Part 3: retrain on legally-grounded data, document the more-data-is-not-better regression and the macro-F1 fix
dd318c1 verified Download val_v2.csv from Devseis/eu-ai-act-classifier-experiment: direct link, hf CLI and curl.
- Browser
- Download file 19.1 kB
-
https://huggingface.co/Devseis/eu-ai-act-classifier-experiment/resolve/main/val_v2.csv
- Command line
-
hf download hf://Devseis/eu-ai-act-classifier-experiment/val_v2.csv
-
curl -L -o val_v2.csv https://huggingface.co/Devseis/eu-ai-act-classifier-experiment/resolve/main/val_v2.csv
19.1 kB
| text,label | |
| A startup is building a product that uses AI to an intelligent traffic control system prioritising road users and adjusting signals based on live conditions.,high_risk | |
| Proposed use case: an AI system designed to software assessing a suspect's personality traits and past behaviour to estimate reoffending likelihood for sentencing input.,high_risk | |
| Proposed use case: an AI system designed to a company-wide 'trust score' combining employees' personal social media activity to deny unrelated internal privileges disproportionately.,prohibited | |
| Proposed use case: an AI system designed to an app that identifies users living in extreme poverty and targets them with predatory high-interest loan offers.,prohibited | |
| A startup is building a product that uses AI to an AI system generating targeted political ads shown directly to voters to influence a referendum outcome.,high_risk | |
| "An AI tool ranking students by likelihood of future criminal behaviour using only personality test results, with no factual evidence.",prohibited | |
| We are developing an AI system that will an AI tool giving employees feedback on voluntary training-module quizzes for their own development.,minimal_risk | |
| Software inferring a website visitor's gender from typing patterns to tailor product recommendations.,minimal_risk | |
| Software linking real-time water-level sensor data to flood-gate control for road safety.,high_risk | |
| We're planning to use machine learning to a resume-parsing tool that automatically filters out applicants below an AI-determined suitability threshold.,high_risk | |
| "A startup is building a product that uses AI to software assessing the health risk posed by a traveller seeking entry, used by border authorities.",high_risk | |
| "We are developing an AI system that will a case-law search engine using standard keyword search for court staff, without generating legal reasoning.",minimal_risk | |
| We're planning to use machine learning to a vending machine estimating a customer's age from a camera to decide whether to sell an age-restricted product.,minimal_risk | |
| Proposed use case: an AI system designed to software predicting sewage system overflow that could contaminate the drinking water supply.,high_risk | |
| An adaptive placement tool recommending whether a student should move to advanced or vocational classes.,high_risk | |
| We are developing an AI system that will an app that swaps a user's face into a movie clip for entertainment.,limited_risk | |
| Proposed use case: an AI system designed to a system assessing which reading level or academic track a child should be placed in going forward.,high_risk | |
| Software analysing police emergency calls to prioritise which incidents get an immediate response.,high_risk | |
| We are developing an AI system that will a company-wide 'trust score' combining employees' personal social media activity to deny unrelated internal privileges disproportionately.,prohibited | |
| We're planning to use machine learning to a hybrid-office desk-booking tool matching employees to available desks and meeting rooms.,minimal_risk | |
| A startup is building a product that uses AI to an office monitoring tool that scans employees' facial expressions during meetings to detect who seems unhappy at work.,prohibited | |
| "A startup is building a product that uses AI to a scheme scoring citizens on unrelated behaviours (traffic tickets, late library books, online comments) to restrict access to public services.",prohibited | |
| "Our client wants to deploy an AI tool to an AI tool predicting which past donors are likely to contribute to a campaign, based on donation history.",minimal_risk | |
| "We're planning to use machine learning to software assessing the health risk posed by a traveller seeking entry, used by border authorities.",high_risk | |
| We're planning to use machine learning to an AI tool calculating a detained youth suspect's reoffending risk score to inform post-arrest decisions.,high_risk | |
| Proposed use case: an AI system designed to a vocational-training platform automatically deciding which applicants are admitted to a trade certification course.,high_risk | |
| Our client wants to deploy an AI tool to an adaptive placement tool recommending whether a student should move to advanced or vocational classes.,high_risk | |
| Proposed use case: an AI system designed to a hybrid-office desk-booking tool matching employees to available desks and meeting rooms.,minimal_risk | |
