Just 4 hours Google «nurture" the best chess player in history

Chess - not an easy game. At least by human standards. But for heavy duty artificial intelligence is the usual fun that can be learned in a very short time. In an article published in Arxiv.org online library, from Google engineers in detail tell about how their newest AI configuration AlphaZero could develop a "superhuman" in chess, just a few hours.

Just 4 hours Google «nurture

Once the system have only a basic set of rules (no strategies), AlphaZero took only four hours in which to master the game so well that before it fell even the strongest chess engine in the world Stockfish.

After a series of 100 games against Stockfish AI algorithm AlphaZero won 25 times a game with white pieces and three times per game blacks. All other parties, both systems are tied. As a result, Stockfish failed to win a single time, and AlphaZero, in turn, not even once did not lose.

"We have a new master of chess. This no doubt will revolutionize the game, but it would be nice to think about how it can be applied outside of chess ", - commented on the results, David Kramaley researcher chess game Chessable and chief editor of the site.

AlphaZero system is based on the achievements of AI algorithms and AlphaGo AlphaGo Zero, the creators of which are engineers Artificial Intelligence Laboratory DeepMind, by Google. DeepMind working to improve these AI algorithms for several years, along the way defeating them with the help of the world's best players in an ancient game of logic. The culmination of winning the series took place in October of this year, when a new and completely self-contained version of AI algorithm AlphaGo enrolled solely on the game with itself, not against human opponents, won all its previous versions. At the same time the previous version AlphaGo Zero algorithm partially studied the game with the help of observations of how people play. Thus, the developers wanted to help her study the strategies of the game, but as it turned out, in fact it only slowed her development. Full provision itself in the training affected the significantly better results AlphaGo Zero in the competition one by one.

"It's like watching an alien civilization, inventing their own math", - said in an interview with Gizmodo portal in October, Nick Hynes, the MIT scientist, working in the field of computer science.

"What we see here is an independent, self-sufficient model created with excluded factor of human prejudices. It alone is able to learn what she thinks best, that certainly will differentiate it from our own concepts. "

However, the development of AI sphere is so rapid that what has been achieved, there could already be out of date to October of this year. In the published article engineers DeepMind noted as the latest version of AlphaZero moved to a new level and is able to address a wider range of tasks. This means that now AlphaZero capable of not only great to play chess. It also copes with Shogi (Japanese board game), and the second, in which she was able to achieve perfection in just 2 and 8 hours, respectively. So far, neither Google engineers or engineers DeepMind openly make comments on the new results of their work, they expect a third-party and independent peer review articles in Arxiv.org. However, it is clear that the ascent to the heights of the algorithm of artificial intelligence is still far from being completed, but now it is so steep that it is recognized even grandmasters.

"I always wanted to learn how to look a game of chess between the more advanced species, have landed on Earth. Now I know, "- commented on the results AlphaZero grandmaster Peter Heine Nielsen.