🇰🇷 Lee Sedol

Move 78. That is what history kept. On March 13, 2016, in the third hour of Game 4 of the DeepMind Challenge Match in Seoul, Lee Sedol — the greatest Go player of his generation, a man who had won eighteen world championships and who had said publicly that he expected to lose perhaps one game in five — placed a white stone at a position that no one in the watching room, human or machine, had anticipated. AlphaGo, computing at speeds incomprehensible to biological minds, calculated its response, calculated again, and then — for the first time in the match — began to make errors. Lee Sedol won Game 4. He became, and remains, the only human being in history to have defeated AlphaGo in an official game.

Lee Sedol was born on March 2, 1983, on Bigeumdo, a small island off the southwestern coast of South Korea. He learned Go from his older brother at age five. At twelve, he relocated to Seoul to study under the professional training system of the Korea Baduk Association, and at fourteen years and two months he received his professional 1-dan certification — the youngest person in Korean history to do so. The pace of his development was extraordinary even by the standards of Korean Go, where national competition produces talents at industrial scale. Within three years of turning professional he had won his first major international title.

Over the two decades that followed, Lee Sedol rewrote the record books of East Asian Go. He won the Samsung Cup five times, the LG Cup twice, the Korean domestic Guksu tournament multiple times, and accumulated a total of eighteen international championship titles — a figure that places him among the four or five greatest Go players who ever lived. His playing style became legendary for its creativity: he was willing, in ways that more conventionally trained players were not, to pursue high-risk, high-reward positions that violated established theory. He lost games other champions would have drawn. He won games no one else would have attempted. His fans called this quality "Lee Sedol-ness" — a combination of artistry, aggression, and a deep comfort with uncertainty that is difficult to systematize and impossible to replicate.

The DeepMind Challenge Match of March 2016 was announced as a five-game series between AlphaGo and Lee Sedol, the world's top player at that time. The prize fund was $1 million. Before the match, a global consensus among Go professionals held that even the most advanced AI could not beat a top human at Go within the decade — the game's complexity was, in the popular imagination, essentially infinite. Lee Sedol shared this view. He predicted a 5-0 or 4-1 win for himself. AlphaGo won Games 1, 2, and 3. The shock was global, covered by every major news organization on earth. More than 200 million people watched the series.

In Game 4, down 0-3 and facing a machine that had proven definitively superior in three consecutive games, Lee Sedol played Move 78. The move — a white stone placed in what Go professionals call a "wedge" tesuji, a forcing move that attacked AlphaGo's stone formation at an angle no conventional analysis would have suggested — was described by commentator Gu Li as a "divine move," and by DeepMind's own team as a position they had not properly trained AlphaGo to handle. AlphaGo's win probability collapsed from over 70 percent to roughly 45 percent in the minutes following the move. Lee won by resignation after 180 moves. "I've never been so happy to lose three games," he said afterward.

In November 2019, Lee Sedol announced his retirement from professional Go. In his retirement statement, he said that continued competition had become meaningless because "there is an entity that cannot be defeated" — a reference to AI Go programs that had, since 2016, advanced far beyond even AlphaGo's capabilities at the time of their match. He was thirty-six years old. He had won eighteen world titles. He had beaten AlphaGo once — the only person alive who could say that. His final ranked game was against HanDol, a domestic Korean AI. He lost, in three games, giving himself two-stone handicap advantages, and still lost the third game. He retired the following day. The record — one human win against AlphaGo — stands permanent and untouchable, because AlphaGo itself retired in 2017. It will never be broken.