER consistently distinguished \npatients at elevated risk of ICI-P from those without the event, with area under \nthe curves (AUCs) ranging from 0.77 to 0.85. In head-to-head benchmarking, \nCIPHER achieved an AUC of 0.83, outperforming the clinical, radiomics and \nensemble models. In the external validation cohort, CIPHER maintained high \nperformance (AUC=0.83; balanced accuracy=81.7%), exceeding the radiomics model \n(DeLong p=0.0318) and demonstrating superior specificity without sacrificing \nsensitivity. By contrast, the radiomics model, despite high sensitivity (85.0%), \nshowed markedly lower specificity (45.8%). Confusion matrix analyses confirmed \nCIPHER's robust classification, correctly identifying 80 of 96 non-ICI-P cases \nand 16 of 20 ICI-P cases.\nCONCLUSIONS: We developed and externally validated CIPHER, a CT-based imaging \nbiomarker for pretreatment ICI-P risk stratification in NSCLC. CIPHER shows \npromise as a non-invasive tool for ICI-P risk assessment but warrants \nprospective validation before clinical translation.\n\n© Author(s) (or their employer(s)) 2026. Re-use permitted under CC BY-NC. No \ncommercial re-use. See rights and permissions. Published by BMJ Group.\n\nDOI: 10.1136/jitc-2026-015014\nPMID: 42759982 [Indexed for MEDLINE]\n\nConflict of interest statement: Competing interests: TC reports speaker \nfees/honoraria (including travel/meeting expenses) from: ASCOPost, AstraZeneca, \nBio Ascend, Bristol Myers Squibb, Clinical Care Options, IDEOlogy Health, \nMedical Educator Consortium, Medscape, OncLive, PeerView, Physicians’ Education \nResource, Targeted Oncology; advisory role/consulting fees (including \ntravel/meeting expenses) from AstraZeneca, Bristol Myers Squibb, Daiichi Sankyo, \nGenentech, Johnson & Johnson, Merck, Nuvalent, oNKo-innate, Pfizer, and RAPT \nTherapeutics; and institutional research funding from AstraZeneca, Bristol Myers \nSquibb and Merck. NV receives consulting fees from Sanofi, Regeneron, Oncocyte, \nand Eli Lilly, and research funding from Mirati, outside the submitted work. DLG \nhas served on scientific advisory committees for Menarini Ricerche, 4D Pharma, \nOnconova, and Eli Lilly, and has received research support from Takeda, \nAstellas, NGM Biopharmaceuticals, Boehringer Ingelheim, and AstraZeneca. JVH \nreports scientific advisory roles for AstraZeneca, Boehringer Ingelheim, \nGenentech, GlaxoSmithKline, Eli Lilly, Novartis, Spectrum, EMD Serono, Sanofi, \nTakeda, Mirati Therapeutics, BMS, and Janssen; research support from \nAstraZeneca, Takeda, Boehringer Ingelheim, and Spectrum; and licensing fees from \nSpectrum. JZ reports research funding from Helius, Johnson and Johnson, Merck, \nNovartis, Summit, and personal fees from AstraZeneca, Catalyst, GenePlus, \nHelius, Hengrui, Innovent, Johnson and Johnson, Novartis, Oncohost, Takeda and \nVarian outside the submitted work. JW reports research funding from Siemens \nHealthcare. The remaining authors declare no competing interests.", "extra": {"pmid": "42759982"}}
{"source": "pubmed", "title": "10. BMJ Health Care Inform. 2026 Sep 18;33(1):e101830. doi: ", "url": "https://pubmed.ncbi.nlm.nih.gov/42759972/", "summary": "10. BMJ Health Care Inform. 2026 Sep 18;33(1):e101830. doi: \n10.1136/bmjhci-2025-101830.\n\nLeveraging time-series electronic health records with large language models for \nchronic kidney disease diagnosis in primary care.\n\nWu J(1)(2), Zheng Y(2), He Z(3), Zuo Q(2), Zhang W(4), Li P(5)(2), Luxia \nZ(1)", "content": "10. BMJ Health Care Inform. 2026 Sep 18;33(1):e101830. doi: \n10.1136/bmjhci-2025-101830.\n\nLeveraging time-series electronic health records with large language models for \nchronic kidney disease diagnosis in primary care.\n\nWu J(1)(2), Zheng Y(2), He Z(3), Zuo Q(2), Zhang W(4), Li P(5)(2), Luxia \nZ(1)(2).\n\nAuthor information:\n(1)National Institute of Health Data Science, Peking University, Beijing, China.\n(2)Advanced Institute of Information Technology, Peking University, Hangzhou, \nChina.\n(3)Independent Researcher, Hangzhou, China.\n(4)Weinan Municipal Health Commission, Weinan, Shaanxi, China.\n(5)National Institute of Health Data Science, Peking University, Beijing, China \npfli@bjmu.edu.cn.\n\nOBJECTIVES: Chronic kidney disease (CKD) presents a growing public health \nchallenge in China, exacerbated by low patient awareness and limited nephrology \nresources. This study evaluated the potential of large language models (LLMs) to \nsupport CKD diagnosis in primary care using time-series electronic health record \n(EHR) data.\nMETHODS: Longitudinal data were extracted from the EHR system of Weinan, China. \nAmong 29 963 adults meeting inclusion criteria, 2300 participants were randomly \nsampled to generate clinical vignettes. CKD status was determined using \ndiagnosis data, and individuals with early kidney injury indicated by laboratory \nabnormalities who lack a formal diagnosis were also identified. Using the \nDeepSeek LLM with prompt engineering, we generated binary CKD diagnoses and \nprobability scores based on EHR data. Diagnostic performance was evaluated using \naccuracy, sensitivity, specificity, F1 score, area under the curve (AUC) and the \ndetection rate for early kidney injury and compared with traditional machine \nlearning (ML) models.\nRESULTS: Using 1-month EHR data, the LLM achieved an accuracy of 87.0%, AUC of \n0.921, F1 score of 0.579, sensitivity of 89.1%, specificity of 86.8%, and \ndetection rate for early kidney injury of 42.2%. With 1-year data, sensitivity \nincreased to 92.2% and early detection to 52.6%.\nDISCUSSION: The LLM outperformed ML models across most metrics, achieving \ndiagnostic performance comparable to clinicians and further enhancing detection \nfor early kidney injury.\nCONCLUSIONS: LLMs applied to time-series EHR data may serve as a clinical \ndecision support tool to improve CKD diagnosis in primary care, with particular \nvalue in resource-limited settings.\n\n© Author(s) (or their employer(s)) 2026. Re-use permitted under CC BY-NC. No \ncommercial re-use. See rights and permissions. Published by BMJ Group.\n\nDOI: 10.1136/bmjhci-2025-101830\nPMID: 42759972 [Indexed for MEDLINE]\n\nConflict of interest statement: Competing interests: None declared.", "extra": {"pmid": "42759972"}}
{"source": "pubmed", "title": "11. BMJ Health Care Inform. 2026 Sep 18;33(1):e102188. doi: ", "url": "https://pubmed.ncbi.nlm.nih.gov/42759970/", "summary": "11. BMJ Health Care Inform. 2026 Sep 18;33(1):e102188. doi: \n10.1136/bmjhci-2026-102188.\n\nPredicting zero-dose vaccination status in 27 sub-Saharan African countries: a \nmachine learning approach.\n\nEndehabtu BF(1)(2), Addisu Taye E(3)(2).\n\nAuthor information:\n(1)Health Informatics, University of Gon", "content": "11. BMJ Health Care Inform. 2026 Sep 18;33(1):e102188. doi: \n10.1136/bmjhci-2026-102188.\n\nPredicting zero-dose vaccination status in 27 sub-Saharan African countries: a \nmachine learning approach.\n\nEndehabtu BF(1)(2), Addisu Taye E(3)(2).\n\nAuthor information:\n(1)Health Informatics, University of Gondar College of Medicine and Health \nSciences, Gondar, Ethiopia.\n(2)Center for Digital Health Implementation Science, University of Gondar \nCollege of Medicine and Health Sciences, Gondar, Ethiopia.\n(3)Health Informatics, University of Gondar College of Medicine and Health \nSciences, Gondar, Ethiopia eliyasaddisu12@gmail.com.\n\nOBJECTIVE: To develop a machine learning (ML)-based predictive model for \nassessing the risk of zero-dose children using Demographic Health Survey data in \nsub-Saharan Africa.\nMETHODS: This study analysed pooled Demographic and Health Survey data from 27 \ncountries in sub-Saharan Africa collected between 2016 and 2024. Data \npreprocessing included imputation, balancing of unequal classes and systematic \nfeature selection. Seven ML models were trained and evaluated using performance \nmetrics such as accuracy, recall and F1-score. Feature importance was \ninterpreted using SHapley Additive exPlanations (SHAP)-based analysis.\nRESULTS: Among the seven models evaluated, LightGBM demonstrated the best \noverall performance, achieving the highest accuracy (80%), sensitivity (91%), \nand area under the receiver operating characteristic curve (AUROC = 0.78). SHAP \nanalysis identified place of delivery, antenatal care attendance, maternal \ntetanus vaccination, household wealth and maternal education as top predictors.\nDISCUSSION: Our findings suggest that machine learning, particularly LightGBM \nand XGBoost, is moderately effective in predicting the risk of zero-dose \nvaccination among children. This approach facilitates the easy identification of \nchildren most at risk of missing life-saving vaccinations.\nCONCLUSION: The LightGBM model predict children at zero-dose risk, enabling \nearly identification and supporting intervention strategies. Future studies will \nscale these models using larger, routine datasets for practical implementation.\n\n© Author(s) (or their employer(s)) 2026. Re-use permitted under CC BY-NC. No \ncommercial re-use. See rights and permissions. Published by BMJ Group.\n\nDOI: 10.1136/bmjhci-2026-102188\nPMID: 42759970 [Indexed for MEDLINE]\n\nConflict of interest statement: Competing interests: None declared.", "extra": {"pmid": "42759970"}}
{"source": "pubmed", "title": "12. J Am Acad Psychiatry Law. 2026 Sep 18;54(3):275-281. doi: ", "url": "https://pubmed.ncbi.nlm.nih.gov/42759958/", "summary": "12. J Am Acad Psychiatry Law. 2026 Sep 18;54(3):275-281. doi: \n10.29158/JAAPL.260081-26.\n\nFuture Directions in the Evaluation of Forensic Cases Involving Artificial \nIntelligence \"Chatbots\".\n\nSimpson JR(1), Pierre JM(2).\n\nAuthor information:\n(1)Dr. Simpson is a Clinical Associate Professor, Departme", "content": "12. J Am Acad Psychiatry Law. 2026 Sep 18;54(3):275-281. doi: \n10.29158/JAAPL.260081-26.\n\nFuture Directions in the Evaluation of Forensic Cases Involving Artificial \nIntelligence \"Chatbots\".\n\nSimpson JR(1), Pierre JM(2).\n\nAuthor information:\n(1)Dr. Simpson is a Clinical Associate Professor, Department of Psychiatry and \nBehavioral Sciences, University of Nevada, Reno, Reno, NV; Dr. Pierre is a \nHealth Sciences Clinical Professor, Department of Psychiatry and Behavioral \nSciences, University of California, San Francisco, San Francisco, CA. \njrsimpsonmd@gmail.com.\n(2)Dr. Simpson is a Clinical Associate Professor, Department of Psychiatry and \nBehavioral Sciences, University of Nevada, Reno, Reno, NV; Dr. Pierre is a \nHealth Sciences Clinical Professor, Department of Psychiatry and Behavioral \nSciences, University of California, San Francisco, San Francisco, CA.\n\nDOI: 10.29158/JAAPL.260081-26\nPMID: 42759958", "extra": {"pmid": "42759958"}}
