AlphaGo is a computer program developed by DeepMind to play the board game Go, a game long considered a major unsolved challenge for Artificial intelligence because of its vast search space, far larger than chess, which made brute-force approaches like those used in Deep Blue computationally infeasible. AlphaGo combined deep neural networks with Monte Carlo tree search, using networks trained on both human expert games and self-play Reinforcement learning to evaluate board positions and select moves.
The Lee Sedol match
AlphaGo became internationally famous through its 2016 match against Lee Sedol, one of the strongest professional Go players in the world, held in Seoul, South Korea, from March 9 to March 15, 2016. AlphaGo won the five-game match 4 games to 1, an outcome that surprised much of the Go and AI communities, who had generally expected human mastery of Go to hold out against machines for at least another decade given the game's complexity. The match was broadcast globally and watched by an estimated 200 million people, making it one of the most visible public demonstrations of AI capability to that point, arguably the moment deep learning entered mainstream public awareness in the way ChatGPT's launch would seven years later.
Move 37
Game two of the match produced a moment that became emblematic of AlphaGo's novelty: on move 37, AlphaGo played a move on the fifth line of the board that professional commentators and Lee Sedol himself initially considered highly unusual, even a likely mistake, since it contradicted centuries of accumulated human Go strategy. The move later proved pivotal to AlphaGo's win of that game, and it was widely cited afterward as evidence that the system had derived genuinely novel strategic insight rather than simply imitating or interpolating between human play, a claim later analyses of the system's internal move-probability estimates broadly supported. Move 37 became a frequently referenced example, in both AI research and popular writing, of a machine learning system producing creative, non-human-like problem-solving.
Predecessor to AlphaZero
AlphaGo's training initially relied on a large dataset of recorded human expert games to bootstrap its policy network before self-play reinforcement learning refined it further. DeepMind subsequently developed AlphaGo Zero, announced in 2017, which learned Go purely through self-play from random initialization, without using any human game data, and which surpassed the original AlphaGo's playing strength. This line of research led directly to AlphaZero, a generalization of the same self-play approach to chess and shogi in addition to Go, released later in 2017.
Reception and legacy
AlphaGo's victory over Lee Sedol is widely regarded as a landmark event in the history of AI, comparable in cultural significance to Deep Blue's 1997 defeat of Garry Kasparov in chess but distinguished by AlphaGo's reliance on learned pattern recognition rather than exhaustive search, reflecting the shift in AI research from symbolic and brute-force methods toward deep-learning-based approaches. Lee Sedol retired from professional Go competition in 2019, citing in part the sense that AI had become unbeatable at the game, remarking that even as the top player he could not defeat "an entity that cannot be defeated." DeepMind's AlphaGo research program is frequently cited alongside AlphaFold as evidence of DeepMind's broader strategy of applying reinforcement learning and deep learning to well-defined, previously intractable problems.