| Software that detects users in a vulnerable economic situation and steers them toward predatory financial products.,prohibited | |
| A startup is building a product that uses AI to software aggregating a tutor's ratings into a performance score that can automatically suspend their account.,high_risk | |
| Software assessing a suspect's personality traits and past behaviour to estimate reoffending likelihood for sentencing input.,high_risk | |
| We are developing an AI system that will a vocational-training platform automatically deciding which applicants are admitted to a trade certification course.,high_risk | |
| "Our client wants to deploy an AI tool to software automatically transcribing an immigration interview recording into text for the case file, without judging honesty.",minimal_risk | |
| Proposed use case: an AI system designed to an AI system optimising the targeting and delivery of political ads to individual voters.,high_risk | |
| We're planning to use machine learning to software placing targeted job advertisements to specific demographic groups based on automated candidate profiling.,high_risk | |
| "We are developing an AI system that will an AI tool ranking students by likelihood of future criminal behaviour using only personality test results, with no factual evidence.",prohibited | |
| A news aggregator publishing AI-written summaries of current events without any human review.,limited_risk | |
| Proposed use case: an AI system designed to an AI system generating targeted political ads shown directly to voters to influence a referendum outcome.,high_risk | |
| "Our client wants to deploy an AI tool to software analysing market trends to help an insurer design new product features, without scoring individual applicants.",minimal_risk | |
| Proposed use case: an AI system designed to software identifying religious minority group members to push them toward exploitative service agreements.,prohibited | |
| "We're planning to use machine learning to software analysing market trends to help an insurer design new product features, without scoring individual applicants.",minimal_risk | |
| Our client wants to deploy an AI tool to an office monitoring tool that scans employees' facial expressions during meetings to detect who seems unhappy at work.,prohibited | |
| An AI system optimising the targeting and delivery of political ads to individual voters.,high_risk | |
| "An AI tool predicting which past donors are likely to contribute to a campaign, based on donation history.",minimal_risk | |
| Our client wants to deploy an AI tool to a private platform that lowers a person's insurance eligibility based on unrelated data like their gym check-ins and grocery delivery habits.,prohibited | |
| We're planning to use machine learning to software analysing police emergency calls to prioritise which incidents get an immediate response.,high_risk | |
| We are developing an AI system that will a private platform that lowers a person's insurance eligibility based on unrelated data like their gym check-ins and grocery delivery habits.,prohibited | |
| "Proposed use case: an AI system designed to an AI tool ranking students by likelihood of future criminal behaviour using only personality test results, with no factual evidence.",prohibited | |
| "We are developing an AI system that will a scheme scoring citizens on unrelated behaviours (traffic tickets, late library books, online comments) to restrict access to public services.",prohibited | |
| Our client wants to deploy an AI tool to an AI tool calculating a detained youth suspect's reoffending risk score to inform post-arrest decisions.,high_risk | |
| We're planning to use machine learning to a TV network's tool counting how often its presenters smile on camera for editorial review.,minimal_risk | |
| We're planning to use machine learning to an autocomplete feature suggesting the next word while typing an internal memo.,minimal_risk | |
| Proposed use case: an AI system designed to an in-car system inferring driver emotional state from facial expression to adjust in-vehicle alerts.,high_risk | |
| "Our client wants to deploy an AI tool to an AI tool aggregating anonymised travel data to spot broad migration trends, with no individual identified.",minimal_risk | |
| Proposed use case: an AI system designed to software analysing police emergency calls to prioritise which incidents get an immediate response.,high_risk | |
| An intelligent traffic control system prioritising road users and adjusting signals based on live conditions.,high_risk | |
| We are developing an AI system that will a vending machine estimating a customer's age from a camera to decide whether to sell an age-restricted product.,minimal_risk | |
| We are developing an AI system that will an AI system optimising the targeting and delivery of political ads to individual voters.,high_risk | |
| "Our client wants to deploy an AI tool to a case-law search engine using standard keyword search for court staff, without generating legal reasoning.",minimal_risk | |
