Go to Contents Go to Navigation

(LEAD)(News Focus) S. Korean Go player on challenger's side against Google AI

All News 19:28 March 11, 2016

(ATTN: CORRECTS the KBA's request to Google in 10th para, ADDS Google's comment)
By Joo Kyung-don

SEOUL, March 11 (Yonhap) -- At first glance, South Korean Go player Lee Se-dol was the favorite as he accepted the challenge from Google's artificial intelligence (AI) program AlphaGo. But after losing his first two matches this week, the tables have turned.

Lee, an 18-time world champion, is down 2-0 at the five-round Go tournament in Seoul after throwing his stones on Wednesday and Thursday against AlphaGo, the AI developed by Google's London-based firm DeepMind.

Lee, a ninth-dan player who went pro at the age of 12, can technically still win if he takes all of the remaining three matches that are slated for Saturday, Sunday and next Tuesday at the Four Seasons Hotel in Seoul. But it seems AlphaGo has already made too much of an impact, already leaving people "surprised" and "speechless."

After all, Lee is human, too.

If his playing level hasn't gone down, Lee's confidence level for sure has. For emotionless AlphaGo, that's no problem.

The 33-year-old South Korean last month said that he will win the historic five-match man-versus-computer showdown 5-0 or 4-1 in his favor. A day before the first match, Lee backed off by saying he isn't so sure his previous predicted score would happen, and after losing the opening contest Wednesday, he went for 50-50.

Taking another punch from the human-like algorithm on Thursday, Lee said he will now try to win at least one.

As the expectations for a human victory turned sour, the South Korean Go community is saying maybe the matchup was unfavorable to Lee from the start.

"AlphaGo has been hiding itself, but it has all the game records from Lee Se-dol," Korea Baduk Association (KBA) secretary-general Yang Jae-ho, who is also a ninth-dan Go player, told Yonhap News Agency Friday. "Lee is facing an opponent who knows A to Z about him. This isn't fair play."

Go, known as "baduk" in Korea, is a territory-winning board game that originated from China more than 2,500 years ago. Reading the opponent's style of play and using it against it to win territories are considered critical parts in leading the game. The KBA tried to acquire the playing records for AlphaGo from Google, but the request was never officially made.

Having only seen an inferior version of AlphaGo playing against European champion Fan Hui last October, during which the AI had a 5-0 victory, Lee's information on the AlphaGo was already old. Facing improved algorithms, the South Korean Go master said it will be impossible to read AlphaGo's "fighting spirit" and its style of play.

"It will feel like I'm playing the game alone," he said on Tuesday before facing AlphaGo.

Following the criticism that the match was played under unfair conditions to Lee, Google said Friday that it has worked closely with the KBA to prepare the match and tried to provide conditions that can satisfy the national governing body's request."

To find out AlphaGo's character, Lee, who is known for his aggressive and unconventional style of plays, tried to tweak the game and lure AlphaGo into his traps in their first match. But the cold-blooded AlphaGo didn't bother. It actually countered back with a more unconventional move that hammered Lee and eventually forced him to drop the game in just 186 moves.

In the second match, Lee went opposite of his usual style of play, making his sequences calmly which ninth-dan player Yoo Chang-hyuk, who was commentating the match, described as "playing like a different person." Lee's strategy pushed AlphaGo into overtime use, but couldn't beat the program's cold calculation of territories in the end. Lee resigned from the game in 211 moves.

"AlphaGo seems to have a characteristic that it places stones just enough to collect a win," said Lee Ha-jin, a secretary general at International Go Federation and a three-dan player. "It doesn't push the game hard even though it is winning."

If understanding the machine's style of play is a difficult task, there is another disadvantage for Lee: He gets tired, while AlphaGo doesn't.

"Lee's condition isn't good in this week," said Yang from the KBA. "Before this historic event, Lee had a busy schedule as he conducted many media interviews and played tough matches at the Nongshim Cup in China."

But what might make Lee, or other humans, really afraid is that no one knows whether these matches were AlphaGo's best performance. The human-like algorithm keeps evolving, as long as its power is on.

"The reason why we have this match is that we need to see any weakness we don't know about," Demis Hassabis, the CEO of Google's London-based AI company DeepMind, said on Tuesday. "We have not experienced the ceiling level of AlphaGo and how far it can improve."


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