The match between AlphaGo and Lee Sedol was a five-game Go series held in Seoul, South Korea, from March 9 to March 15, 2016. It pitted DeepMind's AlphaGo program against Lee Sedol, an 18-time world champion and one of the strongest Go players of his era. AlphaGo won four games to one, a result that surprised much of the Go community, which had generally expected human players to hold their advantage in the game for years longer.
Background
Go had long been treated as a harder problem for computers than chess, which Deep Blue had already conquered in 1997, owing to its vast branching factor and its reliance on pattern recognition and intuition that resisted brute-force search. AlphaGo combined deep neural networks trained through Supervised learning on a large corpus of human games with a policy and value network refined by Reinforcement learning through self-play, coupled with Monte Carlo tree search to select moves. The project was led by David Silver at DeepMind, whose founder Demis Hassabis had a longstanding interest in games as a proving ground for general intelligence.
The match
AlphaGo won the first three games, securing the match before the final two were played. Game two produced a widely discussed moment, move 37, an unconventional shoulder hit that commentators initially thought was a mistake but that later analysis judged highly creative, a move a human professional was unlikely to have played. Lee Sedol won game four with a sequence starting at move 78, sometimes called the "hand of God" move, briefly exploiting a weakness in AlphaGo's evaluation of an unusual position. AlphaGo won game five to close out the match 4-1.
Aftermath and significance
The result was widely covered outside specialist Go and AI circles and is frequently cited as a milestone in Deep learning research, comparable in cultural weight to Deep Blue's 1997 win over Garry Kasparov in chess. Lee Sedol described the loss as difficult personally, later saying that even becoming the top human player would no longer mean being unbeatable. He retired from professional play in 2019, citing the rise of AI as a factor, remarking that there was now "an entity that cannot be defeated."
DeepMind followed the match with successors that generalized the approach: a version trained purely through self-play without human game data, and AlphaZero, which extended the method to chess and shogi from scratch. The match is often referenced in discussions of Artificial general intelligence and of how quickly narrow superhuman performance can arrive once the right combination of architecture, compute, and training method is found, a theme that recurred in later debates over Scaling laws and Emergent abilities in large models.