AlphaGo vs Lee Sedol - Game 4

March 13, 2016. Four Seasons Hotel, Seoul. The Google DeepMind Challenge Match was already decided: AlphaGo had beaten Lee Sedol three times in a row, and a 5-0 sweep seemed inevitable. Yet Game 4, subtitled "Endurance" in the official DeepMind commentary, became the most celebrated game of the entire series. Playing White under Chinese rules with 7.5 komi, Lee Sedol won by resignation after 180 moves, delivering the only victory any human ever scored against AlphaGo in that five-game match. More than 200 million people worldwide had followed the series; this single afternoon would be replayed and studied for years.

The Challenge Match

In March 2016, DeepMind pitted its AlphaGo program against Lee Sedol, a 9-dan professional with 18 international titles and a reputation as the dominant player of his generation. The stakes were $1 million in prize money (pledged to charity if AlphaGo won the match) and, more importantly, a test of whether machine learning had mastered go, a game long considered too intuitive and combinatorial for computers to conquer at the highest level. AlphaGo had already defeated Fan Hui 5-0 in a closed match; now it faced the world's best in a televised event at the Four Seasons Hotel, with English commentary by Michael Redmond 9-dan and Chris Garlock, and stones placed on the board by DeepMind's Aja Huang. Lee lost Games 1, 2, and 3 on March 9, 10, and 12. Game 4 was, in effect, a dead-rubber game. Lee arrived looking calm, the crushing weight of expectation lifted.

Lee's Strategy: Endurance

For Game 4, Lee again took White. The first 11 moves duplicated Game 2 exactly: AlphaGo, playing Black, was nothing if not consistent. Lee then diverged with an unusual tenuki on move 12, testing whether the program would repeat its previous responses. What followed was a deliberate strategy of amashi: taking territory on the sides and in the corners while allowing AlphaGo to build a massive central framework. Fan Hui, commenting live, compared Lee to a wolf in winter, enduring pressure and waiting for a single decisive moment. Lee submitted to painful sequences that left professionals shaking their heads, letting AlphaGo seal in his groups and pile up influence. By move 77, AlphaGo's estimated win rate exceeded 70 percent, and Lee had barely 11 minutes remaining on his clock. Commentators began drifting away from the press room, assuming the match would end 4-0.

Move 78: The Divine Move

Then Lee played move 78, a wedge (wari) that tore through the cracks in Black's central fortress. Gu Li 9-dan, watching from a broadcast studio in China, shouted: "The divine move!" DeepMind later reported that AlphaGo had assigned the play roughly a 1 in 10,000 probability before Lee made it. Lee himself said at the post-game press conference that he had not spent long calculating; he had simply played what felt right, completing preparations he had been building throughout the game.

Professional analysis later suggested that the wedge was not unconditionally correct: a precise Black response could have refuted White's plan. But AlphaGo did not find that response. On move 79, Black pulled back instead of defending properly, and the position turned chaotic. Lee pressed forward; AlphaGo's evaluation swung wildly.

AlphaGo Unravels

What followed baffled everyone in the room. Instead of fighting the central battle, AlphaGo began trying to salvage apparently dead stones on the right side. On move 87, it played an inexplicable wedge that commentator Fan Hui called "completely beyond understanding." David Silver, AlphaGo's lead researcher, explained afterward that the program had been forced to replan from scratch after move 78 and never recovered its footing. David Ormerod of Go Game Guru described Black's play from moves 87 through 101 as characteristic of Monte Carlo search failures: when a critical line falls "off the search radar," even a superhuman engine can spiral into nonsense. By move 92, White had taken the lead; AlphaGo's win rate, once above 70 percent, plunged toward 30 percent and kept falling. Lee entered byoyomi but played with steady accuracy while AlphaGo, still making erratic choices, could not claw back. At move 180, with its winning chances estimated below 20 percent, AlphaGo resigned.

Aftermath

Reporters who had left the venue came rushing back as news spread. Lee entered the press conference to thunderous applause, smiling for the first time in the series. "It's just one game," he said. "I've never been congratulated so much just because I won one game. This win is invaluable and I would not trade it for anything else in the world." Demis Hassabis, DeepMind's CEO, promised the team would analyse why AlphaGo had collapsed once they returned to London. An Younggil of Go Game Guru called the game a masterpiece for Lee Sedol that would become famous in go history. Lee went on to request Black for Game 5, arguing that winning with White had been valuable but winning with Black would mean more; Hassabis agreed. AlphaGo won that final game on March 15, taking the match 4-1.

Why This Game Endures

Game 4 is remembered not because it proved humans superior to AI. AlphaGo won the match convincingly, and stronger successors would soon surpass it entirely. It endures because it captured, in a single afternoon, both the limits of even the most powerful machines and the irreducible spark of human creativity under pressure. Lee Sedol, facing elimination in spirit if not in score, found a move that no one expected and that the world's best go AI could not answer. Move 78 became a symbol: proof that intuition, courage, and the willingness to endure could still, just once, make the stone Buddha of the 21st century blink.

Primary sources for this account include the official DeepMind game commentary (Fan Hui, Gu Li, Zhou Ruiyang), the authenticated SGF record (RE[W+Resign], KM[7.5], 180 moves), the Nihon Ki-in and British Go Association match reports, Wikipedia's documented game analysis, and post-match statements by Lee Sedol, Demis Hassabis, and David Silver.