The first time a computer beat a human chess master happened in 1997 when IBM Deep Blue defeated champion, Gary Kasparov. AI systems have advanced in their ability to play chess to the point where no human has beaten a computer at a tournament in the last 15 years. A team of researchers from Tisch University have developed an AI chess engine that doesn’t seek to beat humans.
Instead, this chess AI engine tries to play like a human to create a more enjoyable chess playing experience for people. Researchers on the project note that it also highlights how computers make decisions differently than humans and how that could potentially help humans learn to do better. One of the researchers in the project, Jon Kleinberg, says that chess is something humans study their entire lives to get good at.
However, computers are better in every possible sense than humans are at playing chess at this point. Kleinberg says chess is a place where we can understand human skill through a lens of super-intelligent AI. The chess engine created during the research is known as Maia.
Maia was released on the free online chess server lichess.org and participated in more than 40,000 matches its first week. The algorithm used in the new chess engine algorithmically characterizes which mistakes are typical of which level of player. It can pinpoint mistakes people should work on and which mistakes they shouldn’t work on because they’re still too difficult for them to grasp.
One challenge with AI systems today is that since they approach problems very differently from humans learning from them is difficult and potentially dangerous. This project aimed to create an AI reducing the disparities between human and algorithmic behavior by training the computer using traces of individual human steps rather than having the AI teach itself to complete an entire task. The AI was trained on individual human chess moves rather than the overall goal of winning a game, essentially teaching it to mimic human behavior.