Maia Chess

Maia is a Leela-style engine that learns from human games rather than its own games, and aims to make human-like rather than optimal moves. In a given position, Maia predicts the exact human move up to 53% of the time, while the Leela and Stockfish versions match the human moves about 43% and 38% of the time, respectively. Thus, Maia is the most natural, human-like chess engine to date and represents a model of human play that we will use to build data-driven chess learning tools.
Maia has 9 versions, one for each rating step between 1100 and 1900, trained on over 100 million chess games between chess players. Each Maia reflects a person's chess playing style at their target level: Maia 1100 most accurately predicts a person's play at the 1100 level, and Maia 1900 most accurately predicts a person's play at the 1900 level.
Maia's goal is to make the average move that players of his target level would make. For example, a Maia 1100 game is more like a game played by 1100-rated players than a single 1100-rated player - they tend to average out their idiosyncratic mistakes (but Maia still makes a lot of human-like mistakes!).
Maia's AI was developed by researchers from Cornell University, the University of Toronto and Microsoft.