{"source": "pubmed", "title": "13. J Pain Symptom Manage. 2026 Sep 18:S0885-3924(26)01000-6. doi: ", "url": "https://pubmed.ncbi.nlm.nih.gov/42759901/", "summary": "13. J Pain Symptom Manage. 2026 Sep 18:S0885-3924(26)01000-6. doi: \n10.1016/j.jpainsymman.2026.09.019. Online ahead of print.\n\nClinicians vs. Artificial Intelligence in Predicting 28-Day ICU Mortality: A \nVignette Study.\n\nSiriseth S(1), Vattanavanit V(2).\n\nAuthor information:\n(1)Division of Internal", "content": "13. J Pain Symptom Manage. 2026 Sep 18:S0885-3924(26)01000-6. doi: \n10.1016/j.jpainsymman.2026.09.019. Online ahead of print.\n\nClinicians vs. Artificial Intelligence in Predicting 28-Day ICU Mortality: A \nVignette Study.\n\nSiriseth S(1), Vattanavanit V(2).\n\nAuthor information:\n(1)Division of Internal Medicine, Faculty of Medicine, Prince of Songkla \nUniversity, Hat Yai, Songkhla, 90110, Thailand.\n(2)Critical Care Medicine Unit, Division of Internal Medicine, Faculty of \nMedicine, Prince of Songkla University, Hat Yai, Songkhla, 90110, Thailand. \nElectronic address: vveerapong@gmail.com.\n\nCONTEXT: The surprise question (SQ) is an intuition-based tool for identifying \npatients who may benefit from palliative care, but its short-term prognostic \nperformance in critical illness remains uncertain. Large language models may \noffer standardized prognostic judgments.\nOBJECTIVES: To compare healthcare professionals and ChatGPT in predicting 28-day \nmortality among critically ill patients using the binary 28-day surprise \nquestion (SQ-28d).\nMETHODS: We conducted a multi-rater, vignette-based cross-sectional study in a \ntertiary medical intensive care unit (ICU) in Thailand. Sixty de-identified \nclinical vignettes were derived from adult patients with known 28-day outcomes. \nConsultant intensivists, internal medicine residents, ICU nurses, and ChatGPT \neach answered the SQ-28d. Primary outcome was correct prognostic classification; \nsecondary outcomes included sensitivity, specificity, predictive values, \nlikelihood ratios, area under the receiver operating characteristic curve \n(AUROC), and inter-rater agreement.\nRESULTS: . Correct classification differed significantly across groups \n(P = 0.046): consultant intensivists, 69.4%; residents, 66.4%; nurses, 64.1%; \nand ChatGPT, 55.0%; however, no pairwise comparison remained significant after \nBonferroni adjustment. AUROC was highest for consultant intensivists (0.734, 95% \nconfidence interval [CI]: 0.670-0.798), followed by residents (0.713), nurses \n(0.657), and ChatGPT (0.657). Consultant intensivists had the highest positive \nlikelihood ratio (2.25) and ChatGPT the lowest negative likelihood ratio (0.14). \nInter-rater agreement was substantial between intensivists and residents and \nmoderate between nurses and physicians.\nCONCLUSIONS: Healthcare professionals demonstrated higher 28-day mortality \nprediction accuracy than ChatGPT using the SQ. The SQ-28d may serve as a \npragmatic trigger for prognostic awareness and palliative care discussions, but \nshould not be used as a standalone prognostic tool.\n\nCopyright © 2026. Published by Elsevier Inc.\n\nDOI: 10.1016/j.jpainsymman.2026.09.019\nPMID: 42759901\n\nConflict of interest statement: Conflict of interests The authors declare that \nthey have no competing interests.", "extra": {"pmid": "42759901"}}
{"source": "pubmed", "title": "14. J Clin Epidemiol. 2026 Sep 18:112517. doi: 10.1016/j.jclinepi.2026.112517. ", "url": "https://pubmed.ncbi.nlm.nih.gov/42759856/", "summary": "14. J Clin Epidemiol. 2026 Sep 18:112517. doi: 10.1016/j.jclinepi.2026.112517. \nOnline ahead of print.\n\nLLM-Based Classifiers Can Reduce the Proportion of Records to Screen with \nMinimal Loss of Relevant Studies: A Case Study.\n\nNussbaumer-Streit B(1), Dobrescu A(2), Sharifan A(2), Klerings I(2), \nNe", "content": "14. J Clin Epidemiol. 2026 Sep 18:112517. doi: 10.1016/j.jclinepi.2026.112517. \nOnline ahead of print.\n\nLLM-Based Classifiers Can Reduce the Proportion of Records to Screen with \nMinimal Loss of Relevant Studies: A Case Study.\n\nNussbaumer-Streit B(1), Dobrescu A(2), Sharifan A(2), Klerings I(2), \nNeubauer-Bruckner C(2), Nowak C(2), Gartlehner G(3).\n\nAuthor information:\n(1)Department for Evidence-based Medicine and Evaluation, University for \nContinuing Education Krems, Austria. Electronic address: \nBarbara.nussbaumer-streit@donau-uni.ac.at.\n(2)Department for Evidence-based Medicine and Evaluation, University for \nContinuing Education Krems, Austria.\n(3)Department for Evidence-based Medicine and Evaluation, University for \nContinuing Education Krems, Austria; Center for Health Economics, Methods, and \nEvidence Synthesis, RTI International, Research Triangle Park, NC, USA.\n\nBACKGROUND: The Epistemonikos Sustainable Knowledge Platform (SKP) integrates a \nsuite of machine learning tools designed to make evidence synthesis more \nefficient, including access to the Epistemonikos Database of Trials (ED-Trials). \nOne feature is the use of classifiers combined with large language models (LLMs) \nto support abstract screening during study selection. LLM-based classifiers can \nbe applied to populations, interventions, and study design features. Their \nprimary aim is to reduce the proportion of records needed to be screenedby \nautomatically excluding records with a high probability of being ineligible. We \naimed to evaluate SKP's LLM-based classifiers on abstract screenings of records \nretrieved via: (scenario 1) a traditional systematic literature search (original \n), and (scenario 2) a search conducted within ED-Trials and subsequently \nforwarded to SKP.\nMETHODS: We used data from a completed but at the time unpublished systematic \nreview and network meta-analysis (NMA)as reference standards. We evaluated 29 \nLLM-based classifier combinations in both scenarios and assessed the proportion \nof records excluded by automated screening as well as whether any relevant study \nwas wrongly excluded and therefore missed. To determine the impact of missed \nstudies, we re-ran the network meta-analyses of the original review, excluding \nmissed studies from the model and we evaluated how missed studies impact the \ncertainty-of-evidence ratings.\nRESULTS: LLM-based classifiers excluded by automated screening 5.8-46.6% of \n1,129 records identified by the original search and 7.7-34.3% of 1,003 records \nbased on the SKP search. More specific classifiers excluded more records than \nbroader ones. Of the 29 combinations applied to the original search results, 10 \nincorrectly excluded the same two relevant studies. Based on the SKP review, \nfour combinations missed one study, and 10 combinations missed two studies. \nHowever, exclusion of these studies resulted in minimal changes to NMA-effect \nestimates (risk ratio changes of 0.0-0.04 in both scenarios) and did not alter \ncertainty-of-evidence ratings.\nDISCUSSION: LLM-based classifiers are a promising strategy for reducing the \namount of records to screen while keeping the risk of missing relevant studies \nlow. However, further evaluations using multiple-use cases across different \nmedical topics are necessary to learn more about the generalizability of our \nresults.\n\nCopyright © 2026. Published by Elsevier Inc.\n\nDOI: 10.1016/j.jclinepi.2026.112517\nPMID: 42759856", "extra": {"pmid": "42759856"}}
{"source": "pubmed", "title": "15. Bioresour Technol. 2026 Sep 18:135884. doi: 10.1016/j.biortech.2026.135884. ", "url": "https://pubmed.ncbi.nlm.nih.gov/42759849/", "summary": "15. Bioresour Technol. 2026 Sep 18:135884. doi: 10.1016/j.biortech.2026.135884. \nOnline ahead of print.\n\nβ-1,3-Glucan biodegradation and utilization: Enzyme discovery, engineering, and \nindustrial applications.\n\nFang K(1), Yuan L(2), Cui Y(1), Zheng X(3), Jia R(2), Liu K(4), Li S(5).\n\nAuthor informa", "content": "15. Bioresour Technol. 2026 Sep 18:135884. doi: 10.1016/j.biortech.2026.135884. \nOnline ahead of print.\n\nβ-1,3-Glucan biodegradation and utilization: Enzyme discovery, engineering, and \nindustrial applications.\n\nFang K(1), Yuan L(2), Cui Y(1), Zheng X(3), Jia R(2), Liu K(4), Li S(5).\n\nAuthor information:\n(1)State Key Laboratory of Microbial Technology, Shandong University, Qingdao \n266237, Shandong, China.\n(2)Angel Yeast (Binzhou) Co., Ltd., Binzhou 256600, Shandong, China.\n(3)Angel Yeast (Binzhou) Co., Ltd., Binzhou 256600, Shandong, China; National \nKey Laboratory of Agricultural Microbiology, AngelYeast Co., Ltd., Yichang \n443003, Hubei, China.\n(4)State Key Laboratory of Microbial Technology, Shandong University, Qingdao \n266237, Shandong, China; Gan Su Institute of Shandong University, Lanzhou \n730000, Gansu, China; Laboratory for Marine Biology and Biotechnology, Qingdao \nMarine Science and Technology Center, Qingdao 266237, Shandong, China. \nElectronic address: liuk@sdu.edu.cn.\n(5)State Key Laboratory of Microbial Technology, Shandong University, Qingdao \n266237, Shandong, China; Laboratory for Marine Biology and Biotechnology, \nQingdao Marine Science and Technology Center, Qingdao 266237, Shandong, China. \nElectronic address: lishengying@sdu.edu.cn.\n\nβ-1,3-Glucans are structurally diverse polysaccharides widely distributed in \nmarine algae, fungi, and plants, and their enzymatic depolymerization is \nimportant for carbon turnover, biomass processing, and polysaccharide \nvalorization. β-1,3-Glucanases are diverse glycoside hydrolases (GHs) that \ncleave β-1,3-glycosidic linkages and enable the biological transformation and \nindustrial utilization of these polymers. These enzymes provide a mechanistic \nand technological bridge between β-1,3-glucan metabolism in nature and the \ndevelopment of sustainable processes for biomass valorization and functional \nproduct generation. This review summarizes recent advances in β-1,3-glucan \nenzymatic bioconversion by integrating GH family classification, structural \ndiversity, catalytic mechanisms, enzyme discovery, molecular engineering, \nproduction optimization, and emerging applications. Particular attention is \ngiven to how variations in fold architecture, active-site organization, and \ncarbohydrate-binding modules (CBMs) collectively determine substrate \nrecognition, catalytic mode, and product specificity toward structurally \ndistinct β-1,3-glucans such as laminarin, curdlan, callose, and fungal cell-wall \nglucans. For an application-oriented perspective, we further discuss how \nsequence-guided discovery, structure-based engineering, and heterologous \nproduction can be integrated to improve thermostability, catalytic efficiency, \nsubstrate adaptability, product control, and scalable enzyme production. \nApplications in agriculture, fermentation processing, biomass refining, \nfunctional oligosaccharide production, and environmental management are also \ndiscussed. Finally, we outline remaining challenges and emerging opportunities, \nparticularly regarding the molecular basis of substrate recognition and product \nspecificity, as well as artificial intelligence (AI) assisted predictive design \nof customized β-1,3-glucanases for next-generation polysaccharide biotechnology.\n\nCopyright © 2026. Published by Elsevier Ltd.\n\nDOI: 10.1016/j.biortech.2026.135884\nPMID: 42759849\n\nConflict of interest statement: Declaration of competing interest The authors \ndeclare that they have no known competing financial interests or personal \nrelationships that could have appeared to influence the work reported in this \npaper.", "extra": {"pmid": "42759849"}}
