Help with OGS's AI for newbies

Hi! This might be a silly request but…

I’ve just become a site supporter but I think I’m too much of a newbie to really take advantage of the AI analysis…
So I would like to kindly ask for anyone to review any game with “full AI analysis” with the mentality of teaching newcomers how to review games using OGS’s AI further than “move X made me loose 80% winrate”
I think it would help me and any other newbie to review more often (and maybe even attract supporters? :sweat_smile:)

Thanks for any possible help!!

PS- I can suggest some game of mine, but I really don’t mind if it’s any other
PS1- I’m aware of the limitations of non-explainable AI’s, I’m just looking for pointers about the available tools :wink:


I’m no expert but my quick tips are:
Use the score change graph and not the win rate - it’s much easier to see how much effect a move had on score prediction rather than trying to interpret the win rate swinging from +99.9 to -99.9 and back!

Where you see AI persistently suggesting the same move or pair of moves then consider why you didn’t notice that that was an important part of the board to play on at that point. E.g. in your linked game, where you are ahead AI constantly wants to play p7 or q6 so we can conclude that that area was the biggest for each player to play in but both ignored it for ages. Why? What were you thinking about at the time? Can you ask yourself if something else is bigger/more important in future?

In the actual game you secured that p7 area so what went wrong elsewhere? Look through what moves the AI thought were better than moves you chose in the area you lost out in (not sure if that makes sense)

Hope helpful - but take it with a pinch of salt as I’m not an ai analysis expert (or Go expert for that matter!)


I usually look for 2 things in AI reviews (although I admit I don’t and can’t utilize them to their full potential):

  1. Check big changes in winning percentage, especially if the color changes. I mean, if I’m black and after my move the winning probability switches to white, it’s probably something I should note.
  2. If I have an idea for a move during the game, or I can’t decide between two moves, sometimes I notice it and see afterwards if there was something useful there. I do this very rarely, though.

P.S. Sometimes, it might help to discuss a big difference in winrate with an opponent. Some opponents might not be interested, but some might want to explore it together.
Especially stronger opponents usually are kind enough to discuss an important move they made.


This is part of our Documentation and FAQ:


Thanks! I read that a while ago and it’s perfect for understanding each element in the interface. (It’s good to revisit it every once in a while, so thanks!)
I was looking more for a “human guide”, for instance I think @teapoweredrobot’s suggestion to use score rather than winrate is a very good idea and I had not thought about it before. It’s silly, I know, but nobody is born knowing…

Any steps/pieces of advice/pointers about how to review is welcome!


Isn’t “score” the default? I think it should be when KataGo is the review AI.

1 Like

Yeah it’s something that makes sense when explained but hard to know why winrate would be worse than score estimate etc.

The thing is, realistically they’re intricately related. What tends to happen is that once the score estimate crosses zero, there’s a big swing in winrate in a lot of cases. It kind of makes sense in that, if the ai thinks one player was behind 5 points but is now ahead by 2-3 points, then who is winning should have changed, and hence the win percentage should probably swing from one side to the other to reflect it.

Big swings in winrate can be both minor point losses eg 1 point behind to 1 point ahead, or from 10 points behind to 10 points ahead. They can look the same in winrate, depending on the certainty of the outcome.

It can be easier to rationalize the points though, because in some cases you can count them on the board. Some one point or 3 point losses, one might be able to physically count on the board in endgame i.e. if I played this other move, then in endgame they wont have this sente move worth an extra point. Or if one misses a move to save a group, there might a point swing about twice the size of the group being captured.

I’ll give reviewing that game a go anyway, and see if there’s anything I can point out related to the ai review.

I watch basics videos whenever I can, I didn’t see these ones!! thanks!!!

I see your point, thanks for the insight on the score/winrate relation!

Thank you very much for the help!!

I should probably go to bed but I’ve left some ideas

They’re probably not all important/useful and I don’t think I’d regularly go into this much detail reviewing my own games. I normally just look for the biggest mistakes or interesting sequences, or some better directions to play in than in the game.

Then just taking away 1-2 things from each game builds up over time :slight_smile:

I’ve also linked a level 2 ai review within the review that you can compare with if you’re interested.

Have fun anyway, and there’s always something one can pick up from stronger players and from ai :slight_smile:

1 Like

Wow!! Thank you so much for the effort!!! I’ll owe you big time :wink:

1 Like

I studied the whole review, so thorough!! Thank you so much!!

I do have a question/suggestion, though. Why is there no graph (score/winrate) in the review window?? doesn’t make much sense, don’t you think?
(I noticed the icon for “Full AI Review” in the menu on the right side, but it throws a error 500)

1 Like

That’s a good question. Could possibly request this be added in as a toggle.

I know some people either don’t like seeing the ai suggestions/winrate and there’s also some issues in analysis mode, where if you played a move the ai suggested, then turn off the ai review, and play the same move it still shows the ai sequence.

I should probably post that bug on github if it’s not already there.