Roundup: AI beats human Go champion with 3-0 lead, dashes hopes for reversal

Artificial intelligence (AI) eventually surpassed humans in the ancient Chinese board game Go as Google's computer program AlphaGo won 3-0 over South Korean Lee Sedol, the world Go champion in the past 10 years, dashing the remaining hopes for reversal in a historic five-game match.

AlphaGo, a computer program developed by Google's London-based AI subsidiary DeepMind, won the third consecutive game in the match with Lee, one of the greatest Go players who has won 18 world championships for 21 years of his professional career.

The 33-year-old resigned as the computer program maintained an unassailable lead for more than four hours in the third match that was launched at 1 p.m. local time (0400GMT) on Saturday at Four Seasons hotel in Seoul.

Lee's defeat was a shock to the world as Go, known as weiqi in China and Baduk in South Korea, was viewed as the last game humans can dominate over machines. The match between Lee and AlphaGo was seen as a representative game between humans and AI.

Dashing hopes for reversal, AlphaGo took a 3-0 lead in the five-game match that kicked off on Wednesday and would last until next Tuesday. The fourth game will be held on Sunday in Seoul.

The AlphaGo's developer will get 1 million U.S. dollars in prize, which will be donated to charities, regardless of whether Lee wins the two remaining games or not.

One commentator said AlphaGo proved to be strong based on the latest three matches, while another commentator said AI turned out to have surpassed humans in the Chinese board game.

Go originated from China more than 2,500 years ago. It involves two players who take turns putting markers on a grid-shaped board to gain more areas on it. One can occupy the markers of the opponent by surrounding the pieces of the other.

"I had originally expected Lee's 5-0 victory, but all changed," Kwon Kap-Yong, Lee's childhood teacher, said while the game was going on. Calling AlphaGo as a "great machine," Kwon said that focus would be moved on how to win at least one game against the AI.

"In a sense, it could be a victory for Lee as this match drew great attention to Baduk from all around the world," Kwon said. "I hope Lee would enjoy the rest of the games and show exciting games to the last minute," he added.

Lee also has lowered confidence in his victory. During the Feb. 22 press conference, he said AlphaGo would by no means win any game, but he got less confident about a 5-0 victory on Tuesday, a day before the opening match, as he listened to how AlphaGo operates to play and predict.

The South Korean Go player expected a 50-50 victory after the first defeat on Wednesday. Lee reduced his confidence further on Thursday, saying that he would do his best to win at least one game.

AlphaGo boasts of a deep learning capability to learn for itself and discover new strategies by playing games against itself and adjusting neural networks based on a trial-and-error process known as reinforcement learning.

Playing black, Lee played aggressively throughout the game as he allegedly analyzed AlphaGo's moves all night after the second loss along with his South Korean professional Go players and decided to play his own style.

Following his second defeat, Lee told a press conference on Thursday that his winning possibility would likely rise if he gets dominant in the first half.

AlphaGo, however, defended his aggression almost completely, maintaining its large groups of white stones in the lowest end of the board into which Lee invaded as the last winning move.

After exchanging 176 moves with AlphaGo that had dominated the game all the way, Lee accepted his loss as the computer program took an unassailable lead. Enditem