DeepMind’s AI became a superhuman chess player in hours

The endgame of DeepMind, Google’s artificial intelligence subsidiary, has never been about beating people at board games. It’s always been about creating something akin to a combustion engine for intelligence – a generic thinking machine that can be applied to a wide range of challenges. The company is still a long way from achieving that goal, but new research published by its scientists this week suggests that they are at least on the right track.

In the article, DeepMind describes how a descendant of the AI ​​program that first conquered the board game Go taught itself to play a number of other games at a superhuman level. After eight hours of self-playing, the program beat the AI ​​which first beat the human Go world champion; and after four hours of training, he beat the current world champion chess program, Stockfish. Then, for a victory lap, he trained for just two hours and beat one of the best in the world shogi-playing programs named Elmo (shogi being a Japanese version of chess that is played on a larger chessboard).

One of the major breakthroughs here is that the new AI program, named AlphaZero, was not specifically designed to play any of these games. In each case, it was given some ground rules (like how knights move in chess, etc.) but was programmed with no other strategies or tactics. It simply got better by replaying itself over and over again at an accelerated pace – an AI training method known as “reinforcement learning”.

Using reinforcement learning in this way is not new in itself. DeepMind engineers used the same method to create AlphaGo Zero; the AI ​​program that was unveiled last October. But, as this week’s article describes, the new AlphaZero is a “more generic version” of the same software, meaning it can be applied to a wider range of tasks without first being primed.

What’s remarkable here is that in less than 24 hours, the same computer program was able to learn to play three complex board games to superhuman levels. This is a new feat for the AI ​​world.

This brings DeepMind one step closer to building the generic thinking machine the company dreams of, but there are major challenges ahead. When DeepMind CEO Demis Hassabis introduced AlphaGo Zero earlier this year, he suggested that a future version of the program could help solve a range of scientific problems, from designing new drugs to discovering new materials. . But these problems are qualitatively very different from simply playing board games, and there is still a lot of work to be done to find out exactly how algorithms can solve them. All we can say for sure now is that artificial intelligence has stopped playing chess for good.

Comments are closed.