Achievement Unlocked: Google AlphaGo A.I. Wins Go Series, 4-1 | KERA News

Achievement Unlocked: Google AlphaGo A.I. Wins Go Series, 4-1

Mar 15, 2016
Originally published on March 17, 2016 6:50 am

The five-game clash pitting man against machine is over, with Google's artificial intelligence program winning the series.The program — called AlphaGo — took four of five games against Korean Lee Sedol, an 18-time world champion of the board game Go.

"I feel a little regrettable because I believe that there is more that a human being could have shown in a match against artificial intelligence," Lee said following the final match.

The matchup had all the mood and atmosphere of a prize fight. A ritzy hotel venue. Enough cameras flashing to temporarily blind you. And play-by-play announcers who analyzed every move.

Of course, in this case, the analysis consisted of discussing strategy while duplicating a board of black and white stones for hours on end. Go is an ancient East Asian board game played with stones on a 19 by 19 board that has long enjoyed popularity in Asia, but has, in the past week of publicity, been introduced to many new corners of the world.

"This is definitely the moment when we realized computers could beat us at this game," says Andrew Okun, the president of the American Go Association.

Google subsidiary DeepMind's artificial intelligence program AlphaGo has been in the works for the past two years. It relied on machine learning and neural networks to study a database of 100,000 human matches, then played against itself millions of times to reprogram itself and improve.

"Go has been an outstanding grand challenge of AI research for at least the last twenty years," says Demis Hassabis, who heads DeepMind. A computer had already beaten a human champion at chess decades ago. But Go has more possible moves than there are atoms in the universe, so it was considered a far more intuitive game for a machine to try and beat.

"So it's really kind of the Mount Everest of perfect information and games," Hassabis said. For an adversary, "We wanted to pick somebody who, in twenty to thirty years time, would still be remembered as a legend of the game."

The challenge riveted South Korea. At Seoul's central train station, we found hundreds huddled around a television to watch the matches. One of them was Lee Song-hee, who has no relation to the player.

"I'm just burning up inside," she said as she watched the final match. "It's like watching my own son play the game. It's like watching my family."

With all eyes on him, Lee lost the first three games to AlphaGo. The losses stunned him, he said. Lee then fought back to defeat the computer once, on Sunday. And after battling it out for every point, Lee ultimately lost the final game, deep in overtime.

So far, Okun says the reaction to this moment is mixed. Not unlike a lot of other reactions to tech innovation.

"Fifty percent sadness and trepidation and 50 percent where can I get my copy? I want this software," Okun says.

AlphaGo's makers don't plan to sell it commercially just yet. But already, it's challenging 2,500 years of traditional Go strategies and thought.

"All the traditional or classical beliefs we have had about Go so far, I have come to question them a little bit based on my experience with AlphaGo," Lee said. "So I think I have more studying to do down the road."

In its historic win over a man, the machine has something to teach — new perspectives on an ancient game.

Haeryun Kang contributed reporting to this story.

Copyright 2017 NPR. To see more, visit http://www.npr.org/.

ARI SHAPIRO, HOST:

A clash of man against machine is over, and the machine won. Google's artificial intelligence program, AlphaGo, took four out of five games in a match of Go, an ancient Chinese board game. The loser was the human reigning world champion. NPR's Elise Hu reports on this computing milestone.

ELISE HU, BYLINE: The matchup had the mood and atmosphere of a prizefight - a ritzy hotel venue, enough cameras flashing to temporarily blind you, and play-by-play announcers.

(SOUNDBITE OF ARCHIVED RECORDING)

UNIDENTIFIED MAN: Black could play here and force white to answer, making a one-point difference. Keeping the tempo, keeping sente.

HU: The face-off was no actual fight, but instead a popular ancient East Asian board game. Go is played with black and white stones, which each player uses to try and gain territory. And the big draw in Seoul was no mortal. It was the artificial intelligence program AlphaGo, which relies on machine learning and neural networks. It studied a database of 100,000 human matches, played against itself millions of times to reprogram itself and improve.

(SOUNDBITE OF ARCHIVED RECORDING)

DEMIS HASSABIS: Go has been an outstanding grand challenge for AI research for at least the last 20 years.

HU: Demis Hassabis is the head of Google's AI subsidiary, DeepMind, which created AlphaGo. The computer had already beaten a human champion at chess decades ago.

(SOUNDBITE OF ARCHIVED RECORDING)

HASSABIS: Really, the only game left above chess is Go. So it's the kind of Mount Everest of perfect information and games.

HU: To really test AlphaGo, the Google DeepMind team needed a worthy adversary.

(SOUNDBITE OF ARCHIVED RECORDING)

HASSABIS: So we wanted to pick somebody who, in 20, 30 years' time, would still be remembered as a legend of the game.

HU: They settled on an 18-time world champion - the Korean Lee Sedol, a baby-faced 33-year-old who turned pro when he was 12. The challenge riveted South Korea. At Seoul Central train station, we found hundreds huddled around a television to watch the matches. One of them was Lee Song-hee, no relation.

LEE SONG-HEE: (Through interpreter) I'm just burning up inside. It's like watching my own son play the game. It's like watching my family.

HU: With all eyes on him, Lee lost the first three games to AlphaGo. The losses stunned him, he said. Lee then fought back to defeat the computer once on Sunday. And after battling it out for every point, Lee ultimately lost the final game deep in over time. He spoke through a translator after the match.

LEE SEDOL: (Through interpreter) I feel little bit regrettable because I believe that there is more that I maybe could have shown in a match against artificial intelligence.

ANDREW OKUN: This is definitely the moment when we realized computers could beat us at this game.

HU: Andrew Okun is the president of the American Go Association. He says the reaction to this moment is mixed, not unlike a lot of other reactions to technology innovation.

OKUN: Fifty percent sadness and trepidation and 50 percent where can I get my copy? I want this software.

HU: AlphaGo's makers don't plan to sell it commercially just yet. But already, it's challenging a couple thousand years of traditional Go strategies. In its historic win over a man, the machine has something to teach - new perspectives on an ancient game. Elise Hu, NPR News, Seoul. Transcript provided by NPR, Copyright NPR.