The researchers used the open source Leela system, based on DeepMind’s AlphaZero, to develop Maia. They trained the model on individual moves from millions of human games online rather than for the sole purpose of winning. This approach allowed researchers to adjust Maia for different skill levels. They have trained several versions of AI at different skill levels. They designed nine robots to play with humans who have notes between 1100 and 1900 (in other words, more novice players to strong amateurs).
In December, researchers made Maia available on the chess site lichess.org. People played over 40,000 games against him in the first week. The researchers found that Maia matched human movements more than half the time at each skill level, with accuracy increasing as the skill level increased. They said the system was able to learn what types of mistakes players were making at different skill levels and recognize the skill level at which people stopped making those mistakes.