Google's AI Beats Go Player Lee Sedol Again to Win Tournament 4-1
AlphaGo took its fourth victory in the final of the 5-game tournament.
It's all over: The historic Go tournament between Google DeepMind's AlphaGo AI and professional human player Lee Sedol came to a close on March 15 with the end of the fifth match between the unlikely adversaries. The robot won.
AlphaGo and Sedol sat down for their final game (figuratively speaking on the part of the AI program) at 1pm Korean Standard Time. Google reports that the match was close right to the end, when Sedol resigned after a long end game and two byo-yomi overtime periods. The final score after five games was 4-1 to AlphaGo, which won all but the fourth game.
In a press conference after the match, DeepMind founder Demis Hassabis said the last game was "the most exciting and stressful one for us."
He pointed to a move AlphaGo made early on that commentators thought was a mistake. It's unclear whether it was a mistake or an unexpected and creative move, but either way AlphaGo brought itself back into the match to eventually win.
"We're not very strong Go players so in the control room we were just relying on AlphaGo's predictions, but certainly the professional commentators seemed to think that some of the sequences in the early part of the game—the "stone killer" tesuji—was a mistake by AlphaGo," said Hassabis. "But in the end AlphaGo was able to get itself back in the match for an incredibly close and tense finish."
"When it comes to Go skills, I don't think that AlphaGo is superior, but when it comes to psychological factors, yes I think that AlphaGo is superior"
Sedol looked resigned after the game, and said he was sorry the tournament had come to an end. In response to questions, he said that he didn't necessarily believe AlphaGo was a better Go player than him and that there was still more a human could have done.
The professional player said that playing against an AI program was different, and he felt he struggled to adapt to this at the beginning of the tournament. He noted that AlphaGo could not feel intimidated or suffer a lapse in concentration. "When it comes to Go skills, I don't think that AlphaGo is superior, but when it comes to psychological factors, yes I think that AlphaGo is superior," he said.
The tournament had a $1 million prize pot, which Google DeepMind is donating to charity.
AlphaGo's victory will no doubt excite AI enthusiasts, all the more so because Sedol had predicted before the tournament that he'd win convincingly.