When you think of AI chess you probably imagine some unbeatable super AI that will be 5 steps ahead and beat you every time. That’s not a bad description for most chess AI.
Maia however isn’t like other chess AI’s. Unlike other artificial intelligence’s it is designed not to win, but to play as much like a human as possible. Maia’s mission is to develop an understanding and be able to predict human moves in an effort to help develop AI that can work better with and better train people.
The Typical AI using deep learning has become extraordinarily good at human games like chess. They play against themselves to become borderline unbeatable.
Maia however draws its knowledge from actual human players. By analysing human games the AI can predict what a human would be most likely to do. A key aspect of Maia that makes it different is understanding mistakes, predicting good moves is easy for an AI but making the moves it shouldn't is much more tricky.
Because of the nature of chess and how objective it is, it’s easy for a computer to find the right moves and avoid any mistakes. Getting a computer to make wrong moves is much more difficult since there is no way to teach a computer why we make them.
Regardless the team behind Maia has said even when human players make “horrific blunders” Maia can predict them around 25% of the time.
The ability to have an AI that can predict the most human decision rather than the best decision could have a massive impact well beyond chess. For example, an AI that could help train doctors on medical imaging may be more effective if it “thinks” like a human and can predict the likely errors.
But what do you think? It may only be a chess AI at the moment but its potential implications are massive. Where else do you think this could be implemented?