Go to Contents Go to Navigation

(LEAD) S. Korean Go player 'speechless' over Google AI

All News 19:14 March 10, 2016

(ATTN: ADDS more comments, photo)
By Joo Kyung-don

SEOUL, March 10 (Yonhap) -- First, he was surprised. Now, he is speechless.

After suffering his second straight defeat against Google's artificial intelligence (AI) program AlphaGo, South Korean Go player Lee Se-dol said Thursday he accepts his "complete defeat" and is "speechless" about the self-learning machine's playing level.

"From the start, I never thought I was taking the lead," he said after the match. "Yesterday, I felt something strange in its moves, but today, I think AlphaGo played a perfect game."

In their second showdown at the Four Seasons Hotel in Seoul, Lee dropped his game in 211 moves after playing nearly four and a half hours. The score is now 2-0 in favor to AlphaGo at the five-round Go tournament named Google DeepMind Challenge Match.

Lee, a ninth-dan player who went professional at the age of 12, did make a little progress from the first match where he threw the towel in 186 moves. This time, the 18-time international event winner forced the AlphaGo to overtime, but couldn't overcome the machine's cold calculation in the end.

"I couldn't find its weaknesses, that's why I lost," Lee said.

After his first loss on Wednesday, Lee said that it's now a 50-50 game, a big retreat from his last month's comment of winning 5-0 or 4-1 in his favor. But the 33-year-old Go master is now lowering his expectations more than that.

"I will now try to put my best foot forward to win at least one game," he said. "If you look at today's game, it's difficult to win when the game moves into middle phase, so I will try to make decisive moves early."

Lee, who has won 47 Go events including 18 international trophies, now has to win all the remaining three matches to collect the US$1 million prize. The matches are slated for Saturday and Sunday at the same venue, before closing out next Tuesday. The prize money will be donated to UNICEF and other charities if AlphaGo wins.

Fellow South Korean professional Go players also praised AlphaGo for its reading of sequence and wrapping the game.

"Unlike yesterday, Lee played too safely," said Yoo Chang-hyuk, a ninth-dan Go player who commentated the match on Thursday. "I thought AlphaGo had a problem with wrapping up the game, but it did very well today."

Demis Hassabis, the CEO of Google's London-based AI company DeepMind, said he is also "speechless" over AlphaGo's performance. The 39-year-old Briton said that its AI had estimated the end of the game.

"We are very pleased that AlphaGo showed pretty beautiful and amazing moves as commentators stated," he said. "AlphaGo seems to know what was happening."


Send Feedback
How can we improve?
Thanks for your feedback!