Have you ever wondered how AI can be used to play ancient board games like Hnefatafl? I came across a fascinating project on Reddit where a developer created an AlphaZero-style system for Hnefatafl, and I’d love to dive deeper into it.
For those who may not know, Hnefatafl is an ancient Scandinavian board game that’s similar to Chess. The game requires strategy and critical thinking, making it an interesting challenge for AI systems.
The developer, who has a background in finance and is self-taught in Python and machine learning, used a combination of self-play, Monte Carlo Tree Search (MCTS), and neural networks to train the AI system. The approach is inspired by the famous AlphaZero paper on playing Go.
What I find interesting is that the developer had to make significant adjustments to the original approach to make it work for Hnefatafl. This highlights the complexity and uniqueness of each board game, even if they share some similarities.
The developer is currently facing some challenges, including limited computing resources, which is forcing them to use shallower searches and fewer games per generation. This has led to limited success so far, but it’s unclear whether the issue lies with the code or the computing limitations.
This project serves as a great example of the challenges and opportunities in applying AI to ancient board games. It also highlights the importance of community feedback and collaboration in overcoming these challenges.
If you’re interested in exploring the code, it’s available on GitHub. The developer is also seeking feedback on their approach, so if you have any expertise in this area, be sure to check it out.
—
*Further reading: [Deep Learning and the Game of Go](https://www.nature.com/articles/nature16961)*