Feedback: AI review after game suggested terrible moves

Hello, I played game https://online-go.com/game/18796966 After game finished, AI made its short review about top 3 most changing moves and they were all nonsense. AI mentioned moves 198,197,199. In my opinion, all those moves are far from being the most game changing moves and for example move 197 is even complete nonsense muchikomi, which would not play even 20 kyu. On the other hand move 216, which allows black group in upper side to live after E12 sequence is ignored, while I personally think, it was one of moves, which changed final score (however not result of the game) quite a lot. Or speaking of that matter, earliest at move 159, black could get ko by playing similar sequence started by E12, which was there for many moves available for black to play or for white to prevent it.

I want to give this feedback in order to improve AI review, but I am not sure if it is proper way to do it or proper forum category for it. I just hope, my feedback will get to proper hands of person, who is responsible for AI reviews after game.

2 Likes

Hi - this is the right place for feedback like this.

I’m not sure if you’re aware, but the initial analysis provided by the AI is “very rough”. It is just a quick pass with a small network.

I took the liberty of triggering a full analysis for you, which is available to supporters, and you will see that it offers a much more detailed and credible analysis.

OGS is experimenting with what kind of AI analysis features to offer for free - at the moment it is the rough analysis you saw. Pretty much, you get what you paid for in this case. I’m reasonably sure that the intent of the “free 3 moves analysis” is that it would be somewhat useful - not complete rubbish - but it also has to be cheap to run. It may be tweaked with experience.

EuG

Hello Eugene, thank you for reply. I perfectly understand, that running analysis has its costs and doing it for free gives you good reputation among OGS users, but nothing else. I am fine with “very rough” analysis, just in that game, I wondered why moves 198, 197, 199 were marked as top changing moves, when they clearly should not. I think, that even “very rough” analysis should not suggest move 197 black M13 as move to play, because it is move on level of total beginner, who puts his stone into atari.

I rather thought, there happened some “glitch” and I wanted to point out possible problem. I thought, that AI probably for some reason limited considered moves beyond what it was supposed to and I just wanted to help with feedback. I did not need extra analysis for me. My intentions were more towards your benefit, rather than mine.

Anyway, thank you for your time and wish you nice day. :slight_smile:

Yes - thanks again for the feedback - it will definitely be worth us taking a look at where those three moves came from - why they were chosen.