Artificial intelligence trains chess players instead of defeating them
Artificial intelligence can help chess players instead of just trying to defeat them. Maia computer algorithms find the mistakes that its opponents have to work on.
A new chess algorithm called Maia can find errors that players of different levels have to work on. Artificial intelligence has become so good at chess that its only rival is now other computer programs. Indeed, man has 15 years to beat a machine in a chess tournament.
It's an impressive technical achievement, but this dominance has also made it a top level chess less imaginative, as players increasingly follow strategies generated by inanimate algorithms.
However, a new research paper shows that artificial intelligence could still pique the interest of chess players. The study authors developed a chess machine, which unlike most of its predecessors, their system is not designed to defeat humans. Instead, she is programmed to play like them.
Researchers believe that Maia could make the game more fun. But it could also help us learn better chess, from the computer.
"Thus chess becomes a place where we can understand human ability, through the perspective of super-intelligent artificial intelligence." said one of the scholars Jon Kleinberg, Professor at Cornell University.
Their system - called Maia - is a customized version of AlphaZero, a program developed by DeepMind Research Lab to dominate chess, Shogi and Go.
Instead of building Maia to win a game of chess, the model was trained in individual moves made by humans. Study co-author Ashton Anderson said this allowed the system to identify where players should work.
Maia has algorithmically identified which errors are typical at which levels, and therefore which errors people should work on and which errors they probably should not, because they are still very difficult to level.
Maia may not be able to teach people how to master AI in chess - but it could help them beat their fellow human beings.
You can read the study document on the arXiv server.