Go to Contents Go to Navigation

(2nd LD) S. Korean Go player loses to Google AI

All News 21:46 March 09, 2016

(LEAD: UPDATES with quotes by Lee and Google official; ADDS photo)
By Joo Kyung-don

SEOUL, March 9 (Yonhap) -- Humans 0, machines 1.

Google's artificial intelligence (AI) program AlphaGo beat top-class South Korean Go player Lee Se-dol in the ancient board game Go Wednesday as the self-learning machine stunned the world by taking the lead at the historic five-match man-versus-computer tournament.

In a landmark showdown held at the Four Seasons Hotel in Seoul, AlphaGo, the AI designed by Google's London-based firm DeepMind, beat Lee by resignation in 186 moves after playing nearly three and a half hours.

"I didn't think I'd lose, and I was quite surprised," Lee told a press conference in Seoul after succumbing to AlphaGo.

Lee said he was surprised with AlphaGo's ability to start the game smoothly and its keymove.

"I thought today's match was going to be difficult for both of us," he said. "There was a keymove which humans wouldn't have thought about making, but I was surprised it actually did it."

Lee, a ninth-dan player who went pro at the age of 12, will play his second match against AlphaGo with a shot at revenge at the same venue on Thursday. The 33-year-old is also slated to face the human-like algorithm on Saturday and Sunday before wrapping up the tourney next Tuesday.

The tournament is scheduled to continue even after three wins by either side, as Google wants to give more data to AlphaGo, which already surprised the Go community after shutting out European Go champion and second-dan Fan Hui 5-0 last October.

Lee, who has won 47 Go events including 18 international trophies, said last month that he would win the tourney with a score of 5-0 or 4-1 in his favor, but just a day before the matchup the Go player stepped back by saying that those scores may not happen.

His concerns turned into reality.

Playing in a room behind closed doors, Lee opened the match by taking the 3-4 point in the right top corner. Going second with white stones, it took 1 minute and 30 seconds for AlphaGo to make its first move. The machine eventually settled for the "flower point" or 4-4 point in the left top corner.

Lee, who brought a cup of coffee with him, then chose the 3-4 point in the right bottom, signaling that he will secure the right area of the board, while AlphaGo, whose moves were actually placed on the board by Google DeepMind programmer and amateur 6-dan Go player Aja Huang, went with another standard move by occupying the flower point in the left bottom to take the left side.

Lee, who is known for his aggressive and unconventional style of play, placed his third black stone near the AlphaGo's left-top flower point, but the engagement broke out in the top right corner and started moving into the center of the board.

Although there were moves from AlphaGo which South Korean commentators tagged as "strange," the machine didn't back off against Lee's effort to tweak the game. Looking frustrated, Lee at one point even left his seat to take a brief rest.

In the latter half of the match, Lee tried to defend his territories from the right side, but AlphaGo's cold, fatal moves narrowed down the winning chances for Lee and forced the South Korean to drop the game.

"As Lee faced an unfamiliar opponent, he looked nervous and made a lot of mistakes which is unusual for him," said Yoo Chang-hyeok, a ninth-dan player from South Korea.

Lee accepted his loss with a look of surprise.

"Speaking of today's match, failures in the opening layouts affected my game till the end," Lee said. "I want to give my respect to programmers who made AlphaGo."

The shocking defeat is now making him more nervous predictions of the five-round tournament.

"I would probably say that I could win if I hadn't seen AlphaGo's surprising and unexpected movement," he said. "But now, it's 50-50."

However, Lee said that there is no regret in accepting Google's offer to face AlphaGo and promised to give his all in the remaining four matches.

"It's shocking that I lost, but I did enjoy the game," he said. "I won't be shaken much by playing just one match."

On the otherside, Google programmers enjoyed a landmark victory for their program.

Demis Hassabis, the CEO of Google DeepMind, said he is expecting more from Lee and AlphaGo as the tournament continues.

"Of course, we are very pleased on AlphaGo's performance," he said. "I'm sure Lee will come with different and new strategies, and I think we will be still excited in next few matches."

David Silver, the team leader of the reinforcement learning research at Google DeepMind, said he expects that the match against Lee will further push the self-learning program.

"This is very historical moment," he said. "He really pushed AlphaGo to its limit."

South Korean professional Go players were said AlphaGo did made mistakes, but its playing style was still challenging.

"It didn't play like a human," said Kim Seong-ryong, a ninth-dan Go player who commentated the match. "It's really interesting that AlphaGo throughout the game keeps its cool even if it makes mistakes."

Go, known as "baduk" in Korea, originated in China more than 2,500 years ago. It involves two players alternately putting black and white stones on a checkerboard-like grid of 19 lines by 19 lines. The object is to claim larger territories than one's opponent by surrounding vacant areas of the board using one's own stones.

Go has been viewed as one of the hardest games for computers to master. Google said the possible number of board configurations of Go is larger than the number of atoms in the universe.

In this special showdown organized by Google and the Korea Baduk Association (KBA), the matches will be played under Chinese rules in which a player who places white stones receives 7.5 compensation points at the end of the match. Scoring calculations will also be based on Chinese rules, also known as "area scoring," in which a player's score is determined by the number of stones that the player has on the board and territories earned.

Each player has two hours per match with three lots of 60 seconds overtime counting after they have finished their allotted time.


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