WASHINGTON, D.C. – Both sides in next month’s big $1 million AI-vs.-human Go match say they’re confident they’ll prevail. But Google DeepMind’s AlphaGo program has a secret weapon: It’s expanding its knowledge of the game exponentially during the buildup to the five-game match against top-ranked player Lee Sedol in Seoul, South Korea.
Last month, researchers at Google DeepMind shook up the Go world with news that its artificial intelligence program bested a European champion, Fan Hui, without being given any advantage to start with. The research, published in Nature, lays out a potentially more powerful approach to AI that combines deep learning with reinforcement learning.
Next month’s match against Lee could be as big for fans of Go (and followers of AI research) as IBM’s Deep Blue victory over chess champion Garry Kasparov was in 1997. The match will be streamed live from Seoul via You Tube from March 9 to 15.
“This really is our Deep Blue moment,” Demis Hassabis, Google DeepMind’s president of engineering, said this weekend at the American Association for the Advancement of Science’s annual meeting in Washington.
Go, which originated in China thousands of years ago, involves the strategic placement of black and white stones on a 19-by-19 board. It may look simple, but it’s considered to be the most complex game ever devised.
AlphaGo beat Fan Hui in all five games of their match, but Lee says he’ll fare better. “I’m honored to play against an AI invented by Google,” Lee said in a statement. “I regard this to be an important match in the history of Go, so I accept the challenge. I’m confident that I can win the match.”
Hassabis said most Go players are giving Lee the edge over AlphaGo. “They give us a less than 5 percent chance of winning … but what they don’t realize is how much our system has improved,” he said. “It’s improving while I’m talking with you.”
Online predictions have shifted back and forth: The current odds on BitBet favor Lee, but the just-for-fun prediction from GoodJudgment gives AlphaGo a little more than a 50-50 chance of winning.
For Hassabis, the AlphaGo project is about much more than beating one of the world’s best Go players. The principles that are being put to work in the program can be applied to other AI challenges as well, ranging from programming self-driving cars to creating more humanlike virtual assistants and improving the diagnoses for human diseases.
“We think AI is solving a meta-problem for all these problems,” Hassabis said.
Even if AlphaGo doesn’t prevail next month, it’s likely to be just a matter of time before the computer wins. That’s how it went for Deep Blue, which lost to Kasparov in 1996 before triumphing a year later. But we can take solace in Hassabis’ assessment that it will be decades before AI programs match humans in what’s known as “general intelligence” – that is, the ability to handle a wide variety of life’s intellectual challenges rather than individual specialized tasks.
In the meantime, we humans can watch from the grandstands as Google’s AlphaGo takes on AI challengers such as Facebook’s Go bot.
Update for 9:15 a.m. PT Feb. 15: This report has been updated with a shift in odds on BitBet, which occurred after this report was published (and mentioned on BitBet).