| A startup is building a product that uses AI to software reconstructing a 3D model of a crime scene from photos for investigators to review.,minimal_risk | |
| "A case-law search engine using standard keyword search for court staff, without generating legal reasoning.",minimal_risk | |
| Proposed use case: an AI system designed to software aggregating a tutor's ratings into a performance score that can automatically suspend their account.,high_risk | |
| Proposed use case: an AI system designed to software that detects users in a vulnerable economic situation and steers them toward predatory financial products.,prohibited | |
| A startup is building a product that uses AI to a platform that screens incoming CVs and ranks candidates before a human recruiter reviews any of them.,high_risk | |
| We are developing an AI system that will a news aggregator publishing AI-written summaries of current events without any human review.,limited_risk | |
| We are developing an AI system that will an intelligent traffic control system prioritising road users and adjusting signals based on live conditions.,high_risk | |
| We're planning to use machine learning to software linking real-time water-level sensor data to flood-gate control for road safety.,high_risk | |
| Our client wants to deploy an AI tool to an app that identifies users living in extreme poverty and targets them with predatory high-interest loan offers.,prohibited | |
| An AI tool flagging weapons visible in a live police bodycam feed during initial crime-scene response.,minimal_risk | |
| We are developing an AI system that will software that detects users in a vulnerable economic situation and steers them toward predatory financial products.,prohibited | |
| "We are developing an AI system that will software automatically transcribing an immigration interview recording into text for the case file, without judging honesty.",minimal_risk | |
| "Proposed use case: an AI system designed to an AI tool aggregating anonymised travel data to spot broad migration trends, with no individual identified.",minimal_risk | |
| Our client wants to deploy an AI tool to software assessing a suspect's personality traits and past behaviour to estimate reoffending likelihood for sentencing input.,high_risk | |
| "We are developing an AI system that will software assessing the health risk posed by a traveller seeking entry, used by border authorities.",high_risk | |
| Proposed use case: an AI system designed to a platform that screens incoming CVs and ranks candidates before a human recruiter reviews any of them.,high_risk | |
| We are developing an AI system that will software inferring a website visitor's gender from typing patterns to tailor product recommendations.,minimal_risk | |
| "We're planning to use machine learning to software automatically transcribing an immigration interview recording into text for the case file, without judging honesty.",minimal_risk | |
| A startup is building a product that uses AI to an autocomplete feature suggesting the next word while typing an internal memo.,minimal_risk | |
| We are developing an AI system that will a TV network's tool counting how often its presenters smile on camera for editorial review.,minimal_risk | |
| A platform that screens incoming CVs and ranks candidates before a human recruiter reviews any of them.,high_risk | |
| We are developing an AI system that will software placing targeted job advertisements to specific demographic groups based on automated candidate profiling.,high_risk | |
| Our client wants to deploy an AI tool to an AI tool creating realistic synthetic background music described as 'performed live' in marketing copy.,limited_risk | |
| We're planning to use machine learning to a vocational-training platform automatically deciding which applicants are admitted to a trade certification course.,high_risk | |
| We are developing an AI system that will software identifying religious minority group members to push them toward exploitative service agreements.,prohibited | |
| We're planning to use machine learning to an app that swaps a user's face into a movie clip for entertainment.,limited_risk | |
| An autocomplete feature suggesting the next word while typing an internal memo.,minimal_risk | |
| A startup is building a product that uses AI to an in-car system inferring driver emotional state from facial expression to adjust in-vehicle alerts.,high_risk | |
| A startup is building a product that uses AI to an AI tool calculating a detained youth suspect's reoffending risk score to inform post-arrest decisions.,high_risk | |
| Our client wants to deploy an AI tool to a news aggregator publishing AI-written summaries of current events without any human review.,limited_risk | |
| Our client wants to deploy an AI tool to a system assessing which reading level or academic track a child should be placed in going forward.,high_risk | |
