By The AI has learned to play chess in ways hitherto seen by grandmasters
One of DeepMind’s projects, AlphaZero, has broken new ground once again by learning to play chess like a human It can now develop its own tactics, understand positioning, and has amazed one of the greatest players of all time.
DeepMind’s AlphaZero AI is the most advanced of its kind, and works by thinking like a human brain. Known as a neural network, where the AI keeps information to hand when making decisions, AlphaZero is able to remember previous outcomes or results, and use those to influence decision making for similar problems in the future.
Its creators used this to play chess, starting it out with only the game’s basic rules. AlphaZero then played against itself millions of times, learning and remembering as it went. Within an incredibly short period of time, a matter of hours infact, it was playing at a world-class level.
After just a few hours of mental sparring, AlphaZero wasset against the world’s most powerful chess machine, Stockfish, which is able to calculate 60 million moves a second. AlphaZero came out comfortably victorious. In a 1,000-game match, AlphaZero bested Stockfish 155 times, losing only six games and drawing the remaining fixtures.
The victory itself is impressive, but what has experts stunned is how AlphaZero was able to achieve the win.
A chess machine’s main advantage is its ability to calculate significantly more moves than its opponent, calculating how to keep its own pieces and take the enemy’s. AlphaZero, which can only process 60,000 moves a second, went up against Stockfish with its own style and patterns, appearing and playing with formidable creativity and intuition, much like a human.
Human players will make seemingly outlandish moves or sacrifices for future gain, and that’s exactly what AlphaZero did. The AI would move uniquely, gaining and losing positioning or pieces for a long-term line of play, just like grandmasters of the sport have done. However, it was doing this in ways which even the world of chess hadn’t seen before.
Grandmaster Matthew Sadler, who’s been studying AlphaZero for his book, Game Changer, describes watching the AI like “discovering the secret notebooks of some great player from the past.”
AlphaZero came up with tactics which went against the long-standing, tried and tested chess game-theory. It would move its own king in a risky manner, or surround the opponent’s in an equally unorthodox style. However, opening after opening passed, and these tactics held.
To the satisfaction of its creators, DeepMind has taken historical theory and turned it on its head. The ambition for the AI has always been to use it in solving humanity’s greatest problems. However, by succeeding in this original manner in the game of chess, there’s sincere hope that such achievements can transcend into hugely important scientific fields.
DeepMind has recently bested experts at predicting the 3D shapes of genes, a mind-bogglingly complex yet incredibly important part of science. However, if it starts looking at problems in entirely new ways, then solutions to some of our greatest challenges could be just around the corner.
One of the greatest players of all time, Gary Kasparov, marvelled at how the AI plays. “I can’t disguise my satisfaction that it plays with a very dynamic style, much like my own!” he said, humbly. In 1997 Kasparov lost to IBM’s Deep Blue and took the loss pretty badly, so to gain the approval of a famously difficult to impress player is quite an achievement.
Comments