Moves that defy AI expectations

Ueno Asami played Nyuu Eiko in Female Meijin league.

It is said that this 1-3 combination by Ueno Asami is better than computer suggestions. Including FineArt and Golaxy.

Indeed on the relay FineArt suggest extend with slight disadvantage for black and then flips to black advantage

Though on my laptop b18c384 with default config eventually finds 1.

But not 3.

Though it agrees it’s good after it’s played.

Looks like ko on the left and all that reading at the bottom with additional reading in top-right makes too many moves to read?


But… what does it mean to be better than the AI’s move? How do you measure that?


Don’t know what is so special about this tpoic. :grin:



It’s a move that the AI didn’t think about, but once played, the AI thinks it’s better than what the AI had in mind.


Also there is a very large unresolved ko, which might make all sequences harder to read since at any point, a player could tenuki and start playing the ko.