{"source": "pubmed", "title": "16. Eur J Vasc Endovasc Surg. 2026 Sep 18:S1078-5884(26)00907-X. doi: ", "url": "https://pubmed.ncbi.nlm.nih.gov/42759832/", "summary": "16. Eur J Vasc Endovasc Surg. 2026 Sep 18:S1078-5884(26)00907-X. doi: \n10.1016/j.ejvs.2026.09.035. Online ahead of print.\n\nComprehensive Artificial Intelligence Based Analysis of Pre- and Post-operative \nComputed Tomography Scans in Patients Undergoing Endovascular Abdominal Aortic \nRepair.\n\nGarrido", "content": "16. Eur J Vasc Endovasc Surg. 2026 Sep 18:S1078-5884(26)00907-X. doi: \n10.1016/j.ejvs.2026.09.035. Online ahead of print.\n\nComprehensive Artificial Intelligence Based Analysis of Pre- and Post-operative \nComputed Tomography Scans in Patients Undergoing Endovascular Abdominal Aortic \nRepair.\n\nGarrido-Oliver J(1), Tenezaca-Sari X(2), Ribeiro T(3), Dux-Santoy L(4), Capó \nXF(2), Lopez-Gutierrez P(1), Cuellar-Calabria H(5), Gonçalves FB(6), \nBellmunt-Montoya S(2), Guala A(7); of the collaborative study group.\n\nCollaborators: Bragulat-Arévalo M(8), Catala-Santarrufina A(9), \nRodriguez-Palomares J(10).\n\nAuthor information:\n(1)Vall d'Hebron Institut de Recerca, Barcelona, Spain; Department of Medicine, \nUniversitat Autònoma de Barcelona, Bellaterra, Spain.\n(2)Vascular and Endovascular Surgery, Hospital Universitari Vall d'Hebron, \nBarcelona, Spain.\n(3)Vascular Surgery Department, ULS São José, Lisbon, Portugal; Centro Clínico \nAcadémico de Lisboa, Lisboa, Portugal.\n(4)Vall d'Hebron Institut de Recerca, Barcelona, Spain. Electronic address: \nlydia.duxsantoy@vhir.org.\n(5)Vall d'Hebron Institut de Recerca, Barcelona, Spain; Department of Medicine, \nUniversitat Autònoma de Barcelona, Bellaterra, Spain; Radiology Department, \nHospital Universitari Vall d'Hebron, Barcelona, Spain.\n(6)Vascular Surgery Department, ULS São José, Lisbon, Portugal; Centro Clínico \nAcadémico de Lisboa, Lisboa, Portugal; NOVA Medical School, Faculdade de \nCiências Médicas, NMS | FCM, Universidade Nova de Lisboa, Lisboa, Portugal; \nHospital CUF Tejo, Lisboa, Portugal.\n(7)Vall d'Hebron Institut de Recerca, Barcelona, Spain; CIBER-CV, Instituto de \nSalud Carlos III, Madrid, Spain.\n(8)Vall d'Hebron Institut de Recerca, Barcelona, Spain; Department of Medicine, \nUniversitat Autònoma de Barcelona, Bellaterra, Spain; CIBER-CV, Instituto de \nSalud Carlos III, Madrid, Spain.\n(9)Vall d'Hebron Institut de Recerca, Barcelona, Spain.\n(10)Vall d'Hebron Institut de Recerca, Barcelona, Spain; Department of Medicine, \nUniversitat Autònoma de Barcelona, Bellaterra, Spain; CIBER-CV, Instituto de \nSalud Carlos III, Madrid, Spain; Cardiology Department, Hospital Universitari \nVall d'Hebron, Barcelona, Spain.\n\nOBJECTIVE: Endovascular aortic repair (EVAR) requires lifelong imaging \nsurveillance to monitor potential complications such as endoleaks (EL) or \naneurysm growth. This study aimed to develop and validate a comprehensive \nartificial intelligence (AI) pipeline for analysing pre- and post-EVAR contrast \nenhanced computed tomography angiographies (CTAs) in abdominal aortic aneurysm \n(AAA) patients.\nMETHODS: Consecutive patients who underwent elective EVAR with available pre- \nand post-EVAR CTA were retrospectively identified at two independent centres. \nFour convolutional neural networks (nnU-Net) were developed to automatically \nidentify pre-EVAR (lumen, thrombus, six anatomical landmarks) and post-EVAR CTAs \n(lumen, excluded aorta, six anatomical and three prosthetic landmarks) and to \ndetect EL. Maximum AAA diameter and its progression during follow up were \nautomatically extracted and compared with manual annotations across experts and \nprotocols. Data from the internal centre were used to train and validate the \nmodels, while data from the external centre served for external validation.\nRESULTS: A total of 372 patients (1 133 CT scans) were included: 135 patients \n(295 CTAs) for model development and 237 (838 CTAs) for external validation. \nOverall, age was 74 [69 - 79] years; 15 were female (4%), follow up was 8.1 \n[3.1, 27.9] months, pre-EVAR maximum AAA diameter was 59 [55, 68] mm, and 150 \npatients developed an EL during follow up. All AI segmentations were excellent \n(all Dice scores > 0.9) and landmarks were accurately identified, with errors \ncomparable with the interobserver variability. EL detection achieved an area \nunder the curve of 0.87 and 0.85 in the internal and external cohorts, \nrespectively. Differences between AI derived and manual maximum AAA diameters \nwere minor (median error ≤ 2 mm), with excellent correlations across centres \n(all R ≥ 0.97, p < .001; intraclass correlation ≥ 0.97). Disagreement in follow \nup diameter change between AI and manual measurements was lower than the \ninterobserver variability.\nCONCLUSION: The AI based pipeline enabled accurate, generalisable, and \nexplainable assessment of AAA patients, including both pre-EVAR evaluation and \npost-EVAR follow up.\n\nCopyright © 2026 The Author(s). Published by Elsevier B.V. All rights reserved.\n\nDOI: 10.1016/j.ejvs.2026.09.035\nPMID: 42759832", "extra": {"pmid": "42759832"}}
{"source": "pubmed", "title": "17. Gastrointest Endosc. 2026 Sep 18:S0016-5107(26)07066-5. doi: ", "url": "https://pubmed.ncbi.nlm.nih.gov/42759726/", "summary": "17. Gastrointest Endosc. 2026 Sep 18:S0016-5107(26)07066-5. doi: \n10.1016/j.gie.2026.09.022. Online ahead of print.\n\nEndoscopic Artificial Intelligence for Invasion-Depth Prediction in Superficial \nUpper Gastrointestinal Cancer: A Diagnostic Test Accuracy Meta-Analysis.\n\nLee H(1), Min YW(2), Kim S(3", "content": "17. Gastrointest Endosc. 2026 Sep 18:S0016-5107(26)07066-5. doi: \n10.1016/j.gie.2026.09.022. Online ahead of print.\n\nEndoscopic Artificial Intelligence for Invasion-Depth Prediction in Superficial \nUpper Gastrointestinal Cancer: A Diagnostic Test Accuracy Meta-Analysis.\n\nLee H(1), Min YW(2), Kim S(3), Yoo H(3), Kim TJ(4).\n\nAuthor information:\n(1)Department of Medicine, Samsung Medical Center, Sungkyunkwan University \nSchool of Medicine, Seoul, Republic of Korea; Samsung Precision Genome Medicine \nInstitute, Samsung Medical Center, Seoul, Republic of Korea. Electronic address: \nleehyuk@skku.edu.\n(2)Department of Medicine, Samsung Medical Center, Sungkyunkwan University \nSchool of Medicine, Seoul, Republic of Korea; Samsung Precision Genome Medicine \nInstitute, Samsung Medical Center, Seoul, Republic of Korea.\n(3)Samsung Precision Genome Medicine Institute, Samsung Medical Center, Seoul, \nRepublic of Korea.\n(4)Division of Gastroenterology, Department of Internal Medicine, Hallym \nUniversity Dongtan Sacred Heart Hospital, Hallym University College of Medicine, \nHwaseong, Republic of Korea.\n\nBACKGROUND AND AIMS: Pre-resection prediction of submucosal invasion is pivotal \nfor endoscopic submucosal dissection (ESD) candidate selection in early gastric \nand esophageal cancers. We performed a diagnostic test accuracy meta-analysis of \nartificial intelligence (AI) for invasion-depth prediction.\nMETHODS: PubMed/MEDLINE, Embase, Cochrane CENTRAL, and Web of Science were \nsearched through 11 June 2026. Eligible studies applied AI to endoscopic images \nor videos of superficial upper gastrointestinal cancers with histopathological \nreference standards. Pooled sensitivity and specificity were estimated with a \nbivariate random-effects model; subgroup and meta-regression analyses examined \nvalidation framework, histology, AI architecture, and imaging modality. Risk of \nbias was assessed with QUADAS-2.\nRESULTS: Twenty-six studies were eligible; 18 (14 gastric adenocarcinoma, 4 \nesophageal squamous cell carcinoma) contributed to the primary pool of 6,568 \nvalidation/test patients or lesions. Pooled sensitivity was 0.77 (95% CI \n0.71-0.82) and specificity 0.88 (0.85-0.90); positive and negative likelihood \nratios were 6.30 and 0.26. Externally validated specificity was lower than \ninternally validated specificity (0.85 vs. 0.93); validation framework was the \nonly covariate significantly associated with performance in prespecified \nmeta-regression (specificity pseudo-R2 77.9%), whereas AI architecture and \nimaging modality were not. In within-study paired comparisons (n = 8 studies), \nmean per-study sensitivity was higher for AI alone than for unassisted \nendoscopists (82.5% vs. 63.8%; +18.7 percentage points; paired t = 3.29, p = \n0.013); specificity did not differ significantly. Limited AI-assisted \nreader-comparison data were directionally consistent with a second-reader role \nbut remained exploratory.\nCONCLUSIONS: Endoscopic AI shows clinically meaningful, but \nvalidation-dependent, accuracy for invasion-depth prediction, supporting \nstronger rule-in than rule-out utility for submucosal invasion. Its most \nplausible near-term role is second-reader support rather than stand-alone \ndetermination of ESD candidacy; independent patient-level external validation \nshould be considered essential before clinical deployment. (PROSPERO \nregistration: CRD420261400366).\n\nCopyright © 2026 American Society for Gastrointestinal Endoscopy. Published by \nElsevier Inc. All rights reserved.\n\nDOI: 10.1016/j.gie.2026.09.022\nPMID: 42759726", "extra": {"pmid": "42759726"}}
