Google’s Go-Playing AI Wins Second Match Against Pro Player Lee Sedol

Google DeepMind's AlphaGo has won its second Go game against professional player Lee Sedol. The victory puts the AI at a convincing lead in the $1 million human-computer tournament, which will consist of five games in total.

"Yesterday I was surprised, but today I am more than that; I am quite speechless," Sedol said in a videoed post-match press conference. He said that it was a clear loss on his part, and that he never felt like he had the upper hand in the game.

"Today I really feel that AlphaGo played a near-perfect game; there was not a moment where I thought AlphaGo's moves were unreasonable," he said.

The game began at 1pm Korean Standard Time on March 10, or 11pm ET on March 9. AlphaGo and Sedol both used up their full two-hours of time for the game, which ended in "byo-yomi" periods, a type of overtime.

Demis Hassabis, founder and CEO of DeepMind, said the game was "unbelievably exciting and incredibly tense." "I think it's testament to Lee Sedol's incredible skills, and it was a very exciting match, with some very beautiful and quite surprising moves according to the commentators," he added.

#AlphaGo wins match 2, to take a 2-0 lead!! Hard for us to believe. AlphaGo played some beautiful creative moves in this game. Mega-tense…
— Demis Hassabis (@demishassabis) March 10, 2016

Though AlphaGo won the first match earlier this week, the second win still comes as something of a surprise. Many have been expecting Sedol, a South Korean 9-dan player (the highest Go ranking), to take the tournament. Speaking at the start of the second game, commentator Michael Redmond—himself a 9-dan player—suggested that Sedol might have underestimated his AI opponent the first time around.

By all accounts, the second game was a tough match. Redmond noted that AlphaGo's opening seemed stronger than in previous games, while Korean commentator Yoo Changhyuk suggested that Sedol made some mistakes in his byo-yomi periods.

Go is a notoriously strategic game, which makes it a significant challenge for AI—beyond, say, Chess, which IBM's Deep Blue conquered 20 years ago. DeepMind describes its methods as combining "Monte-Carlo tree search with deep neural networks that have been trained by supervised learning, from human expert games, and by reinforcement learning from games of self-play."

With three matches left, Sedol needs to take the next one if he's going to have a chance of winning the tournament and claiming a victory for Team Human. The third match will take place in Seoul on March 12 at 1pm local time, which is 11pm on March 11 on the East Coast, and will be livestreamed on the DeepMind YouTube channel.