Maia Chess

Maia is a chess engine designed to play like a human, not like a computer. Most chess engines aim to play perfectly and crush any opponent; Maia deliberately plays at human skill levels, from beginner (1100 rating) to advanced amateur (1900 rating). The benefit is that you get a chess opponent that makes the kinds of mistakes and strategic choices a real person would make, rather than inhuman perfection.

The way it works is through a neural network trained on millions of real games played on Lichess, an online chess platform. Researchers fed the engine games from players at different skill levels, and it learned to mimic the decision-making patterns of each group. When you ask it to make a move, it doesn't search dozens of moves ahead like traditional engines do; instead, it uses what it learned from human players to pick a single best move. You can play against three of these models right now on Lichess under the names maia1, maia5, and maia9.

This project is useful for several audiences. Chess students can practice against an opponent at their own level without the frustration of playing a superhuman AI or the unpredictability of random opponents online. Researchers interested in AI alignment—making AI systems behave more like humans want them to—can study how the engine captures human behavior. Game developers or chess trainers might also use these models to power better teaching tools.

The repository includes the trained models ready to download and use, along with code for training your own models from any set of chess games. It also includes tools for analyzing and predicting blunders in chess positions. While the README doesn't spell out all the technical details, the project is grounded in academic research (published at a major AI conference) and is designed for reproducibility, so others can retrain the models or build on the approach.