{"source": "pubmed", "title": "18. J Am Acad Dermatol. 2026 Sep 18:S0190-9622(26)03559-0. doi: ", "url": "https://pubmed.ncbi.nlm.nih.gov/42759719/", "summary": "18. J Am Acad Dermatol. 2026 Sep 18:S0190-9622(26)03559-0. doi: \n10.1016/j.jaad.2026.09.050. Online ahead of print.\n\nImplementing a Domain-Adapted GPT Model to Support Genetic Counseling in a \nGenodermatology Clinic.\n\nUdupa M(1), Mija LA(2), Bendayan E(1), Alamari A(3), Cutumisu M(4), Li Y(5), \nJoly", "content": "18. J Am Acad Dermatol. 2026 Sep 18:S0190-9622(26)03559-0. doi: \n10.1016/j.jaad.2026.09.050. Online ahead of print.\n\nImplementing a Domain-Adapted GPT Model to Support Genetic Counseling in a \nGenodermatology Clinic.\n\nUdupa M(1), Mija LA(2), Bendayan E(1), Alamari A(3), Cutumisu M(4), Li Y(5), \nJoly-Chevrier M(2), Leblanc J(6), Tardif J(6), Martinez-Jaramillo E(7), Nedjar \nH(8), Rahme E(8), Foulkes WD(9), Marino TC(10), Netchiporouk E(11).\n\nAuthor information:\n(1)Faculty of Medicine and Health Sciences, McGill University, Montreal, QC, \nCanada.\n(2)Faculté de Médecine, Université de Montréal, Montreal, QC, Canada.\n(3)Division of Dermatology, Department of Medicine, McGill University Health \nCentre, Montreal, QC, Canada.\n(4)Department of Educational and Counselling Psychology, Faculty of Education, \nMcGill University, Montreal, QC, Canada.\n(5)Quantitative Life Sciences Program, McGill University, Montreal, QC, Canada; \nSchool of Computer Science, McGill University, Montreal, QC, Canada.\n(6)Medical Genetics Service, Department of Laboratory Medicine, CIUSSS \nSaguenay-Lac-Saint-Jean, Qc, Canada.\n(7)Division of Experimental Medicine, McGill University Health Centre, Montreal, \nQC, Canada.\n(8)Division of Clinical Epidemiology, Department of Medicine, McGill University, \nMontreal, QC, Canada; Centre for Outcomes Research and Evaluation, Research \nInstitute of the McGill University Health Centre, Montréal, QC, Canada.\n(9)Division of Experimental Medicine, McGill University Health Centre, Montreal, \nQC, Canada; Division of Medical Genetics, Department of Specialized Medicine, \nMcGill University, Montreal, QC, Canada; Cancer Research Program, Research \nInstitute of the McGill University Health Centre, Montreal, QC, Canada; \nDepartment of Human Genetics, McGill University, Montreal, QC, Canada; Cancer \nAxis, Lady Davis Institute, Jewish General Hospital, Montreal, QC, Canada.\n(10)Division of Medical Genetics, Department of Specialized Medicine, McGill \nUniversity, Montreal, QC, Canada.\n(11)Division of Dermatology, Department of Medicine, McGill University Health \nCentre, Montreal, QC, Canada; Division of Experimental Medicine, McGill \nUniversity Health Centre, Montreal, QC, Canada. Electronic address: \nelena.netchiporouk@mcgill.ca.\n\nDOI: 10.1016/j.jaad.2026.09.050\nPMID: 42759719", "extra": {"pmid": "42759719"}}
{"source": "pubmed", "title": "19. Clin Med (Lond). 2026 Sep 18:100642. doi: 10.1016/j.clinme.2026.100642.", "url": "https://pubmed.ncbi.nlm.nih.gov/42759707/", "summary": "19. Clin Med (Lond). 2026 Sep 18:100642. doi: 10.1016/j.clinme.2026.100642.\nOnline  ahead of print.\n\nArtificial Intelligence for Detecting Fall Events, Assessing Fall Risks, and \nPreventing Fall Occurrences in the Elderly: A Narrative Review.\n\nShih CC(1), Lin SH(2), Yen WJ(3), Cheng GC(4), Chen WE(5", "content": "19. Clin Med (Lond). 2026 Sep 18:100642. doi: 10.1016/j.clinme.2026.100642.\nOnline  ahead of print.\n\nArtificial Intelligence for Detecting Fall Events, Assessing Fall Risks, and \nPreventing Fall Occurrences in the Elderly: A Narrative Review.\n\nShih CC(1), Lin SH(2), Yen WJ(3), Cheng GC(4), Chen WE(5), Shiao CC(6).\n\nAuthor information:\n(1)Department of Neurology, Mennonite Christian Hospital; No. 44, Minquan Road, \nHualien City, Hualien County 97059, Taiwan ROC. Electronic address: \nCcshih@mch.org.tw.\n(2)Department of Rehabilitation, Camillian Saint Mary's Hospital Luodong; No. \n160, Zhongzheng S. Rd., Luodong Township, 265, Yilan County, Taiwan ROC. \nElectronic address: phillip31150@hotmail.com.\n(3)Department of Rehabilitation, Camillian Saint Mary's Hospital Luodong; No. \n160, Zhongzheng S. Rd., Luodong Township, 265, Yilan County, Taiwan ROC. \nElectronic address: wei25180@gmail.com.\n(4)Department of Rehabilitation, Camillian Saint Mary's Hospital Luodong; No. \n160, Zhongzheng S. Rd., Luodong Township, 265, Yilan County, Taiwan ROC. \nElectronic address: V0909345769@gmail.com.\n(5)Department of Computer Science and Information Engineering, National Ilan \nUniversity; No.1, Section 1, Shennong Rd., Yilan City, Yilan County 260, Taiwan \nROC. Electronic address: wechen@niu.edu.tw.\n(6)Division of Nephrology, Department of Internal Medicine, Camillian Saint \nMary`s Hospital Luodong, No. 160, Zhongzheng S. Rd., Luodong Township, 265, \nYilan County, Taiwan ROC. Electronic address: chungyy2001@yahoo.com.tw.\n\nFalls in the elderly represent a critical public health crisis, necessitating a \nshift from reactive to proactive management. This review evaluates artificial \nintelligence (AI) across four primary modalities: wearable sensors, vision-based \nsystems, ambient devices, and natural language processing and large language \nmodels applied to electronic health records. Machine learning and deep learning \nalgorithms enable real-time detection, near-fall identification, and predictive \nrisk stratification. Hybrid edge-cloud frameworks and multimodal fusion enhance \naccuracy and scalability. However, challenges include the \"simulation gap\" in \ntraining data, \"black box\" interpretability, and privacy concerns, \"alarm \nfatigue\" among nursing staff, and \"class imbalance.\" Future directions emphasize \nhybrid systems, integrating multiple sensor streams with edge-cloud computing, \nto enhance ecological validity and responsiveness. Ultimately, seamless \nintegration of interpretable AI into clinical workflows offers a transformative \npath toward personalized prevention, reducing fall-related morbidity and \nempowering elderly independence.\n\nCopyright © 2026. Published by Elsevier Ltd.\n\nDOI: 10.1016/j.clinme.2026.100642\nPMID: 42759707\n\nConflict of interest statement: Conflict of interest Statement None declared \nDeclaration of Competing Interest The authors declare that they have no known \ncompeting financial interests or personal relationships that could have appeared \nto influence the work reported in this paper.", "extra": {"pmid": "42759707"}}
{"source": "pubmed", "title": "20. Clin Gastroenterol Hepatol. 2026 Sep 18:S1542-3565(26)00682-8. doi: ", "url": "https://pubmed.ncbi.nlm.nih.gov/42759683/", "summary": "20. Clin Gastroenterol Hepatol. 2026 Sep 18:S1542-3565(26)00682-8. doi: \n10.1016/j.cgh.2026.09.008. Online ahead of print.\n\nImplementing Artificial Intelligence in a Gastroenterology Practice.\n\nAisenberg J(1), Bamji N(1), Hertz D(1), Allen JI(2).\n\nAuthor information:\n(1)New York Gastroenterology Ass", "content": "20. Clin Gastroenterol Hepatol. 2026 Sep 18:S1542-3565(26)00682-8. doi: \n10.1016/j.cgh.2026.09.008. Online ahead of print.\n\nImplementing Artificial Intelligence in a Gastroenterology Practice.\n\nAisenberg J(1), Bamji N(1), Hertz D(1), Allen JI(2).\n\nAuthor information:\n(1)New York Gastroenterology Associates LLP, 311 East 79(th) Street, New York, \nNY 10075.\n(2)Department of Medicine. University of Michigan School of Medicine Ann Arbor, \nMichigan (Retired).\n\nDOI: 10.1016/j.cgh.2026.09.008\nPMID: 42759683", "extra": {"pmid": "42759683"}}
{"source": "pubmed", "title": "21. J Control Release. 2026 Sep 18:115370. doi: 10.1016/j.jconrel.2026.115370. ", "url": "https://pubmed.ncbi.nlm.nih.gov/42759666/", "summary": "21. J Control Release. 2026 Sep 18:115370. doi: 10.1016/j.jconrel.2026.115370. \nOnline ahead of print.\n\nImmune memory engineering in mRNA therapeutics.\n\nWen J(1), Tao W(2), Zhou X(3).\n\nAuthor information:\n(1)McWilliams School of Biomedical Informatics, The University of Texas Health \nScience Center ", "content": "21. J Control Release. 2026 Sep 18:115370. doi: 10.1016/j.jconrel.2026.115370. \nOnline ahead of print.\n\nImmune memory engineering in mRNA therapeutics.\n\nWen J(1), Tao W(2), Zhou X(3).\n\nAuthor information:\n(1)McWilliams School of Biomedical Informatics, The University of Texas Health \nScience Center at Houston, Houston, TX 77030, USA. Electronic address: \nJianguo.Wen@uth.tmc.edu.\n(2)Molecular Diagnostics Laboratory, Division of Pathology-Lab Medicine, \nUniversity of Texas MD Anderson Cancer Center, Houston, TX 77030, USA.\n(3)McWilliams School of Biomedical Informatics, The University of Texas Health \nScience Center at Houston, Houston, TX 77030, USA. Electronic address: \nXiaobo.Zhou@uth.tmc.edu.\n\nMessenger RNA (mRNA) therapeutics have revolutionized vaccine development and \nare rapidly expanding into cancer immunotherapy, infectious diseases, \nregenerative medicine, and protein replacement therapies. While considerable \nefforts have focused on maximizing antigen expression and short term immune \nresponses, durable therapeutic efficacy ultimately depends on the establishment \nof long lasting immune memory. Emerging evidence suggests that immune memory is \nnot merely a passive consequence of antigen expression but can be actively \nshaped through the rational engineering of antigen persistence, delivery, and \nimmune programming. In this review, we propose immune memory engineering as a \nunifying framework for mRNA therapeutics that achieve durable immune protection. \nWe examine three interconnected biological determinants of immune memory, \nincluding antigen persistence, antigen trafficking, and innate and spatial \nprogramming, and discuss how LNP engineering, formulation parameters, and tissue \nspecific delivery quantitatively influence memory formation. We further \nhighlight how artificial intelligence and multiomics approaches can integrate \nthese determinants into predictive design strategies for optimizing durable \nimmune responses. Together, these advances position immune memory engineering as \nan emerging design paradigm for developing next generation mRNA therapeutics \nwith durable, controllable, and increasingly personalized immune protection.\n\nCopyright © 2026. Published by Elsevier B.V.\n\nDOI: 10.1016/j.jconrel.2026.115370\nPMID: 42759666\n\nConflict of interest statement: Declaration of competing interest The authors \ndeclare no competing interests.", "extra": {"pmid": "42759666"}}
