2016
AlphaGo defeats Lee Sedol
DeepMind's AlphaGo defeats world Go champion Lee Sedol 4–1, twenty years earlier than experts expected.
The game no computer was supposed to win
In March 2016, DeepMind's AlphaGo played a five-game match against Lee Sedol, one of the greatest Go players of all time. AlphaGo won 4–1. The result shocked the AI world and the general public. Go was long considered the last board game beyond the reach of computers — too complex for brute-force search, requiring a kind of intuition that was thought to be uniquely human. Most experts had predicted human superiority for at least another decade.
How AlphaGo worked
AlphaGo combined two deep neural networks — a policy network that selected candidate moves, and a value network that evaluated board positions — with Monte Carlo tree search. The policy network was first trained on a dataset of 160,000 expert games (supervised learning), then improved by playing millions of games against itself (reinforcement learning). This combination of supervised and reinforcement learning was the key innovation that made AlphaGo possible.
Move 37
In game 2, AlphaGo played a move — later called Move 37 — that stunned the Go world. The move was so unexpected that commentators initially thought it was an error. Professional Go players rated it as one in ten thousand. Lee Sedol left the room for 15 minutes. When he returned, he had to admit it was a brilliant move. That moment — when a machine played a move so creative that humanity needed time to comprehend it — became a symbol of a new era in AI.
Legacy
AlphaGo was followed by AlphaGo Zero (2017), which learned entirely through self-play without human game data, and AlphaZero, which mastered chess, shogi, and Go simultaneously. DeepMind's follow-up, AlphaFold 2 (2020), applied similar techniques to solve the protein folding problem — demonstrating that the combination of deep learning and reinforcement learning could tackle problems far beyond games.
Sources
- Silver, D. et al. (2016). Mastering the game of Go with deep neural networks and tree search. Nature, 529, 484–489.
- Wikipedia — AlphaGo versus Lee Sedol