Twelve-game comparative study · 16

Twelve classic games show that the hardest move is not always the AI best move

From Jowa's ghost moves and Shusaku's ear-reddening move to AlphaGo 37 and Lee Sedol 78, twelve variation trees repeat one boundary: objective value, practical difficulty, and historical courage are not one metric.

1. Split one brilliant move into three different questions

   First ask objective value: against perfect replies, how many points separate the move from the best candidate? Second ask practical difficulty: how many concrete decisions does it force the opponent to solve? Third ask historical courage: under the period's knowledge, rules, match pressure, and visible information, what boundary did the choice cross? KataGo is strongest on the first and helps expand the second; it cannot decide the third alone.

2. Jowa's first ghost move shows resistance can lose evaluation

   White 68 in the blood-vomiting game became the first ghost move, yet this deep search raises Black from about 92.3% to 96.7% and widens the score lead. The move still matters because Jowa declined the cleanest survival line and pulled the left side and center into a fight that demanded calculation. History praises sustained responsibility under disadvantage, not a coordinate certified by AI.

3. The ear-reddening move separates whole-board vision from top value

   Shusaku's Black 127 attended to several directions, and an observer read its force in Gennan's reddening ears; modern search prefers more direct candidates. The useful response is not to reverse the legend into “bad move,” but to distinguish expressive whole-board organization, its shock to human attention, and a current model's fine value ordering. The lesson is how the move organizes the board, not its central coordinate.

4. AlphaGo 37 made unfamiliarity and correctness overlap

   The shock of the P10 shoulder hit came not only from high value. Contemporary professional policy rarely prioritized it, yet its purpose—press White low while retaining central flexibility—can be explained in Go terms. Unlike a human classic that chooses complexity under pressure, the machine expanded the boundary of human candidate generation. Search moved an alien-looking point into an intelligible high-value region.

5. Lee Sedol 78 demonstrates best play against a particular opponent

   Modern KataGo still gives AlphaGo about 98.7% after L11; the move did not instantly reverse a static number. It found a chain of questions the machine had to answer accurately in an almost lost game. Against perfect defense, another move may maximize objective survival. If the goal is to create a real error path inside roughly 1.7% survival, L11 is an irreplaceable match masterpiece.

6. Lee Sedol's broken ladder turns local loss into a whole-board ledger

   In the 2003 ladder game, Black 67 through Black 97 let stones be chased while repeatedly cashing sente elsewhere. A local view says the ladder fails; a one-node label misses the accounting route. The variation tree shows whether local loss purchases outside order, thickness, or time, rather than turning a proverb into a disabled button.

7. Keep two answers when reviewing your own game

   Preserve the objective route favored by adequate search and the route that you or your opponent would find hardest in practice. Compare which risk each removes, how many consecutive choices it creates, and whether it depends on an error. When ahead, move closer to the objective route; when behind, complexity may buy chances, but “hard to answer” must not be disguised as “numerically best.” Keep decision conditions, not a gallery of divine coordinates.