{"source": "pubmed", "title": "22. Appl Clin Inform. 2026 Sep 18. doi: 10.1055/a-2962-8004. Online ahead of", "url": "https://pubmed.ncbi.nlm.nih.gov/42759594/", "summary": "22. Appl Clin Inform. 2026 Sep 18. doi: 10.1055/a-2962-8004. Online ahead of\nprint.\n\nEvaluation of a Large Language Model Discharge Summary Hospital Course Tool: \nImproved Quality but Longer Documentation Time.\n\nDobek AJ(1), Naeem F(1), DeBenedictus C(2), Patel S(3), Suarez M(4), Suarez \nS(1)(5).\n\nA", "content": "22. Appl Clin Inform. 2026 Sep 18. doi: 10.1055/a-2962-8004. Online ahead of\nprint.\n\nEvaluation of a Large Language Model Discharge Summary Hospital Course Tool: \nImproved Quality but Longer Documentation Time.\n\nDobek AJ(1), Naeem F(1), DeBenedictus C(2), Patel S(3), Suarez M(4), Suarez \nS(1)(5).\n\nAuthor information:\n(1)University of Miami Miller School of Medicine, Department of Medicine, \nFlorida, United States, Miami.\n(2)University of Miami Miller School of Medicine, Division of Internal \nMedicine-Pediatrics, Department of Medicine, Florida, United States, Miami.\n(3)University of Miami Miller School of Medicine, Division of Hospital Medicine, \nDepartment of Medicine, Florida, United States, Miami.\n(4)University of Miami Miller School of Medicine, Division of General Internal \nMedicine, Department of Medicine, Florida, United States, Miami.\n(5)University of Miami Miller School of Medicine, HHospital Medicine IT Lead, \nDivision of Hospital Medicine, Florida, United States, Miami.\n\nBACKGROUND: No studies have examined the effect of artificial intelligence \n(AI)-generated hospital courses on discharge summary documentation time.\nOBJECTIVES: To evaluate the effect of an AI-generated hospital course tool on \ndischarge summary edit time, quality, and clinician perceptions.\nMETHODS: Observational study between a \"Pre-AI period\" (March 16, 2025 to \nNovember 19, 2025) and a \"Post-AI period\" (November 20, 2025 to February 11, \n2026). An AI hospital course tool using a GPT 4.1 model to generate a draft \nhospital course integrated into mandatory discharge summary templates on \nNovember 20, 2025. We included 8,298 hospitalized adults cared for by hospital \nmedicine clinicians. The primary outcome was discharge summary edit time, \ndefined as time from note creation to signature across all sessions. A subset of \n27 encounters was reviewed for cohesiveness, comprehensiveness, conciseness, \npotential for harm, and overall quality. Thirty-six clinicians completed a \nsurvey assessing usability, efficiency, trustworthiness, and quality.\nRESULTS: Implementation of the AI tool was not associated with a significant \nchange in discharge summary edit time between study periods (6.60 [IQR 3.4-13.5] \nvs 6.28 [IQR 3.2-13.0] minutes, p=0.11). Among 2,206 encounters in the Post-AI \nperiod, edit time was longer when the AI tool was used (9.20 [IQR 4.9-16.3] vs \n4.93 [IQR 2.5-9.7] minutes, p<0.001). In multivariable analysis, AI tool was \nassociated with a 32% increase in edit time (95% CI 24%-41%, p < 0.001; adjusted \nR² = 0.34). Faculty review found AI-generated summaries had higher quality and \nlower harm scores but were less concise. 31 (86.1%) clinicians agreed the tool \nimproved efficiency.\nCONCLUSIONS: Discharge summary edit time did not differ between study periods \nbut was longer when the AI tool was used. However, there was increased \nperception of efficiency. The tool improved summary quality and reduced \npotential harm but may increase documentation time.\n\nThieme. All rights reserved.\n\nDOI: 10.1055/a-2962-8004\nPMID: 42759594\n\nConflict of interest statement: The authors declare that they have no conflict \nof interest.", "extra": {"pmid": "42759594"}}
{"source": "pubmed", "title": "23. J Neurosci Methods. 2026 Sep 18:110909. doi: 10.1016/j.jneumeth.2026.110909. ", "url": "https://pubmed.ncbi.nlm.nih.gov/42759576/", "summary": "23. J Neurosci Methods. 2026 Sep 18:110909. doi: 10.1016/j.jneumeth.2026.110909. \nOnline ahead of print.\n\nTopoAdapter: a plug-and-play multi-hop topology adapter for MI-EEG decoding.\n\nCai Y(1), Guo Q(2), Lin X(3).\n\nAuthor information:\n(1)Academy of Artificial Intelligence, Beijing Institute of Petro", "content": "23. J Neurosci Methods. 2026 Sep 18:110909. doi: 10.1016/j.jneumeth.2026.110909. \nOnline ahead of print.\n\nTopoAdapter: a plug-and-play multi-hop topology adapter for MI-EEG decoding.\n\nCai Y(1), Guo Q(2), Lin X(3).\n\nAuthor information:\n(1)Academy of Artificial Intelligence, Beijing Institute of Petrochemical \nTechnology, Beijing 102617, China. Electronic address: 15260903118@163.com.\n(2)Academy of Artificial Intelligence, Beijing Institute of Petrochemical \nTechnology, Beijing 102617, China. Electronic address: guoqj@iccas.ac.cn.\n(3)Academy of Artificial Intelligence, Beijing Institute of Petrochemical \nTechnology, Beijing 102617, China. Electronic address: linxiaozhu@bipt.edu.cn.\n\nBACKGROUND: Motor imagery electroencephalography (MI-EEG) decoding is limited by \nlow signal-to-noise ratio, non-stationarity, inter-subject variability, and \nsmall calibration sets. Lightweight decoders are attractive for online BCI but \noften learn channel relations only from limited training data.\nNEW METHOD: We introduce TopoAdapter, a plug-and-play input module that injects \na fixed electrode-layout prior into existing EEG backbones. It builds a physical \nelectrode graph from the montage, computes cumulative multi-hop \nchannel-to-neighborhood contrasts, and adds a learnable low-amplitude residual \nwhile preserving input shape.\nRESULTS: Under an aligned 500-epoch, five-seed ATCNet protocol, mean accuracy \nchanges were +1.17 and +0.13 percentage points on the BCI Competition IV-2a and \nZhou2016 motor-imagery datasets, respectively, and +0.36 points on the \nHigh-Gamma executed-movement dataset. All three dataset-level means were \npositive; the BCI IV-2a and High-Gamma bootstrap intervals excluded zero, \nalthough no paired test remained significant after three-dataset Holm \ncorrection. In a separate BCI IV-2a compatibility study, all seven selected \nbackbones improved on average and three retained Holm-adjusted Wilcoxon \nevidence.\nCOMPARISON WITH EXISTING METHODS: Unlike graph neural decoders that redesign the \nbackbone, TopoAdapter keeps downstream components unchanged. In a matched \ncomparison with the open-source Adaptive Channel Mixing Layer (ACML), \nTopoAdapter attained 60.52% versus 60.42% accuracy with 70 versus 506 added \nparameters; the direct paired difference was unresolved.\nCONCLUSIONS: TopoAdapter provides an explicit, ultra-lightweight spatial prior \nfor motor EEG decoding. Positive mean changes on two motor-imagery datasets, one \nexecuted-movement dataset, and seven heterogeneous BCI IV-2a backbones support \nportability and a favorable cost-benefit profile, while effect magnitude remains \ndataset- and subject-dependent.\n\nCopyright © 2026 Elsevier B.V. All rights reserved.\n\nDOI: 10.1016/j.jneumeth.2026.110909\nPMID: 42759576\n\nConflict of interest statement: Declaration of Competing Interest The authors \ndeclare that they have no known competing financial interests or personal \nrelationships that could have appeared to influence the work reported in this \npaper. Declaration of competing interest The authors declare no competing \ninterests.", "extra": {"pmid": "42759576"}}
{"source": "pubmed", "title": "24. Ophthalmology. 2026 Sep 18:S0161-6420(26)00687-1. doi: ", "url": "https://pubmed.ncbi.nlm.nih.gov/42759575/", "summary": "24. Ophthalmology. 2026 Sep 18:S0161-6420(26)00687-1. doi: \n10.1016/j.ophtha.2026.09.015. Online ahead of print.\n\nCentralized Multidisciplinary Retinoblastoma Care Closes the Outcome Gap: \nSurvival and Globe Salvage for 1000 Eyes from 601 Patients.\n\nYousef YA(1), Halalsheh H(2), Mohammad M(3), Khzou", "content": "24. Ophthalmology. 2026 Sep 18:S0161-6420(26)00687-1. doi: \n10.1016/j.ophtha.2026.09.015. Online ahead of print.\n\nCentralized Multidisciplinary Retinoblastoma Care Closes the Outcome Gap: \nSurvival and Globe Salvage for 1000 Eyes from 601 Patients.\n\nYousef YA(1), Halalsheh H(2), Mohammad M(3), Khzouz J(4), Laban DA(5), Al Jabari \nR(3), Haboob H(5), Shawagfeh M(6), Jaradat I(7), Sultan I(8), Al-Hussaini M(9), \nMehyar M(3), Al-Nawaiseh I(3).\n\nAuthor information:\n(1)Departments of Surgery/Ophthalmology. Electronic address: yyousef@khcc.jo.\n(2)Departments of Pediatric Oncology.\n(3)Departments of Surgery/Ophthalmology.\n(4)Departments of Pathology and Laboratory Medicine.\n(5)Departments of Radiology.\n(6)Departments of Anesthesia.\n(7)Departments of Radiation Oncology.\n(8)Departments of Pediatric Oncology; Departments of Artificial Intelligence \nOffice.\n(9)Departments of Pathology and Laboratory Medicine; Cell Therapy and Applied \nGenomics, King Hussein Cancer Centre, Amman, Jordan.\n\nPURPOSE: Retinoblastoma (RB) outcomes in low- and middle-income countries remain \ninferior to those achieved in high-income countries, largely due to delayed \ndiagnosis, limited access to specialized care, and fragmented treatment systems. \nThe aim of this study is to evaluate the long-term impact of a two-decade \nstrategy based on centralized multidisciplinary care, structured capacity \nbuilding, and a telemedicine-supported international twinning partnership in \nreducing disparities in retinoblastoma (RB) outcomes in a resource-limited \nsetting.\nDESIGN: Retrospective, single-center cohort study.\nSUBJECTS: 601 children with RB (1000 affected eyes) treated at King Hussein \nCancer Center (KHCC) between March 2003 and December 2025.\nMETHODS: Demographic characteristics, tumor classification, treatment \nmodalities, globe salvage, metastasis, secondary malignancies, and survival \noutcomes were reviewed. Overall survival was estimated using the Kaplan-Meier \nmethod.\nMAIN OUTCOME MEASURES: Globe salvage rate, rates of metastasis and secondary \nmalignancy, and 5-year overall survival.\nRESULTS: Of 601 children, 314 (52%) were male, 399 (66%) had bilateral disease, \nand 346 (58%) were non-Jordanian, referred from 15 countries. Of 1000 affected \neyes, 120 (12%) were enucleated before referral; among the remaining eyes, 7 \n(0.7%) had extraocular disease, and by International Classification, 52 (5.2%) \nwere Group A, 134 (13%) Group B, 204 (20%) Group C, 391 (39%) Group D, and 92 \n(9.2%) Group E. Of 880 eyes treated at KHCC, primary enucleation was performed \nin 157 (18%), and conservative eye-salvage therapy was attempted in 723 (82%). \nAmong the 723 eyes that underwent eye-salvaging treatment, systemic chemotherapy \nwas administered in 713 (99%), transpupillary thermo therapy (TTT) in all 723 \n(100%), cryotherapy in 248 (34%), intra-arterial chemotherapy in 34 (4.7%; 11 \nprimary and 23 secondary), intravitreal chemotherapy in 43 (5.9%), subtenon \nchemotherapy in 24 (3.3%), and iodine-125 plaque brachytherapy in 23 (3.2%). Ten \nGroup A eyes in 10 patients were treated with TTT and/or cryotherapy alone \nwithout systemic chemotherapy. Overall eye salvage rate was 61% (538/880); among \nconservatively treated eyes, salvage rate was 74.4% (538/723), ranging from 98% \n(Group A) to 17% (Group E). At median follow-up of 120 months, metastasis \noccurred in 26 patients (4.3%) and secondary malignancy in 7 (1.2%). Estimated \n5-year overall survival was 96% (95% CI, 94%-98%).\nCONCLUSIONS: A centralized multidisciplinary RB program integrating \ntelemedicine, international collaboration, and advanced globe-salvage therapies \nachieved survival and eye-preservation outcomes comparable to high-income \nsettings despite resource constraints. Over two decades, KHCC evolved into a \nmajor regional referral center, offering a scalable model for reducing global \ndisparities in RB care.\nFUNDING: None.\n\nCopyright © 2026 American Academy of Ophthalmology, Inc. Published by Elsevier \nInc. All rights reserved.\n\nDOI: 10.1016/j.ophtha.2026.09.015\nPMID: 42759575", "extra": {"pmid": "42759575"}}
