Google DeepMind’s AlphaGo artificial intelligence program will take home the $1 million prize after winning the first three games in its Go showdown with South Korean champion Lee Sedol.
“Folks, you saw history made here today,” webcast host Chris Garlock said.
But today’s third win isn’t the end of the historic match in Seoul: The last two games will still be played, with Lee hoping to demonstrate that it’s possible for a human to beat the computer program.
“I think it’s going to be tough going,” match commentator Michael Redmond said during today’s webcast. Lee was never able to achieve an advantage in the third game, which lasted more than four hours. More than 65,000 viewers watched the YouTube webcast at its peak.
After today’s game, DeepMind co-founder Demis Hassabis paid tribute to Lee, and particularly to the “really huge ko fight” that the champion executed during the endgame.
“To be honest, we are a bit stunned and speechless,” Hassabis told reporters. “Lee Sedol put up an incredible fight again.”
Lee apologized for his performance, and said he let the pressure get to him during the third game. “I should have shown a better outcome. … I kind of felt powerless,” he said.
#AlphaGo won game 3 and the match! Historic moment. In complete awe of Lee Sedol’s incredible genius, and proud of the amazing AlphaGo team!
— Demis Hassabis (@demishassabis) March 12, 2016
The duel marks a milestone for AI, and for the millennia-old game of Go. Comparisons have been drawn to chess champion Garry Kasparov’s defeat in a 1997 match against IBM’s Deep Blue computer, and the triumph of IBM’s Watson computer over human champions in the “Jeopardy” TV quiz show.
Go, which involves putting black and stones down on a 19-by-19 board, may look simpler than chess. But it’s considered more intellectually challenging because there are so many more possible permutations of the pieces.
“Go is a very beautiful game, and I think it teaches a lot about life – much more so than a game like chess,” Sergey Brin, Google’s co-founder and the president of its holding company, Alphabet, said during today’s press conference. “When you watch great Go players play, it is like a thing of beauty. So I’m very excited that we have been able to instill that kind of beauty in our computers.”
Before AlphaGo broke onto the scene, experts in AI thought it might take a decade for game-playing programs to match the best professional players. But Google DeepMind changed the terrain when it announced in January that AlphaGo had beaten European Go champion Fan Hui.
AlphaGo took advantage of two strategies that give it a more humanlike learning style, but at an accelerated pace. It drew upon a database of high-level Go games played in the past, and then played against itself millions of times to learn how to value various board positions.
Even after AlphaGo beat Fan Hui in five straight games, the program continued to improve. By the time the Seoul showdown began, AI experts were predicting that AlphaGo would prevail, said Oren Etzioni, CEO of the Seattle-based Allen Institute for Artificial Intelligence.
Wow: the moment #Go fell to ML
AlphaGo 3 – Lee Sedol 0
What an incredible achievement by the #AlphaGo team! pic.twitter.com/JYU6BEF08q
— Chris Burgess (@cpburgess_) March 12, 2016
“It’s not ‘man versus’ machine, but rather an impressive triumph for the brilliant (human) researchers and engineers at DeepMind and Google,” Etzioni told GeekWire in an email. “Of course, they are building on a great deal of earlier work in AI. I’d say that their ‘overnight’ success has been 50-plus years in the making!”
Etzioni emphasized that AlphaGo’s success doesn’t translate into the more generalized intelligence that humans draw upon every day. That’s a big topic for the Allen Institute for Artificial Intelligence, also known as AI2.
“It will take decades to generalize this success from a board game to real-life situations that have nuance, ambiguity and far less labeled data,” he said. “Still, we are in an ‘AI spring,’ and AI2 is growing commensurately.”
Don’t expect AlphaGo to be spending the million-dollar prize on computer-chip bling: DeepMind’s Hassabis says the money will be donated to UNICEF, science education programs and Go organizations.
The fourth game will be webcast starting at 8 p.m. PT today. The final fifth game gets under way at 8 p.m. PT Monday. And there’s already talk about a follow-up match between AlphaGo and Chinese champion Ke Jie, who is currently the world’s top-rated Go player. (Lee is rated No. 4.)
