A Computer Beats One of the World's Best Players at Go

Over five games in Seoul, a program built by Google's DeepMind lab defeated Lee Sedol, a champion of the ancient board game Go, four games to one. Most experts had guessed a computer wouldn't manage this for another decade.

The final game ended this afternoon in Seoul. AlphaGo, a computer program built by DeepMind — a London research lab that Google bought two years ago — beat Lee Sedol, one of the strongest Go players alive, to finish the match four games to one.

Go is a board game, older than chess, played with black and white stones on a grid. The rules fit on an index card, but the number of possible positions is larger than the number of atoms in the universe, so a computer can't simply calculate every move the way chess programs do. Top players talk about intuition — a position just "looks right." That is why Go was supposed to be the game computers couldn't crack. Before this week, most researchers put a win like this ten years away.

AlphaGo learned differently from the chess programs that came before it. Instead of being given rules by its programmers, it studied millions of moves from human games, then played millions more games against copies of itself, keeping whatever worked. Nobody told it how to play well. It found out.

Lee won the fourth game, on Sunday, with a move commentators called brilliant, and the crowd in the hotel ballroom cheered as if their country had scored a goal. In the second game, the machine played a move — the 37th of the game — that professional players first assumed was a mistake. It wasn't. Several have said they would never have thought of it.

Two hundred million people watched some part of the match, most of them in China, Korea, and Japan, where Go is followed the way baseball is in America. Lee said afterward that he was surprised, that he had never once felt in control, and that he would need to think about what it meant. So will the rest of us.