{"source": "pubmed", "title": "25. Physiol Meas. 2026 Sep 18. doi: 10.1088/1361-6579/aea9ee. Online ahead of", "url": "https://pubmed.ncbi.nlm.nih.gov/42759567/", "summary": "25. Physiol Meas. 2026 Sep 18. doi: 10.1088/1361-6579/aea9ee. Online ahead of\nprint.\n\nStage-specific EEG spectral signatures of primary insomnia: regional and \nsex-related differences across NREM and REM sleep.\n\nPremakumara N(1), Yang CY(2), Sudantha BH(3), Chen KW(4).\n\nAuthor information:\n(1)Nation", "content": "25. Physiol Meas. 2026 Sep 18. doi: 10.1088/1361-6579/aea9ee. Online ahead of\nprint.\n\nStage-specific EEG spectral signatures of primary insomnia: regional and \nsex-related differences across NREM and REM sleep.\n\nPremakumara N(1), Yang CY(2), Sudantha BH(3), Chen KW(4).\n\nAuthor information:\n(1)National Taipei University, No. 151, University Rd,, Sanxia Dist,, New Taipei \nCity, Taipei City, 237303, Taiwan.\n(2)Department of Electrical Engineering, National Taipei University, No.151, \nUniversity Rd, Sanxia Dist, New Taipei City, Taipei City, 237303, Taiwan.\n(3)University of Moratuwa, Katubedda, Moratuwa, WP, 10400, Sri Lanka.\n(4)Department of Internal Medicine, Cheng Hsin General Hospital, No. 45, \nZhenxing Street, Ronghua Village, Beitou District, Taipei City, 112, Taiwan.\n\nSleep architecture is characterized by dynamic transitions between non-rapid eye \nmovement (NREM) and rapid eye movement (REM) stages, each defined by distinct \noscillatory signatures. Although quantitative EEG has been widely used to study \nprimary insomnia (PI), reported findings remain inconsistent, partly due to \nlimited consideration of cortical regional differences, sleep-stage specificity, \nand sex-related variability. This study integrates spectral power analysis with \nexplainable machine learning to provide a comprehensive characterization of \nneural dynamics in PI. EEG recordings from 100 adults (50 healthy sleepers and \n50 PI patients) were analyzed across frontal (F3, F4), central (C3, C4), and \noccipital (O1, O2) regions. Stage-specific classifiers were developed for NREM \nand REM using LASSO, random forest, CatBoost and XGBoost models, and Shapley \nAdditive Explanations (SHAP) were employed to quantify the contribution of each \nspectral and regional feature to model predictions. Models trained on NREM \nfeatures demonstrated superior discrimination compared with those trained on REM \nfeatures, suggesting that slow-wave neural activity captures insomnia-related \nalterations more reliably than REM dynamics. SHAP analysis identified reduced \ndelta power and elevated beta activity most prominently in central regions as \nthe dominant predictors of PI during NREM sleep. In contrast, REM-based models \npointed to prefrontal beta1 and occipital alpha as secondary discriminative \nfeatures. Sex-stratified analyses further revealed systematically higher \nspectral power in females and distinct SHAP feature-importance patterns between \nsexes. Collectively, this work establishes an interpretable, stage-specific \nframework for identifying EEG biomarkers of insomnia and demonstrates the value \nof explainable artificial intelligence in advancing the neurophysiological \nunderstanding of sleep disturbance.\n\n© 2026 Institute of Physics and Engineering in Medicine. All rights, including \nfor text and data mining, AI training, and similar technologies, are reserved.\n\nDOI: 10.1088/1361-6579/aea9ee\nPMID: 42759567", "extra": {"pmid": "42759567"}}
{"source": "pubmed", "title": "26. Physiol Meas. 2026 Sep 18. doi: 10.1088/1361-6579/aea9ea. Online ahead of", "url": "https://pubmed.ncbi.nlm.nih.gov/42759561/", "summary": "26. Physiol Meas. 2026 Sep 18. doi: 10.1088/1361-6579/aea9ea. Online ahead of\nprint.\n\nImproving sleep monitoring in athletes: a comparative evaluation of four \nautomated sleep scoring algorithms on 204 polysomnography recordings.\n\nCrosbie F(1), Hanif U(2), Aloulou A(2), Chauvineau M(3), Chennaoui M(", "content": "26. Physiol Meas. 2026 Sep 18. doi: 10.1088/1361-6579/aea9ea. Online ahead of\nprint.\n\nImproving sleep monitoring in athletes: a comparative evaluation of four \nautomated sleep scoring algorithms on 204 polysomnography recordings.\n\nCrosbie F(1), Hanif U(2), Aloulou A(2), Chauvineau M(3), Chennaoui M(4), Nedelec \nM(5), Léger D(2).\n\nAuthor information:\n(1)INSEP, 11 Avenue du Tremblay, Paris, 75012, France.\n(2)VIFASOM, Université Paris Cité, 1 Parvis Notre Dame, Paris, Île-de-France, \n75004, France.\n(3)VIFASOM, Université Paris Cité, 1 Parvis Notre Dame, Paris, 75006, France.\n(4)IRBA (Institut de Recherche biomédical des armées), Place General Valerie \nAndre, Brétigny-sur-Orge, 91220, France.\n(5)INSEP, 11 Avenue du Tremblay, Paris, Île-de-France, 75012, France.\n\nObjective&#xD;With advancements in AI, automated sleep scoring offers the \npotential for fast, precise, and cost-effective analysis. This study aimed to \nevaluate the accuracy, robustness, and clinical agreement of four algorithms \ncompared to manual PSG in athletes to determine their suitability for real-life \nuse.&#xD;Approach&#xD;Data from 75 athletes (204 polysomnography recordings) \nwere retrospectively collected from studies conducted in our laboratory. Manual \nscoring was compared to four automated sleep staging algorithms: Luna, U-Sleep, \nYASA, and GSSC. Performance was evaluated based on accuracy (per stage, macro \nand weighted F1), sensitivity (confusion matrices), robustness (signal quality: \nR 2 ; individual differences: beta coefficients) and clinical relevance \n(prediction of total sleep time, sleep onset latency and wake after sleep \nonset).&#xD;Main results&#xD;All algorithms demonstrated strong agreement with \nmanual scoring, with macro F1 scores of 0.76 (GSSC), 0.71 (U-Sleep), 0.70 \n(YASA), and 0.69 (Luna). GSSC was the most accurate and robust across sport, \nphysiological, behavioral, and anthropometric variables. It also reliably \npredicted total sleep time, sleep onset latency and wake after sleep onset, \nshowing minimal bias and narrow limits of agreement. GSSC and Luna were the most \nresilient to signal degradation (R² = 0.08), compared to YASA (R² = 0.16) and \nU-Sleep (R² = 0.29).&#xD;Significance&#xD;Automated sleep staging algorithms \ndemonstrated accuracy comparable to typical inter-rater agreement, supporting \ntheir broader use in athletes. GSSC emerged as the most accurate and robust \noption, particularly resilient to signal quality degradation, a common challenge \nin ecological sport settings. By reducing scoring time and removing the need for \nexpert raters, these algorithms could make polysomnography more accessible for \nboth routine sleep monitoring and large-scale research, enabling more \nindividualized recovery strategies and deeper insights into how sleep \narchitecture influences athletic performance.\n\nCreative Commons Attribution license.\n\nDOI: 10.1088/1361-6579/aea9ea\nPMID: 42759561", "extra": {"pmid": "42759561"}}
{"source": "pubmed", "title": "27. Neuron. 2026 Sep 18:S0896-6273(26)00643-4. doi: 10.1016/j.neuron.2026.08.013.", "url": "https://pubmed.ncbi.nlm.nih.gov/42759509/", "summary": "27. Neuron. 2026 Sep 18:S0896-6273(26)00643-4. doi: 10.1016/j.neuron.2026.08.013.\n Online ahead of print.\n\nThe recipe for intelligence in natural and artificial systems.\n\nSummerfield C(1), Stachenfeld K(2).\n\nAuthor information:\n(1)Department of Experimental Psychology, University of Oxford, Oxford O", "content": "27. Neuron. 2026 Sep 18:S0896-6273(26)00643-4. doi: 10.1016/j.neuron.2026.08.013.\n Online ahead of print.\n\nThe recipe for intelligence in natural and artificial systems.\n\nSummerfield C(1), Stachenfeld K(2).\n\nAuthor information:\n(1)Department of Experimental Psychology, University of Oxford, Oxford OX13UD, \nUK. Electronic address: christopher.summerfield@psy.ox.ac.uk.\n(2)Center for Theoretical Neuroscience, Columbia University, New York, NY 10027, \nUSA; Google DeepMind, New York, NY 10011, USA.\n\nTen years ago, two landmark papers discussed open problems in AI research from \nthe standpoint of neuroscience and cognitive science. Today, many of these \nproblems have been solved, mostly without direct inspiration from the study of \nthe human brain. In this paper, we consider what is still shared between the \ntwin endeavors of understanding natural brains and building artificial ones. We \nargue that, as well as understanding the \"ingredients\" of intelligence (its \ncomponent parts), both fields are now coming to grasp the \"recipe\" for \nintelligence (how they fit together). Surprisingly, this recipe has many \nparallels in biological and artificial systems. Despite rapid progress in AI \nresearch, there are many computational problems that remain unsolved. We discuss \nhow these problems may be addressed in natural intelligence.\n\nCopyright © 2026 The Author(s). Published by Elsevier Inc. All rights reserved.\n\nDOI: 10.1016/j.neuron.2026.08.013\nPMID: 42759509\n\nConflict of interest statement: Declaration of interests K.S. is an employee and \nshareholder of Google DeepMind. C.S. was an employee and shareholder of Google \nDeepMind until 2023. C.S. now works for the UK AI Security Institute. These \norganizations had no role in funding this work.", "extra": {"pmid": "42759509"}}
