Google’s Artificial Intelligence Trounces World Go Champ in 1st Match

An artificial intelligence system developed by Google has trounced a world champion  Go player in the first of five planned matches.

The program, called AlphaGo, is currently facing off with former  world champion Lee Sedol in a historic five-day matchup in South Korea. Both players are considered the best of their kind: AlphaGo recently bested the European Go champion, Fan Hui, while Sedol is considered the best  living player of the strategy game, which many consider the most complicated game in the world.

So far, things are not looking good for the human.

In the first match, AlphaGo roundly defeated Sedol, who was visibly shocked by the outcome.

“I didn’t know AlphaGo would play such a perfect game,” Sedol said  in a news conference Wednesday (March 9).

Other professional players were also blown away by AlphaGo’s performance.

“When Google said the odds were fifty-fifty, it seems they  weren’t joking,” Cho Hanseung, a top professional Go player, told GoGameGuru, a Go enthusiast website. “I still can’t believe its performance, even though I just saw it with my own eyes.”

AI Grand Challenge

lee-sedol-plays-alphagoThe defeat of Hui in January was seen as a triumph of artificial intelligence. Go has often been considered one of the grand challenges in the field, owing to the incredible  number of potential moves ­ — each turn requires deciding among 200 potential moves, compared to a relatively modest 20 in chess.

“It’s probably the most complex game devised by humans,” AlphaGo co-inventor Demis Hassabis, a computer scientist at Google DeepMind in London, said Jan. 26 in a news conference. “It has 10 to the power 170 possible board positions, which is greater than the  number of atoms in the universe.”

To break this problem into more manageable bites, the team coded AlphaGo (which was developed by Google’s DeepMind unit) to use separate “learning” networks. A so-called value network looks at the state of the board to decide who is winning, while a policy network  evaluates potential moves. The program trains itself by watching and then playing millions and millions of games.

While AlphaGo may, in the end, rout Sedol, it’s not completely fair to say the competition was an even fight. AlphaGo practiced for his matchup against Sedol by  watching many of the reigning champion’s matches, while Sedol had much less exposure to AlphaGo’s playing style. In addition, Sedol seemed to be off his game and made a few errors, but still managed to retake the lead at several points during the game, according to GoGameGuru’s commentary of the match.

A few years ago, or even a few months ago, this result would have been inconceivable. Up until now, the best artificial intelligence Go players were only as good  as amateur players, and most experts in the field believed it would be years before the best AI players could compete against the best humans.