In March 2016, in a hotel in Seoul, one of the world's best Go players sat down across from a computer. Lee Sedol was an 18-time world champion and the dominant player of his decade. His opponent was AlphaGo, a program built by a lab called DeepMind. Go is played on a 19 by 19 grid, and its possible positions outnumber the atoms in the observable universe. Experts had long assumed computers would fail at it for decades. Lee shared that confidence. Before the first stone was placed, he predicted he would win 5-0, or perhaps 4-1. The winner's prize was $1 million, which DeepMind pledged to give to charity. This is the story of a single move in that match, a move so strange that the champion, returning to the board, needed nearly 15 minutes to answer it. What did the machine see that he did not?
AlphaGo's Move 37
Lee Sedol was stunned by AlphaGo's Move 37, a fifth-line stone a human would play only once in 10,000.
A Champion Across the Board
Frequently Asked Questions
What was Move 37 in the AlphaGo match against Lee Sedol?
Move 37 was a stone AlphaGo played on the fifth line from the edge in the second game. Human players are taught to build on the third or fourth lines, so it looked like a mistake. It worked because it built influence in the center of the board.
Why did Move 37 surprise professional Go players?
Traditional theory treats fifth-line moves in the opening as an inefficient way to control space. AlphaGo's own policy network estimated that a human professional would play the move only one time in 10,000. The program had no fear of looking foolish, so it played what its calculations favored.
What was the Divine Move in the AlphaGo match?
The Divine Move was Lee Sedol's 78th move in the fourth game. He placed a wedge stone in the middle of AlphaGo's territory, which exposed a rare weakness in its calculations. AlphaGo resigned on move 180, and it was Lee's only win of the match.
Learn more
To continue the story, download the Chunks Microlearning app
Or get it straight away: