AlphaGo Beats a Go Champion Four Games to One

Between March 9 and March 15, 2016, professional Go player Lee Sedol faced AlphaGo, a computer program developed by Google DeepMind, in a five-game match at the Four Seasons Hotel in Seoul, South Korea. Lee was a 9-dan master with eighteen international titles and a reputation for creative play. AlphaGo represented a new approach to game-playing artificial intelligence, relying on deep neural networks and Monte Carlo tree search rather than hand-crafted evaluation rules. After defeating European champion Fan Hui in late 2015, the program entered the Seoul contest with observers widely divided over whether an algorithm could challenge a player at the absolute summit of the game.

Go had long stood as a benchmark that resisted computer automation. The vast branching factor of the nineteen-by-nineteen board made the exhaustive search methods that had powered chess engines ineffective. When IBM computer Deep Blue defeated Garry Kasparov in 1997, chess became a conquered domain for machines, yet a computer victory over a professional Go player had been widely thought to be at least a decade away. AlphaGo countered the complexity of the game by bootstrapping its neural networks on thirty million moves from human games and then playing millions of matches against copies of itself through reinforcement learning.