{"source": "pubmed", "title": "28. Med. 2026 Sep 18:101296. doi: 10.1016/j.medj.2026.101296. Online ahead of", "url": "https://pubmed.ncbi.nlm.nih.gov/42759502/", "summary": "28. Med. 2026 Sep 18:101296. doi: 10.1016/j.medj.2026.101296. Online ahead of\nprint.\n\nDeep learning-enabled screening of chronic kidney disease from echocardiography.\n\nYuan V(1), Ieki H(2), Sandhu A(3), Nguyen LH(4), Cheng PP(2), Chang ST(2), \nAmbrosy AP(5), Kwan AC(6), Go AS(5), Cheng S(6), Ouyang ", "content": "28. Med. 2026 Sep 18:101296. doi: 10.1016/j.medj.2026.101296. Online ahead of\nprint.\n\nDeep learning-enabled screening of chronic kidney disease from echocardiography.\n\nYuan V(1), Ieki H(2), Sandhu A(3), Nguyen LH(4), Cheng PP(2), Chang ST(2), \nAmbrosy AP(5), Kwan AC(6), Go AS(5), Cheng S(6), Ouyang D(7).\n\nAuthor information:\n(1)Department of Cardiology, Smidt Heart Institute, Cedars-Sinai Medical Center, \nLos Angeles, CA, USA; David Geffen School of Medicine, University of California, \nLos Angeles, Los Angeles, CA, USA.\n(2)Division of Cardiology, Department of Medicine, Stanford University, Palo \nAlto, CA, USA.\n(3)David Geffen School of Medicine, University of California, Los Angeles, Los \nAngeles, CA, USA; Division of Cardiology, Department of Medicine, Stanford \nUniversity, Palo Alto, CA, USA.\n(4)Division of Gastroenterology, Massachusetts General Hospital, Boston, MA, \nUSA.\n(5)Division of Research, Kaiser Permanente Northern California, Pleasanton, CA, \nUSA.\n(6)Department of Cardiology, Smidt Heart Institute, Cedars-Sinai Medical Center, \nLos Angeles, CA, USA.\n(7)Department of Cardiology, Smidt Heart Institute, Cedars-Sinai Medical Center, \nLos Angeles, CA, USA; Division of Research, Kaiser Permanente Northern \nCalifornia, Pleasanton, CA, USA. Electronic address: david.ouyang@kp.org.\n\nBACKGROUND: Chronic kidney disease (CKD) affects nearly 850 million individuals \nglobally. The prevalence of undiagnosed CKD is over 60%.\nMETHODS: Taking advantage of the relationship between CKD and cardiovascular \ndisease, we developed a deep learning (DL) model to detect CKD from parasternal \nlong-axis (PLAX) videos, using 325,377 PLAX videos from 62,818 patients at \nCedars-Sinai Medical Center (CSMC). We externally validated our model in two \nindependent cohorts of 2,762 patients at Stanford Healthcare (SHC) and 41,611 \npatients at Kaiser-Permanente Northern California (KPNC).\nFINDINGS: In a held-out test cohort at CSMC, our model detected any stage of CKD \nwith an area under the curve (AUC) of 0.756 (95% confidence interval: \n0.749-0.763), with consistently strong performance in the KPNC (AUC: 0.718 \n[0.714-0.723]) and SHC (AUC: 0.719 [0.704-0.735]) cohorts. Our model performed \nwell across subgroups with and without metabolic comorbidities, suggesting that \nit learned imaging features specific to CKD.\nCONCLUSIONS: Our DL echo model detected CKD with robust performance at two \nexternal clinical sites, thus offering an avenue for noninvasive screening and \nimproved detection rates.\nFUNDING: This work was supported by the Sarnoff Cardiovascular Research \nFoundation, the American Heart Association (25POST1357984 and 25AHAI1487693), \nand the National Institutes of Health (R00HL157421, R01HL173487, and \nR01HL173526).\n\nCopyright © 2026 The Author(s). Published by Elsevier Inc. All rights reserved.\n\nDOI: 10.1016/j.medj.2026.101296\nPMID: 42759502\n\nConflict of interest statement: Declaration of interests S.T.C. reports research \nsupport from Siemens Healthcare. A.P.A. reports research support through grants \nfrom the National Institutes of Health (R01HL173866 and R01AG091005), the Food \nand Drug Administration (U01FD008704), the American Heart Association, the \nPermanente Medical Group, Kaiser Permanente Northern California Community \nBenefits Programs, Garfield Memorial Fund, Abbott Laboratories, Abiomed, Amarin \nPharma, Inc., Bayer Pharma AG, Boehringer Ingelheim, Cordio Medical, Edwards \nLifesciences LLC, Esperion Therapeutics, Inc., Merck, and Novartis; and \nconsulting for Bristol Meyer Squibb, Bayer Pharma AG, Corstasis, Merck, Novo \nNordisk, Pharmacosmos A/S, Pfizer, scPharma, SQ Innovation, Reprieve \nCardiovascular, and VisCardia. A.S.G. has received research support from the \nNational Heart, Lung and Blood Institute and the National Institute of Diabetes, \nDigestive and Kidney Diseases. D.O. reports consulting, equity, or honoraria for \nlectures from AstraZeneca Alexion, EchoIQ, Ultromics, Pfizer, InVision, the \nKorean Society of Echocardiography, and the Japanese Society of \nEchocardiography.", "extra": {"pmid": "42759502"}}
{"source": "pubmed", "title": "29. Nurse Educ Today. 2026 Sep 15;168:107421. doi: 10.1016/j.nedt.2026.107421. ", "url": "https://pubmed.ncbi.nlm.nih.gov/42759434/", "summary": "29. Nurse Educ Today. 2026 Sep 15;168:107421. doi: 10.1016/j.nedt.2026.107421. \nOnline ahead of print.\n\nFunctional connectivity theory-based neurophysiological profiles associated with \nemotional competence and clinical judgment in nursing students: A cluster \nanalysis.\n\nCho YK(1).\n\nAuthor informati", "content": "29. Nurse Educ Today. 2026 Sep 15;168:107421. doi: 10.1016/j.nedt.2026.107421. \nOnline ahead of print.\n\nFunctional connectivity theory-based neurophysiological profiles associated with \nemotional competence and clinical judgment in nursing students: A cluster \nanalysis.\n\nCho YK(1).\n\nAuthor information:\n(1)Department of Nursing, Yeungnam University College, Daegu, Republic of Korea. \nElectronic address: lieben59@naver.com.\n\nBACKGROUND: Artificial intelligence (AI) is transforming nursing education by \nsupporting information processing and clinical reasoning; however, human-centred \ncompetencies such as emotional competence and clinical judgment remain essential \nfor psychiatric nursing practice. Guided by Functional Connectivity Theory, this \nstudy investigated whether neurophysiological profiles based on hemispheric \nbalance and interhemispheric connectivity were associated with emotional \ncompetence and clinical judgment among nursing students.\nOBJECTIVES: To identify neurophysiological profiles using a person-centred \ncluster analytic approach and to examine differences in emotional competence and \nclinical judgment across the identified profiles.\nDESIGN: A secondary analysis using a cross-sectional, person-centred cluster \nanalytic design.\nPARTICIPANTS/SETTINGS: A total of 212 fourth-year nursing students who completed \npsychiatric nursing practicum at a university in the Republic of Korea.\nMETHODS: Hierarchical cluster analysis (Ward's method) was performed using \nhemispheric balance and interhemispheric connectivity measured with a portable \ntwo-channel EEG device. Independent-samples t-tests compared emotional \ncompetence and clinical judgment between the identified neurophysiological \nprofiles.\nRESULTS: Two distinct neurophysiological profiles were identified: an integrated \nprofile (n = 124) and a less-integrated profile (n = 88). Students in the \nintegrated profile demonstrated significantly higher emotional utilization \n(p = .033), emotional self-efficacy (p = .009), and overall clinical judgment \n(p = .028), including the domains of noticing, interpreting, responding, and \nreflecting. No significant difference was observed in overall emotional \ncompetence.\nCONCLUSIONS: Guided by Functional Connectivity Theory, this study demonstrates \nthat functional brain organization is associated with higher-order human-centred \ncompetencies rather than with emotional competence as a global construct. \nNeurophysiological profiling may provide a neuroscience-informed framework for \nunderstanding individual differences in nursing students' learning and may \nsupport the development of personalized educational strategies in the AI era.\n\nCopyright © 2026. Published by Elsevier Ltd.\n\nDOI: 10.1016/j.nedt.2026.107421\nPMID: 42759434\n\nConflict of interest statement: Declaration of competing interest The author \ndeclares no competing interests.", "extra": {"pmid": "42759434"}}
{"source": "pubmed", "title": "30. Mol Aspects Med. 2026 Sep 18;112:101504. doi: 10.1016/j.mam.2026.101504.", "url": "https://pubmed.ncbi.nlm.nih.gov/42759410/", "summary": "30. Mol Aspects Med. 2026 Sep 18;112:101504. doi: 10.1016/j.mam.2026.101504.\nOnline  ahead of print.\n\nRole of foundation models in data-driven tissue diagnostics.\n\nBilal M(1), Aadam(2), Raza M(3), Alsuhaibani A(4), Altherwy YN(4), Alabduljabbar \nA(4), Almarshad F(4), Golding P(5), Rajpoot N(6).\n\nAut", "content": "30. Mol Aspects Med. 2026 Sep 18;112:101504. doi: 10.1016/j.mam.2026.101504.\nOnline  ahead of print.\n\nRole of foundation models in data-driven tissue diagnostics.\n\nBilal M(1), Aadam(2), Raza M(3), Alsuhaibani A(4), Altherwy YN(4), Alabduljabbar \nA(4), Almarshad F(4), Golding P(5), Rajpoot N(6).\n\nAuthor information:\n(1)Information Systems Department, College of Computer Engineering and Sciences, \nPrince Sattam bin Abdulaziz University, Al-Kharj, Saudi Arabia. Electronic \naddress: m.farooq@psau.edu.sa.\n(2)Luddy School of Informatics, Computing, and Engineering, Indiana University \nIndianapolis, United States of America.\n(3)Tissue Image Analytics Centre, Department of Computer Science, University of \nWarwick, Coventry, United Kingdom.\n(4)Information Systems Department, College of Computer Engineering and Sciences, \nPrince Sattam bin Abdulaziz University, Al-Kharj, Saudi Arabia.\n(5)Frontier AI, San Francisco Bay Area, United States of America.\n(6)Tissue Image Analytics Centre, Department of Computer Science, University of \nWarwick, Coventry, United Kingdom; Histofy Ltd, The Venture Centre, University \nof Warwick Science Park, Coventry, United Kingdom. Electronic address: \nN.M.Rajpoot@warwick.ac.uk.\n\nArtificial intelligence (AI)-based tissue diagnostics is entering a new phase \ndriven by pathology foundation models: large-scale encoders and multimodal \nsystems pretrained with self-supervised and vision-language objectives. Yet the \nevidence base has expanded faster than the methods used to evaluate and \ninterpret claimed capabilities. The central clinical question is no longer \nwhether these models can achieve strong retrospective performance, but what new \ncapabilities they provide, under what conditions those capabilities are \ndemonstrated, and whether they can translate into reliable diagnostics in \nclinical settings. This review makes three contributions. First, we define \nhistopathology-centric foundation models and distill the technical factors that \nshape their behavior: pretraining data regime, learning objective, and \ndownstream adaptation to clinical endpoints. Second, we introduce a practical \ncapability framing that distinguishes general-purpose capabilities (coverage \nacross tissues, scales, stains, scanners, and institutions) from functional \nbreadth (task primitives, adaptability, and analysis level), while accounting \nfor modality scope spanning vision, language, and genomics. Third, we synthesize \nreported results as an evidence map rather than a leaderboard, clarifying where \ncapability is supported by reported evidence, where reproducibility is \nconstrained by access or reporting limits, and where further validation is \nneeded. We then analyze current benchmarking practice and identify common \nconfounders, including heterogeneous adaptation protocols and \npretraining-evaluation overlap, and propose deployment-aware recommendations \nbuilt around broad-and-deep benchmarks, standardized adaptation \"budgets,\" \noverlap auditing, and domain-agnostic evaluation. Finally, we review emerging \nclinical-utility evidence and argue that foundation models are most compelling \nwhen tied to workflow-defined endpoints, calibrated operating points, and \nmeasurable operational benefit. We conclude with actionable recommendations for \nconverting capability demonstrations into clinically reliable data-driven \ndiagnostic systems.\n\nCopyright © 2026. Published by Elsevier Ltd.\n\nDOI: 10.1016/j.mam.2026.101504\nPMID: 42759410\n\nConflict of interest statement: Declaration of competing interest Nasir Rajpoot \ndeclares his role as a co-founder and shareholder of Histofy Ltd, a university \nspinout focusing on computational pathology software. All other authors declare \nno competing interests.", "extra": {"pmid": "42759410"}}
