Go to Contents Go to Navigation

(LEAD) S. Korean Go player collects first win against Google AI

All News 18:16 March 13, 2016

(ATTN: ADDS details)
By Joo Kyung-don

SEOUL, March 13 (Yonhap) -- It feels like victory for all humans.

South Korean Go player Lee Se-dol finally collected his first win against Google's artificial intelligence (AI) program AlphaGo on Sunday, rebounding from his three consecutive defeats earlier.

In their fourth showdown at the Four Seasons Hotel in Seoul, Lee got the win by resignation over AlphaGo, the AI designed by Google's London-based firm DeepMind, in 180 moves after playing nearly 4 hours and 44 minutes. The score at the special five-round Go tournament named the Google DeepMind Challenge Match is now 3-1.

Lee played conservatively in the opening and started to ruin AlphaGo's attempt to secure territories in the center. While the 33-year-old South Korean was pushed by the time, the AI made several "suicidal" moves in the final section of the match that helped Lee control the game to the end.

Despite the victory, Lee can't win the prize money of US$1 million. AlphaGo had already clinched a series win on Saturday and the money will be donated to UNICEF and other charities.

Lee will have his final match against the Google AI on Tuesday at the same venue.

Lee is a ninth-dan player who went pro at the age of 12 and has won 18 international trophies. AlphaGo first made headlines last October after beating European champion Fan Hui, a second-dan player, in five consecutive matches.

The opening sequence of the fourth round was same as the second match played on Wednesday, and Lee said he believed he would take an early lead. Knowing that AlphaGo's opening moves were same as before, Lee smiled before tweaking the game in the 12th move with his white stone.

After engaging briefly in the lower side of the board, where Lee got an edge in claiming territories, the two players fought hard in left center. AlphaGo started pushing up to connect with its black stones at the top. Lee, who is known for his aggressive and unconventional style of play, then invaded the AI's territory in the upper right corner.

In the mid-phase of the game, the two scuffled in right center and their sequence also moved up. Two hours past the start, Lee, who had use the 60-second overtime slot in the last two matches, had only 40 minutes on his clock, while AlphaGo kept 1 hour, 20 minutes.

The two sides eventually moved to the center of the board, with Lee making some decisive moves to exchange areas to build up territories, while trying to ruin AlphaGo's effort to claim territories in the center.

Closing in on three hours into the game, Lee entered overtime use while the human-like algorithm had more than an hour to spare. But starting from the 87th move, AlphaGo made a couple strange placements in the center right that were viewed as "suicidal."

South Korean commentators even said that if it was not a software bug, those moves would have even surprised Aja Huang, a Google DeepMind programmer and amateur 6-dan Go player who is placing stones for the AI.

"AlphaGo's move just helped Lee's attempt to live," said Song Tae-gon, a ninth-dan player who commentated the match, after watching the unusual 87th move from the AI. "It's a move that no human Go player would ever place."

Suddenly, the match was turning in Lee's favor. Playing with the white stone, the South Korean already had "komi," or 7.5 compensation points, under the Chinese rules. In this tournament, scoring calculations are also based on the Chinese rules, known as "area scoring," in which a player's score is determined by the number of stones the player has on the board and the territories earned.

Though the game turned in Lee's favor, he was still pushed by the time with only one overtime slot remaining. But he did earn a couple of minutes in the final phase of the game by doing something AlphaGo cannot: going to the restroom.

AlphaGo returned to its usual playing style later, but Lee kept his composure and wrapped up the game without a critical mistake.

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 for Go is larger than the number of atoms in the universe.


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