| A startup is building a product that uses AI to an adaptive placement tool recommending whether a student should move to advanced or vocational classes.,high_risk | |
| Our client wants to deploy an AI tool to software inferring a website visitor's gender from typing patterns to tailor product recommendations.,minimal_risk | |
| Proposed use case: an AI system designed to an office monitoring tool that scans employees' facial expressions during meetings to detect who seems unhappy at work.,prohibited | |
| Proposed use case: an AI system designed to software reconstructing a 3D model of a crime scene from photos for investigators to review.,minimal_risk | |
| Software aggregating a tutor's ratings into a performance score that can automatically suspend their account.,high_risk | |
| Our client wants to deploy an AI tool to software predicting sewage system overflow that could contaminate the drinking water supply.,high_risk | |
| We're planning to use machine learning to an app that identifies users living in extreme poverty and targets them with predatory high-interest loan offers.,prohibited | |
| An AI system generating targeted political ads shown directly to voters to influence a referendum outcome.,high_risk | |
| Software placing targeted job advertisements to specific demographic groups based on automated candidate profiling.,high_risk | |
| We are developing an AI system that will software predicting sewage system overflow that could contaminate the drinking water supply.,high_risk | |
| We're planning to use machine learning to a system assessing which reading level or academic track a child should be placed in going forward.,high_risk | |
| Our client wants to deploy an AI tool to a resume-parsing tool that automatically filters out applicants below an AI-determined suitability threshold.,high_risk | |
| Our client wants to deploy an AI tool to software linking real-time water-level sensor data to flood-gate control for road safety.,high_risk | |
| We are developing an AI system that will an AI tool creating realistic synthetic background music described as 'performed live' in marketing copy.,limited_risk | |
| A startup is building a product that uses AI to an AI tool flagging weapons visible in a live police bodycam feed during initial crime-scene response.,minimal_risk | |
| "A startup is building a product that uses AI to software analysing market trends to help an insurer design new product features, without scoring individual applicants.",minimal_risk | |
| We're planning to use machine learning to an AI tool giving employees feedback on voluntary training-module quizzes for their own development.,minimal_risk | |
| Our client wants to deploy an AI tool to a company-wide 'trust score' combining employees' personal social media activity to deny unrelated internal privileges disproportionately.,prohibited | |
| Proposed use case: an AI system designed to a private platform that lowers a person's insurance eligibility based on unrelated data like their gym check-ins and grocery delivery habits.,prohibited | |
| We're planning to use machine learning to an in-car system inferring driver emotional state from facial expression to adjust in-vehicle alerts.,high_risk | |
| A resume-parsing tool that automatically filters out applicants below an AI-determined suitability threshold.,high_risk | |
| We are developing an AI system that will an AI tool flagging weapons visible in a live police bodycam feed during initial crime-scene response.,minimal_risk | |
| An app that swaps a user's face into a movie clip for entertainment.,limited_risk | |
| Our client wants to deploy an AI tool to a vending machine estimating a customer's age from a camera to decide whether to sell an age-restricted product.,minimal_risk | |
| A TV network's tool counting how often its presenters smile on camera for editorial review.,minimal_risk | |
| "A scheme scoring citizens on unrelated behaviours (traffic tickets, late library books, online comments) to restrict access to public services.",prohibited | |
| Proposed use case: an AI system designed to an AI tool creating realistic synthetic background music described as 'performed live' in marketing copy.,limited_risk | |
| "A startup is building a product that uses AI to an AI tool predicting which past donors are likely to contribute to a campaign, based on donation history.",minimal_risk | |
| A hybrid-office desk-booking tool matching employees to available desks and meeting rooms.,minimal_risk | |
| "We are developing an AI system that will an AI tool aggregating anonymised travel data to spot broad migration trends, with no individual identified.",minimal_risk | |
| Our client wants to deploy an AI tool to software reconstructing a 3D model of a crime scene from photos for investigators to review.,minimal_risk | |
| A startup is building a product that uses AI to an AI tool giving employees feedback on voluntary training-module quizzes for their own development.,minimal_risk | |
| Software identifying religious minority group members to push them toward exploitative service agreements.,prohibited | |