Subscribe
E-Paper
 |
| Korea's top go player Lee Se-dol leaves the match venue at the Four Seasons Hotel in Seoul, Thursday, after he was beaten a second time in the five-game series by Google's artificial intelligence program, AlphaGo. / Yonhap |
Go champion admits perfect loss in game 2
By Yoon Sung-won
Google's artificial intelligence (AI) program AlphaGo defeated world go champion Lee Se-dol again in their second game, Thursday, on the back of superior calculations that almost seemed to exceed human brains.
The game, which seemed to be a close one in the middle, drastically inclined to the AI's favor towards the end after its attack on Lee's territory in the center of the board. Lee, who chose relatively safer moves than in the previous round, tried to overturn the game even after he entered overtime counting but ended up resigning again after 211 moves.
After the match, Lee calmly admitted a perfect defeat.
"I was already surprised enough yesterday and I think I have nothing to say now," Lee said. "It was really a perfect defeat. I have never thought I was in the lead for a moment."
He also said AlphaGo played a flawless game.
"I could not find anything particularly strange (from AlphaGo). Yesterday I thought there might be something strange but today AlphaGo played a perfect game," Lee said.
Despite the two consecutive losses, Lee pledged to continue to do his utmost in the remaining games.
"Now that the score is 2-0 and (the victory) is not expected to be easy," he said. "But I will do my best to win at least one round.
"Reflecting on today's match, it is difficult to overturn the game after the middle of it. I will need to try to conclude the game before then to have a better chance of winning."
A match commentator Michael Redmond said, "Unlike in October, AlphaGo was very impressive, played innovatively and adventurously and led dangerously-looking moves to success."
Changing from the game, Lee played second with white stones, receiving 7.5 compensation points as AlphaGo did in the previous game.
Playing first with black stones, AlphaGo began the game at the 3-4 point of top left corner on the board instead of the flower point where it started in the previous round and in all five wins against European go champion Fan Hui in October last year.
In the 13th move, the AI abruptly moved from the bottom right corner and opened a new layout in the upper edge, which the commentator said was "surprising and unprecedented."
From this moment, AlphaGo continued with irregular and unpredictable positions while Lee proceeded with relatively stable moves to counteract the AI. In the meantime, he refrained from his unique aggressive and creative playing style and spent twice as much time as AlphaGo around an hour into the game, thinking deeply about his moves.
Two hours into the game, the AI started combat moves in the bottom left corner. But Lee broke through and succeeded in securing territory.
But when AlphaGo attacked Lee's group in the center, Lee turned to the AI's territory in the top right in exchange, which the commentators said was a bad move for Lee.
Leaving only one minute and 15 seconds, Lee made a key move to turn the tables. But the machine did not falter and drove Lee to overtime counting.
In this special competition between the world's top go player and the AI system developed by Google's London-based subsidiary DeepMind, each player was given two hours per match with three lots of 60 seconds overtime counting after they have finished the allowed time. Each 60-second lot is refreshed if it is not used up.
Lee strived for breakthrough but chose to resign four hours and 25 minutes into the game.
On Wednesday, AlphaGo beat Lee by a resignation in 186 moves after three-and-a-half hours. After AlphaGo's moves that seemed by commentators as "obvious and critical" mistakes in the latter half of that game, Lee tried to push further but failed to secure enough points to overcome the 7.5 compensation points and resigned.
Lee's defeat in the first game stunned the world as many have predicted his win under assumption that go still remains difficult area for machines to outperform humans as the ancient game requires both accurate calculations and intuition.
Last month, Lee showed confidence and said he would clinch a victory with a score of 5-0 or 4-1. But in a press conference on Tuesday, one day before the first game, he said AlphaGo's improved ability to narrow down options for next moves could be more intimidating than he originally expected.
The five-game match between Lee and the AI system will continue on Saturday and Sunday before wrapping up with the fifth-round on March 15. The Google DeepMind Challenge Match, which was organized by Google and the Korea Baduk Association, will continue even after three wins by either side as the event aims at providing the machine with more data.
