Question about AI evaluaiton


I have a question about the results from the AI evaluation. Look at the evaluation of move 7.

The placement is option A given by the review, but gives it a -3.6 result.

Also, this may be what happens when DDK play, but is seems like every move is rated badly by the AI and the winner is the player that makes the least bad mistakes.


The lettering as I understand it represents the order in which the search tree found the move. How green the move is is the actual representation of how good the move is according to the AI. Also the best move may still lead to a decrease in win rate. Imagine if you have a 10 stone chain and a 20 stone chain both in double atari. Let’s say it’s endgame and there are no larger moves available for either player. The AI is going to suggest saving the 20 stone chain, but if it puts you behind on points, the AI will still say it’s a bad result.

Same idea here (to a degree). The approach is not a bad-move, the AI simply sees that in it’s variations, people tend to lose more by starting with that move than throwing in to the corner.