Lee Se-dol, one of Korea’s top professional Go players, is due to play a five-game match against a computer program developed by a Google affiliate starting March 9.
The match has drawn keen attention from Go players and technology experts around the world because Go has thus far been viewed as an unsolved “grand challenge” for artificial intelligence.
Go, which originated in China more than 2,500 years ago, is much more complicated than chess. The classic game is played on a board with a 19×19 grid of lines, much larger than a chess board with its 8×8 grid of squares.
Due to the larger board size, the average number of moves that each player makes in a Go game comes to about 150-200, several times more than the average 37 in a chess game.
What makes Go a daunting challenge for computer programmers is that for each move that one player makes, there are a large number of possible moves the other player can make in response.
Because a computer program must calculate every possible move on each player’s turn, its ability to choose the best plays is sharply reduced when there are a large number of possible moves.
For this reason, prior to 2015, the best Go program only managed to reach the advanced amateur level.
But things began to change with the emergence of AlphaGo, a program developed by Google DeepMind, a Google affiliate in Britain.
In October 2015, AlphaGo surprised many Go players around the world by winning 5-0 in a match against the reigning European Go champion. The victory made AlphaGo the first computer program to ever beat a professional Go player.
Seeking to carry the momentum, Google decided to challenge the world’s best professional Go player. It chose Korea’s Lee Se-dol, the world’s best player over the last decade.
AlphaGo is strong because its machine learning system allows it to improve itself just by watching and playing Go games. The program is based on two neural networks — the “policy network” that suggests intelligent moves to play and the “value network” that evaluates board positions the suggested moves lead to.
Despite its recent triumph against the European Go champion, however, AlphaGo is still believed to be no match for the world-class, top-notch professional players. Go is a game that requires not just a high level of computing power, but such abilities as intuition and feel that computer programs lack.
Lee himself expressed confidence in his victory, saying that he would be able to win 5-0 this time. He said he would welcome it if AlphaGo proposed a return match after losing this one. But he said he might not win a third match with AlphaGo in light of its powerful learning ability.
Lee also acknowledged that it would not be long before computer Go programs beat top professional players.
The coming Go match is meaningful, as it signals that the day will come soon when artificial intelligence is applied to important real-world problems.
Google hopes that its machine learning technology could one day be extended to addressing some of society’s toughest problems, from climate modeling to complex disease analysis.
Lee’s match with AlphaGo should inspire Korea’s efforts to develop artificial intelligence, one of the key emerging technologies to drive the “Fourth Industrial Revolution.”
