Virtuoso Go-playing AI AlphaGo has secured victory against 18-time world champion Lee Se-dol by winning the third straight game of a five-game match in Seoul. AlphaGo is now 3-0 up in the series, but there’s no mercy rule here — the remaining games on Sunday and Tuesday will still be played out. AlphaGo is a program developed by DeepMind, a British AI company acquired by Google two years ago.
I’ve been playing go off and on for nearly 40 years, so I understand the implications of this. While people have gotten used to the fact that computers and apps can beat the best chess players, they generally have no idea of the complexity of go.
Go is in incredibly complicated game. Because there are so many points on the board (361), there are this many legal positions for a game:
(This means positions where stones are allowed to play according to the rules. And I’ve added line breaks so the number doesn’t stretch out off the side of the page.)
What’s interesting about AlphaGo’s performance is not just that it won, but that it played some “creative” moves. In the second game, the AI played a move that all those watching and commenting on the game found to be brilliant.
Go is full of patterns and moves that are considered to be correct, and others that aren’t. Humans generally limit themselves in the moves they play, because of the weight of experience and tradition. But an AI won’t consider what the greats of go played, they’ll play the moves that are the most effective. They’ll eventually introduce new moves that humans haven’t considered, or play moves that humans considered to be incorrect (not wrong, just not optimal).
Take, for example, the “new fuseki” movement in go. (Fuskeki means opening.) In the 1930s, go players, notably including Go Seigen and Kitani Minoru, started playing radically different openings from what was traditional, changing the nature of the game. They experimented with different ways of playing, discarded what didn’t work, and developed a new range of opening strategies. It’s only because they questioned what was traditional that they were able to change the game so much.
An AI does the same thing. It “knows” a corpus of tens of thousands of games, but it can still be free of the limitations that humans have, and try out any new move that seems more effective. Over time, this AI, and others, will lead to changes in the way go is played.
The importance of what AlphaGo did isn’t limited to just go, of course. It shows that AI has made great strides in recent years, and presages many more to come.
Update: Lee Sedol won the fourth game.