The opening of the AI era · game two
AlphaGo's move 37 made everyone ask why
P10 was not magic. It pressed White low on the right while preserving freedom toward the center. What felt alien was a priority humans rarely chose, expressed through perfectly legible Go logic.
AlphaGo (Black) vs Lee Sedol (White) · 2016.03.10 · match game two · Black won by resignation · Chinese 7.5 komi
Before move 37, AI strength still meant 'calculates deeper'
The match moved Go AI into public culture. DeepMind's retrospective says move 37 had about a one-in-10,000 probability under a human move model. That describes human-likeness, not a success rate.
The move appeared while the game was still close and proved nothing by immediately capturing stones. Its logic emerged only through the continuation.
The most common misreading of 'one in 10,000'
The number is a policy probability: how rarely a human-like model expected that choice. It is neither KataGo win rate nor a claim that the move works once in 10,000 games.
Creativity came from searching outside the human distribution, not randomness.
The Go logic of P10
The shoulder hit presses White toward low territory and keeps Black light. Its timing limits White's central growth before the right side is settled.
KataGo sees Black near 49.0% and +0.23 before the move, 49.6% and +0.24 after. P10 is among the top candidates and close to D13 and E12. The classic status comes from geometry, not a giant evaluation swing.
KataGo is no longer surprised; it offers three valid whole-board views
TreeView compares P10, D13, and E12. Similar numbers conceal different priorities: right-side pressure, upper-left settlement, or central initiative.
Follow the lines and the issue becomes move order, not a magic point.
Machine creativity became human vocabulary
Move 37 expanded professional candidate sets. Its later normality proves that Go knowledge changed.
TreeView repeats the learning process: encounter an alien coordinate, follow its consequences, then recover intelligible Go reasoning.