Maia is a neural network chess engine that allows you to play chess like a human. Maia learns the meaning of human chess to realize more human-like behavior than other chess AIs, so chess players can learn what mistakes they will make with an AI model.
Maia is a deep learning framework like AlphaZero, a computer program developed by DeepMind, a Google-affiliated developer of AlphaGo, which gained attention by winning pro Go articles. Maia not only plays chess independently and learns about chess, but also learns chess online like a human. Maia trains in notation for hundreds of times to properly predict the meaning and method of human beings.
When looking at the percentage of chess AI matching with human play, Stockfish showed less than 40% and Leela had more than 40% match, but Maia showed a higher ratio of 50% or more. Among them, it is possible to play the most human chess.
In addition, an experiment is being conducted to train Maia separately using group notation by gathering players playing chess online and dividing them into nine groups. It is expected that this experiment will allow players of a certain level to learn what chess methods to play.
Maia trained by group is called Maia 1100~1900, and the number after Maia trained by a competent player group increases. Learning chess AI focuses on making predictions easier. But Maia predicts human error and can help understand how an inexperienced chess player makes mistakes. In other words, Maia is a useful chess learning tool.
The prediction accuracy can be further improved by training only specific chess player notation for each model Maia 1100-1900. In fact, it is said that Maia 1900 was trained using a specific chess player notation and used for training to predict accuracy up to 75%. The Maia source code is also available on GitHub . More information about Maia can be found here .