AlphaGo

AlphaGo was a computer Go program developed by the British artificial intelligence company DeepMind. It was the first system to defeat a professional human player at Go, a feat long considered a grand challenge for artificial intelligence, and in March 2016 it beat the South Korean grandmaster Lee Sedol four games to one in a match watched around the world. The result is often compared to Deep Blue's victory over Garry Kasparov at chess in 1997, and it helped set off the current era of investment and public attention in deep learning.^[1][2]

History

Go, played on a 19 by 19 board with more positions than there are atoms in the observable universe, had resisted computer mastery for decades: before 2015, the strongest programs played at best at amateur level. AlphaGo combined deep neural networks with Monte Carlo tree search and reinforcement learning, learning first from human games and then by playing itself.^[1][3]

DeepMind announced in January 2016 that AlphaGo had beaten Fan Hui, the European champion, 5 to 0 — the first time a program had defeated a professional player. The five-game DeepMind Challenge Match against Lee Sedol, an 18-time world champion, was played in Seoul from 9 to 15 March 2016 and drew an audience estimated in the tens of millions. AlphaGo won the first three games; Lee then won game four with a celebrated move 78, later called the "divine move", before AlphaGo closed out the match in game five. The US$1 million prize was donated to charity, and the Korea Baduk Association awarded AlphaGo an honorary 9-dan rank.^[1][2]

AlphaGo's victory was followed in 2017 by AlphaGo Zero, which learned Go purely through self-play without human game data and defeated the original program 100 games to nil, and by AlphaZero, which mastered chess and shogi in the same way.^[2][5]

Key concepts and technology

AlphaGo's breakthrough was to replace hand-coded strategy with learned intuition. A policy network narrowed the search to promising moves, while a value network estimated who was winning from a given position; the two guided a Monte Carlo tree search that explored likely future sequences. This combination of deep learning with search allowed the program to find moves that human players had never considered.^[3]

The approach became a template for later systems. AlphaGo Zero showed that reinforcement learning from scratch, without human examples, could exceed human-derived training, and the same techniques underpinned AlphaFold's protein-structure predictions and influenced the design of "reasoning" language models that spend additional computation searching through possible answers before responding.^[2]

Legacy

The match turned AI into a mainstream topic and is credited with accelerating investment and talent flows into machine learning. In 2019, Lee Sedol retired, saying that even at the top of his game there was an entity he could not defeat — a remark that became a touchstone for debates about human achievement and automation. In March 2026, the tenth anniversary of the match was marked with commemorative events in Seoul and a wave of retrospective coverage.^[1][4]

>See Also

• Google DeepMind
• Reinforcement Learning
• Monte Carlo Methods
• Deep Learning
• AlphaFold
• History of Artificial Intelligence