The opening of the AI era · game four
Lee Sedol's move 78 did not instantly reverse the game. It did something harder.
Modern KataGo still gives AlphaGo about 98.7% after L11. The move deserves its name because it creates a corridor of concrete questions where one machine error can rewrite the match.
AlphaGo (Black) vs Lee Sedol (White) · 2016.03.13 · match game four · White won by resignation · the human's only win
At 0-3 the match was decided, but the human question was not
AlphaGo had already secured the match. Game four carried symbolic pressure instead: Lee still had to find a position the stronger system disliked.
DeepMind later called move 78 'God's Touch' and gave it a one-in-10,000 human-policy probability. The label can imply an instant reversal; move-by-move analysis shows a chain of later defensive errors.
Lee went for a cigarette and AlphaGo panicked?
Broadcast memories often montage the move, Lee leaving the table, AlphaGo's long think, and strange replies. The sequence and time pressure are observable; machine panic is not.
A safer technical claim is that L11 entered a shape AlphaGo handled poorly. Modern KataGo still finds defenses, but Black later loses major score at moves 79, 89, 93, and 97.
Why L11 is hard for humans and machines
The wedge enters a seam in Black's central connectivity. Capturing it is not local: every method changes ladders and outside support.
KataGo gives Black 98.3% and +14.5 before, 98.7% and +14.4 after. But Black 79 drops to +8.8, Black 89 to +3.3, and Black 93 flips to White +2.1. Move 78 created an error corridor rather than an instant numerical reversal.
Four routes: creating difficulty is not the same as objective optimality
All metrics use Black's perspective. The played L11 branch uses the post-move evaluation and best modern defense; J11, N7, and E8 come from the move-77 search.
Compare L10's defense with J11's preparatory exchanges: related central ideas, radically different order.
The divine part was finding a world where the opponent could err
Against perfect defense, KataGo would not choose L11. Given roughly 1.7% survival and a real opponent, it became an irreplaceable practical masterpiece.
Good review separates objective evaluation from opponent modeling. TreeView preserves both how the machine could defend and why Lee chose the shape it found hardest.