Google AI teaches itself ‘superhuman’ chess skills in four hours

Human chess grandmaster Peter Heine Nielsen tells the BBC that he’s “always wondered how it would be if a superior species landed on earth and showed us how they played chess.”

Well, move aside, ugly, giant bags of mostly water: now we know, because Google’s “superhuman” AlphaZero artificial intelligence (AI) taught itself chess from scratch in four hours. Then, it wiped the floor with the former world-leading chess software, Stockfish 8.

AlphaZero is actually a game-playing AI created by its Google sibling, DeepMind. DeepMind Technologies Ltd., a Google subsidiary, created a neural network that learns how to play video games in a fashion similar to that of humans.

That neural network had to learn how to play chess – without human interaction, mind you – because until recently it was a Go specialist that had confined itself to going around beating the world’s best Go players in its incarnation as AlphaGo.

Now that AlphaZero has been generalized, it can learn other games. After learning the rules to chess in four hours, it took on a 100-game match with Stockfish 8, which is an open-source chess engine that consistently ranks first or near the top of most chess-engine rating lists.

In the AlphaZero/Stockfish 8 games, AlphaZero won or drew all 100 games, according to a non-peer-reviewed research paper published by the DeepMind crew with Cornell University Library’s arXiv. It garnered 28 wins, 72 draws, and zero losses.

From the paper, whose authors include DeepMind founder Demis Hassabis: a child chess prodigy who reached the rank of chess master at the age of 13:

Starting from random play, and given no domain knowledge except the game rules, AlphaZero achieved within 24 hours a superhuman level of play in the games of chess and shogi [a similar Japanese board game] as well as Go, and convincingly defeated a world-champion program in each case.

Former world chess champion Garry Kasparov told Chess.com that AlphaZero’s performance is “remarkable”:

It’s a remarkable achievement, even if we should have expected it after AlphaGo. It approaches the ‘Type B,’ human-like approach to machine chess dreamt of by Claude Shannon and Alan Turing instead of brute force.

According to Chess.com, AlphaZero is like humans in that it searches far fewer positions than its predecessors. The paper claims that it looks at “only” 80,000 positions per second, compared to Stockfish’s 70 million per second.

In fact, the DeepMind programmers used a specific type of machine learning – reinforcement learning – to train AlphaZero. From Chess.com’s writeup:

Put more plainly, AlphaZero was not “taught” the game in the traditional sense. That means no opening book, no endgame tables, and apparently no complicated algorithms dissecting minute differences between center pawns and side pawns.

This would be akin to a robot being given access to thousands of metal bits and parts, but no knowledge of a combustion engine, then it experiments numerous times with every combination possible until it builds a Ferrari. That’s all in less time than it takes to watch the “Lord of the Rings” trilogy. The program had four hours to play itself many, many times, thereby becoming its own teacher.

Not all grandmasters are fully satisfied with the way the match was set up. They’re debating the processing power of the two adversarial systems, while American GM Hikaru Nakamura reportedly called the match “dishonest”, pointing out that Stockfish’s methodology requires it to have an openings book for optimal performance. Another expert, GM Larry Kaufman, said he wants to see how AlphaZero would do on a home machine, as opposed to Google’s souped-up computers.

But aside from arguments about the fairness of the match, experts say that we’re looking at actual AI at this point. From here, we could see much more than chess wins. Chess.com quotes GM Peter Heine Nielsen:

It goes from having something that’s relevant to chess to something that’s gonna win Nobel Prizes or even bigger than Nobel Prizes. I think it’s basically cool for us that they also decided to do four hours on chess because we get a lot of knowledge. We feel it’s a great day for chess but of course it goes so much further.