Google’s AI just took a huge step forward by winning an ancient game
In one of the bigger sports/gaming upsets of the last few years, Google’s artificial intelligence research team DeepMind defeated a world champion Go player. And it is a really, really big deal.
One of the obvious hurdles facing AI is the ability to adapt and offer – or at least accurately mimic – something akin to intuition. Unlike chess and other games researchers love to throw at AI, Go requires a degree of instinct, evaluation, and adaptability. That’s true of many games, but in most cases AI can “memorize” a strategy based on previous matches someone else play. Go can’t really be won on programming alone.
To test the newest iteration of Google’s AI, AlphaGo, the researchers responsible decided to challenge South Korean Go champion, Lee Se-dol. DeepThought has defeated other Go players in the past, but Lee is considered to be in a different league. The 33-year old champion holds a Go rank of “9-dan,” the highest class of player in the world. An estimated 24 million people play Go worldwide, and Lee is in a class of a few dozen, if that.
Lee and DeepThought set a four-hour time limit on the game. With 28 minutes to go, Lee resigned.
“I was very surprised,” Lee said post-match. “I didn’t expect to lose. I didn’t think AlphaGo would play the game in such a perfect manner.”
The game of Go originated in China over 2,500 years ago, and has been played by amateurs and masters alike around the world. Two players face off on a set board, using white and black stones in an attempt to surround more territory than their opponent. The number of possible outcomes in a game is shockingly vast. To put it in perspective, chess has an estimated 10120 possible games, while Go has 10761.
“I don’t regret accepting this challenge,” Lee said. “I am in shock, I admit that, but what’s done is done. I enjoyed this game and look forward to the next. I think I failed on the opening layout so if I do a better job on the opening aspect I think I will be able to increase my probability of winning.”
Lee agreed to play five games against AlphaGo for a potential prize of $1 million. Given the stakes, and given that Lee admits to underestimating the AI, the remaining matches may have dramatically different results. Regardless the AI win is still a significant accomplishment and a huge step forward for AI in general.
The remaining games will be held on March 10, 12, 13, and 15.