Tech by VICE

In Unexpected Defeat, Google's AI Beat World-Class Go Champion Lee Sedol

And there’s already a second match coming up in a matter of hours. Talk about pressure!

by Rollin Bishop
Mar 9 2016, 9:30pm

Photo: Jaro Larnos/Flickr

In a shocking turn of events, Google DeepMind's AlphaGo—a computer program—has defeated South Korean Go champion Lee Sedol in the first of five total matches.

Lee had previously been heavily favored to win, and possibly even sweep the series, but many were hedging their bets. Now? It's even more unclear.

The first match began at 1 PM local time in Seoul, South Korea on March 9th, meaning this all went down at 11 PM ET on March 8 in North America. Another match is set for 1 PM March 10 (or 11 PM ET March 9), then a break, then two days of matches, a break, and the final match. Now that he's lost the first, Lee will almost certainly be taking the last four even more seriously.

Which isn't to say that Lee's play in the first match was bad, or that he didn't take it seriously. The Verge quotes Lee as saying after the match that he was "surprised" at the manner in which AlphaGo played the game, and that he hadn't expected it "would play the game in such a perfect manner." In February, Lee had predicted the games would be 5-0 or 4-1 in his favor.

That comes after DeepMind announced in January that AlphaGo had defeated European Go champion Fan Hui 5-0 back in October 2015. The rank gap between Fan and Lee is significant, however—Fan is a 2-dan professional Go player while Lee holds a 9-dan rank—so AlphaGo was not expected to win against Lee with only five months of preparation after playing Fan.

Of the two hours allotted to each AlphaGo and Lee, the program had used up nearly all its time with five minutes and 30 seconds remaining, while Lee had 28 minutes and 28 seconds on the clock when he ultimately resigned. It seems clear from the video, and from the professional commentary on the stream, that Lee is aware of what he considers mistakes made by himself and AlphaGo.

"They were neck-and-neck for its entirety, in a game filled with complex fighting."

"They were neck-and-neck for its entirety, in a game filled with complex fighting," reads an update to the Google Asia Pacific Blog posted after the four-hour stream. "Lee Sedol made very aggressive moves but AlphaGo did not back down from the fights." In other words, the match was too close to call for the majority of it.

Given the quality of play in the first match, it's not like Lee was outclassed entirely or anything, so it's still possible for humanity to come back and win. For Lee, there's both a $1 million prize and his reputation on the line, so he certainly has the motivation. For Google DeepMind and AlphaGo, a loss or a win would be equally useful. If Lee beats the machine, it still gives the researchers a new trove of information to work with. If AlphaGo wins, the team will have accomplished what had up until recently been considered impossible or at least improbable.

With the added pressure of being behind, it'll be interesting to watch Lee take on AlphaGo again.

Artificial Intelligence
motherboard show
lee sedol
fan hui