Google DeepMind’s AlphaGo artificial intelligence program won the first of five Go games in a milestone million-dollar match against South Korean champion Lee Sedol today – marking another milestone for machine learning.
“I am in shock. … But I am looking forward to tomorrow,” Lee was quoted as saying afterward.
AlphaGo notched its first victories against a professional Go player in October when it beat European champion Fan Hui, five games out of five. But experts in the centuries-old game thought the AI program would have a harder time with Lee, who is more highly ranked on the Go circuit.
Lee ran out of options for the endgame and surrendered after about three and a half hours of play. “A big surprise, I think,” commentator Michael Redmond said during the webcast from Seoul.
Lee Sedol arriving for post-game press conference pic.twitter.com/27iAqcx94w
— American Go Assoc. (@theaga) March 9, 2016
Go is arguably the world’s most complex game because its 19-by-19 playing board provides for 10170 legal positions, compared with 1050 positions for chess. This week’s match has been compared to the historic chess duel between IBM’s Deep Blue computer and chess champion Garry Kasparov in 1997. In both cases, many experts doubted whether the machine could beat the human.
Because Go is more complex than chess, AlphaGo had to be programmed to take an innovative approach to learning the game. Researchers had the computer analyze a database of Go games, then play against itself millions of times to improve its performance.
Google Deepmind’s Demis Hassabis has said the approach can bring about more practical advances in artificial intelligence, ranging from better smartphone virtual assistants to AI-assisted scientific research.
The next game is due to begin today at 8 p.m. PT. Every game will be webcast via Google DeepMind’s YouTube channel.