AlphaGo

In March 2016 DeepMind's AlphaGo beat Lee Sedol, one of the best Go players in the world, four games to one. Experts had expected this to be a decade away.

Context

Go is much harder for computers than chess because the board has far more possible positions, so brute-force search does not work. AlphaGo combined neural networks trained on human games and on self-play with a tree search to choose moves. The match took place in Seoul in March 2016 over five games, with a prize of one million dollars that DeepMind said it would give to charity.

Lee Sedol, a South Korean professional with many international titles, lost the first three games. He won the fourth with an unexpected move that the program seemingly missed, and AlphaGo won the fifth. In the second game, a move by the program was seen by commentators as surprising and creative.

Earlier, in October 2015, AlphaGo had beaten the European champion Fan Hui five games to nil, though he was a weaker player than Lee. The 2016 match is therefore the landmark result. Later versions of the software were far stronger, and the claim is about beating a top professional in a formal match, not about perfect play. DeepMind later published scientific papers describing the approach, and the match drew an estimated audience of tens of millions online, mostly in East Asia.