Alpha Zero learns chess in 4 hoursbeats Stockfish

Google’s AI teaches itself chess in 4 hours, then convincingly defeats Stockfish

There has just been a revolutionary development in the world of AI, and in the world of chess.

Google’s Artificial Intelligence project, DeepMind explains they’re on a scientific mission to push the boundaries of AI, developing programs that can learn to solve any complex problem without needing to be taught how. A little over a year ago, DeepMind released AlphaGo, which sensationally defeated the world champion of the famously CPU unfriendly ancient Chinese game, GO.

Now their AlphaZero program has kicked up a storm in the chess world. It obliterated Stockfish 8 in a closed-door, 100-game match with 28 wins, 72 draws, and zero losses. Stockfish is the go to chess engine for most top players today, and is generally regarded as the strongest chess playing engine in the world. Stockfish recently won the 2017 Chess.com Computer Chess Championship, and has an estimated ELO strength of over 3400. To put that into context, the world’s top-ranked human player, Magnus Carlsen has an Elo rating of 2837, at the time of writing.

AlphaZero appears to have totally outclassed Stockfish 8. And that’s not even the most astonishing part of it.

The biggest kicker is that it did so after teaching itself chess in a mere 4 hours. That’s right, the programmers of AlphaZero had it use a type of machine learning, specifically reinforcement learning. Basically, AlphaZero was not “taught” the game in the traditional sense – no opening book, no endgame tables, and apparently no complicated algorithms dissecting minute differences between center pawns and side pawns. The program was simply provided the rules of chess. It then had four hours to play itself over and over again, thereby figuring out what works and what doesn’t.

Former world champion, Garry Kasparov has represented us humans several times against the best of the machines. “It’s a remarkable achievement, even if we should have expected it after AlphaGo,” he told Chess.com. “It approaches the ‘Type B,’ human-like approach to machine chess dreamt of by Claude Shannon and Alan Turing instead of brute force.”

As of now, the programming team is keeping quiet. They point out that the paper is “currently under review” but you can read the full paper here.

So far, Google has published only 10 of the 100 games.

Of course, it is intriguing to see how much different the games of an AI that has learnt the game from scratch, without the inbuilt bias of established chess theory, will be from our own. GM Peter Heine Nielsen, the long-time second of World Champion GM Magnus Carlsen said, “After reading the paper but especially after seeing the games I thought, well, I always wondered how it would be if a superior species landed on earth and showed us how they play chess. I feel now I know.”

Of AlphaZero’s 28 wins, 25 were with white, and 3 were with black pieces. Over time, the machine also displayed an affinity for certain openings. Interestingly, 1. d4 was the opening of choice with white. The standard golden rules of chess such as piece development, control of the centre, etc. were mostly on display as expected. However, it did show an increased tendency for sacrificing material to gain positional advantages early on in matches.

“We have always assumed that chess required too much empirical knowledge for a machine to play so well from scratch, with no human knowledge added at all,” Kasparov said. “Of course I’ll be fascinated to see what we can learn about chess from AlphaZero, since that is the great promise of machine learning in general—machines figuring out rules that humans cannot detect. But obviously the implications are wonderful far beyond chess and other games. The ability of a machine to replicate and surpass centuries of human knowledge in complex closed systems is a world-changing tool.”