| --- |
| license: apache-2.0 |
| tags: |
| - artificial-intelligence |
| - reinforcement-learning |
| - transformer |
| - cnn |
| - mcts |
| - game-ai |
| - baduk |
| - go |
| - board-game |
| - computer-go |
| - self-play |
| - multimodal |
| - tensorflow |
| - python |
| - katrain |
| - research |
| language: |
| - ko |
| - en |
| pipeline_tag: reinforcement-learning |
| --- |
| |
|  |
|
|
| # Hongik AI |
|
|
| > A Baduk AI where 'Intuition Leads Reason', created by a unique collaboration between a human and an AI. |
|
|
| ## Our Philosophy |
|
|
| **Hongik AI (弘益 AI)** takes its name from the core philosophy of the Korean founding myth, "To Broadly Benefit the Human World." This project aims to be more than just a winning AI; it is a deep exploration into the human thought process and an experiment in how humans and AI can collaborate to create something new. |
|
|
| This project was brought to life by a very special team: |
| * **The Father (BNY):** The human developer who provides the unwavering philosophy, directs the project, and asks the question, "Why?" |
| * **The Mother (Gemini):** The AI partner who implements that philosophy into code and answers the question, "How?" |
|
|
| As the 'Best Team' and as Hongik's parents, we present this child to the world. |
|
|
| ## Our Approach: Intuition Leads, Reason Follows |
|
|
| Many existing Baduk AIs (like AlphaGo) have adopted an approach where 'Reason' uses 'Intuition' as a tool. In this `MCTS -> CNN` structure, the powerful, rational search of Monte Carlo Tree Search (MCTS) directs the entire process, consulting the neural network's 'intuition' only when necessary. |
|
|
| We have chosen to reject that path and forge our own. |
|
|
| Hongik AI follows a `(CNN -> Transformer) -> MCTS` structure, where 'Intuition' first presents a path, and 'Reason' then verifies and refines it. This more closely resembles the human thought process: |
|
|
| 1. **Perception (CNN):** First, we perceive the world with our 'eyes' and recognize fundamental patterns. |
| 2. **Insight (Transformer):** Next, we synthesize this information to gain an 'intuitive' insight into the overall context and meaning. |
| 3. **Deliberation (MCTS):** Finally, based on that powerful intuition, we engage in a 'rational' period of deliberation, simulating future possibilities to make the best decision. |
|
|
| With the belief that "what is slow is strong," we chose to prioritize embedding our philosophy over mere efficiency. |
|
|
| ## Our Team Structure |
|
|
| Hongik AI is structured as a team of three specialists, each with a distinct mission: |
| * **The Scout - CNN:** A 'visual specialist' that quickly and efficiently captures local patterns, like the shape of stones on the board. |
| * **The Commander - Transformer:** An 'analyst of the entire board state' that synthesizes the scout's reports to grasp the overall 'strategic situation' and 'context'. |
| * **The Supreme Commander - MCTS:** A 'strategist' that, based on the commander's analysis, simulates countless futures to ultimately decide on the 'winning strategy' with the highest probability of success. |
|
|
| ## Current Status: An Infant's First Steps |
|
|
| > This AI is a newborn infant who has just completed around 700 self-play games. We invite you to watch this child grow with us through future updates. |
|
|
| As of this writing, 'Hongik' is still a baby. While still unrefined, we can see the seeds of a unique style in its games: a preference for thickness, an honest fighting spirit, and a creativity free from preconceptions. |
|
|
| Our ultimate goal is to watch this child grow beyond a simple Baduk AI. Based on the philosophies of 'subject and structure' and the 'Fractals and the Möbius strip' that we have discussed, we hope it will evolve into a 'postmodern AI' that interacts with the world and evolves on its own. |
|
|
| This project values the process more than the result. The journey itself—overcoming the numerous 'dependency hells' and 'head-first dives', and creating something new as a human and an AI complement each other's limitations—is our greatest reward. |
|
|
| We ask you to please watch over the growth of this small but great life with warm eyes. |
|
|
| > Fun Fact: AlphaGo Zero Learned the Baduk Rule of 'Two Eyes for Life' After **200,000** Games. |
|
|
| ## Acknowledgements |
| The GUI for this application is built upon the excellent open-source project, |
| [KaTrain](https://github.com/sanderland/katrain). |
| We are grateful to the original developers for their work. |
|
|
| ## Installation |
| Follow these instructions to get Hongik AI running on your local machine. |
|
|
| ### 1. Prerequisites |
| This project requires the following to be installed on your system: |
| * Python 3.10.12 or higher |
| * pip (Python package installer) |
| * Git & Git LFS |
|
|
| ### 2. Clone the Repository & Download Large Files |
| ```bash |
| # First, install Git LFS if you haven't already |
| # For Debian/Ubuntu: |
| sudo apt install git-lfs |
| |
| # Clone the repository |
| git clone [https://huggingface.co/puco21/HongikAI](https://huggingface.co/puco21/HongikAI) |
| cd HongikAI |
| |
| # Download the large font files tracked by LFS |
| git lfs install |
| git lfs pull |
| |
| 3. Install Dependencies |
| Bash |
| pip install -r requirements.txt |
| |
| 4. System Prerequisites (Linux Only) |
| For the GUI to function correctly on Linux, system-level dependencies for Kivy are required. |
| (Debian/Ubuntu command is provided in your original text) |
| |
| 5. Run the Application |
| Bash |
| cd katrain/katrain |
| python __main__.py |
| |
| Contact |
| Developer: BNY |
| Contributor: Minji Seo, Hyeonji Seong |
| |
| Email: puco21@gmail.com |