{"source": "stackexchange", "title": "What is the difference between supervised learning and unsupervised learning?", "url": "https://stackoverflow.com/questions/1832076/what-is-the-difference-between-supervised-learning-and-unsupervised-learning", "summary": "Score: 293, Answers: 27", "content": "What is the difference between supervised learning and unsupervised learning?", "extra": {"score": 293, "answer_count": 27}}
{"source": "stackexchange", "title": "Artificial Intelligence Compiler", "url": "https://stackoverflow.com/questions/3455062/artificial-intelligence-compiler", "summary": "Score: 20, Answers: 4", "content": "Artificial Intelligence Compiler", "extra": {"score": 20, "answer_count": 4}}
{"source": "stackexchange", "title": "How to pick a language for Artificial Intelligence programming?", "url": "https://stackoverflow.com/questions/1877505/how-to-pick-a-language-for-artificial-intelligence-programming", "summary": "Score: 37, Answers: 12", "content": "How to pick a language for Artificial Intelligence programming?", "extra": {"score": 37, "answer_count": 12}}
{"source": "stackexchange", "title": "Learning Lisp and artificial intelligence", "url": "https://stackoverflow.com/questions/7753858/learning-lisp-and-artificial-intelligence", "summary": "Score: 4, Answers: 4", "content": "Learning Lisp and artificial intelligence", "extra": {"score": 4, "answer_count": 4}}
{"source": "stackexchange", "title": "Drools Artificial Intelligence", "url": "https://stackoverflow.com/questions/6490916/drools-artificial-intelligence", "summary": "Score: -4, Answers: 1", "content": "Drools Artificial Intelligence", "extra": {"score": -4, "answer_count": 1}}
{"source": "stackexchange", "title": "Backward chaining in artificial intelligence", "url": "https://stackoverflow.com/questions/40027883/backward-chaining-in-artificial-intelligence", "summary": "Score: 1, Answers: 1", "content": "Backward chaining in artificial intelligence", "extra": {"score": 1, "answer_count": 1}}
{"source": "stackexchange", "title": "What&#39;s artificial intelligence?", "url": "https://stackoverflow.com/questions/4469040/whats-artificial-intelligence", "summary": "Score: 8, Answers: 8", "content": "What&#39;s artificial intelligence?", "extra": {"score": 8, "answer_count": 8}}
{"source": "stackexchange", "title": "how to start artificial-intelligence programming?", "url": "https://stackoverflow.com/questions/8208302/how-to-start-artificial-intelligence-programming", "summary": "Score: 2, Answers: 1", "content": "how to start artificial-intelligence programming?", "extra": {"score": 2, "answer_count": 1}}
{"source": "stackexchange", "title": "Artificial Intelligence in games", "url": "https://stackoverflow.com/questions/14423212/artificial-intelligence-in-games", "summary": "Score: -2, Answers: 1", "content": "Artificial Intelligence in games", "extra": {"score": -2, "answer_count": 1}}
{"source": "stackexchange", "title": "Artificial Intelligence for Decision Making", "url": "https://stackoverflow.com/questions/5320506/artificial-intelligence-for-decision-making", "summary": "Score: -1, Answers: 5", "content": "Artificial Intelligence for Decision Making", "extra": {"score": -1, "answer_count": 5}}
{"source": "stackexchange", "title": "Artificial Intelligence and Chat Filters", "url": "https://stackoverflow.com/questions/3992513/artificial-intelligence-and-chat-filters", "summary": "Score: 3, Answers: 2", "content": "Artificial Intelligence and Chat Filters", "extra": {"score": 3, "answer_count": 2}}
{"source": "stackexchange", "title": "Artificial intelligence libraries", "url": "https://stackoverflow.com/questions/16778463/artificial-intelligence-libraries", "summary": "Score: 5, Answers: 2", "content": "Artificial intelligence libraries", "extra": {"score": 5, "answer_count": 2}}
{"source": "stackexchange", "title": "What&#39;s the utility theory in artificial intelligence?", "url": "https://stackoverflow.com/questions/5332380/whats-the-utility-theory-in-artificial-intelligence", "summary": "Score: 12, Answers: 1", "content": "What&#39;s the utility theory in artificial intelligence?", "extra": {"score": 12, "answer_count": 1}}
{"source": "stackexchange", "title": "Artificial intelligence functionality in three.js", "url": "https://stackoverflow.com/questions/30911988/artificial-intelligence-functionality-in-three-js", "summary": "Score: 0, Answers: 4", "content": "Artificial intelligence functionality in three.js", "extra": {"score": 0, "answer_count": 4}}
{"source": "stackexchange", "title": "How to write artificial intelligence in .net ( C # )", "url": "https://stackoverflow.com/questions/4638213/how-to-write-artificial-intelligence-in-net-c", "summary": "Score: 20, Answers: 5", "content": "How to write artificial intelligence in .net ( C # )", "extra": {"score": 20, "answer_count": 5}}
{"source": "stackexchange", "title": "What kind of artificial intelligence jobs are out there?", "url": "https://stackoverflow.com/questions/715181/what-kind-of-artificial-intelligence-jobs-are-out-there", "summary": "Score: 20, Answers: 5", "content": "What kind of artificial intelligence jobs are out there?", "extra": {"score": 20, "answer_count": 5}}
{"source": "stackexchange", "title": "Artificial Intelligence - Machine Learning", "url": "https://stackoverflow.com/questions/10002491/artificial-intelligence-machine-learning", "summary": "Score: 5, Answers: 3", "content": "Artificial Intelligence - Machine Learning", "extra": {"score": 5, "answer_count": 3}}
{"source": "stackexchange", "title": "Logic aplications in artificial intelligence", "url": "https://stackoverflow.com/questions/61711686/logic-aplications-in-artificial-intelligence", "summary": "Score: 2, Answers: 2", "content": "Logic aplications in artificial intelligence", "extra": {"score": 2, "answer_count": 2}}
{"source": "stackexchange", "title": "Difference between Artificial Intelligence and Machine Learning?", "url": "https://stackoverflow.com/questions/53913030/difference-between-artificial-intelligence-and-machine-learning", "summary": "Score: -6, Answers: 4", "content": "Difference between Artificial Intelligence and Machine Learning?", "extra": {"score": -6, "answer_count": 4}}
{"source": "stackexchange", "title": "artificial intelligence and expert systems", "url": "https://stackoverflow.com/questions/10481316/artificial-intelligence-and-expert-systems", "summary": "Score: 2, Answers: 3", "content": "artificial intelligence and expert systems", "extra": {"score": 2, "answer_count": 3}}
{"source": "stackexchange", "title": "Is it possible to implement artificial intelligence concept in jquery?", "url": "https://stackoverflow.com/questions/4386160/is-it-possible-to-implement-artificial-intelligence-concept-in-jquery", "summary": "Score: 1, Answers: 4", "content": "Is it possible to implement artificial intelligence concept in jquery?", "extra": {"score": 1, "answer_count": 4}}
{"source": "stackexchange", "title": "Artificial Intelligence in Tic-Tac-Toe using C#", "url": "https://stackoverflow.com/questions/5366858/artificial-intelligence-in-tic-tac-toe-using-c", "summary": "Score: 13, Answers: 6", "content": "Artificial Intelligence in Tic-Tac-Toe using C#", "extra": {"score": 13, "answer_count": 6}}
{"source": "stackexchange", "title": "Making my artificial intelligence project known? AI communities?", "url": "https://stackoverflow.com/questions/9592757/making-my-artificial-intelligence-project-known-ai-communities", "summary": "Score: 3, Answers: 2", "content": "Making my artificial intelligence project known? AI communities?", "extra": {"score": 3, "answer_count": 2}}
{"source": "stackexchange", "title": "What is STRIPS in artificial intelligence?", "url": "https://stackoverflow.com/questions/41667755/what-is-strips-in-artificial-intelligence", "summary": "Score: -1, Answers: 1", "content": "What is STRIPS in artificial intelligence?", "extra": {"score": -1, "answer_count": 1}}
{"source": "stackexchange", "title": "Suggestions about artificial intelligence project", "url": "https://stackoverflow.com/questions/37373355/suggestions-about-artificial-intelligence-project", "summary": "Score: -1, Answers: 1", "content": "Suggestions about artificial intelligence project", "extra": {"score": -1, "answer_count": 1}}
{"source": "stackexchange", "title": "A Social Network with Artificial Intelligence", "url": "https://stackoverflow.com/questions/4272625/a-social-network-with-artificial-intelligence", "summary": "Score: 1, Answers: 2", "content": "A Social Network with Artificial Intelligence", "extra": {"score": 1, "answer_count": 2}}
{"source": "stackexchange", "title": "Python or Ruby for webbased Artificial Intelligence?", "url": "https://stackoverflow.com/questions/2495350/python-or-ruby-for-webbased-artificial-intelligence", "summary": "Score: 4, Answers: 5", "content": "Python or Ruby for webbased Artificial Intelligence?", "extra": {"score": 4, "answer_count": 5}}
{"source": "stackexchange", "title": "Artificial Intelligence Uniform-cost search", "url": "https://stackoverflow.com/questions/6299114/artificial-intelligence-uniform-cost-search", "summary": "Score: 0, Answers: 1", "content": "Artificial Intelligence Uniform-cost search", "extra": {"score": 0, "answer_count": 1}}
{"source": "stackexchange", "title": "Are the actions of an artificial intelligence absolutely static?", "url": "https://stackoverflow.com/questions/57340276/are-the-actions-of-an-artificial-intelligence-absolutely-static", "summary": "Score: 0, Answers: 0", "content": "Are the actions of an artificial intelligence absolutely static?", "extra": {"score": 0, "answer_count": 0}}
{"source": "stackexchange", "title": "Do i need to learn Artificial intelligence before learning Heuristics", "url": "https://stackoverflow.com/questions/6834876/do-i-need-to-learn-artificial-intelligence-before-learning-heuristics", "summary": "Score: 0, Answers: 1", "content": "Do i need to learn Artificial intelligence before learning Heuristics", "extra": {"score": 0, "answer_count": 1}}