I appreciate that OGS had integrated automatic AI analysis after every match and can give us three most terrible moves that caused big win rate change.
However, I found AI useless to me and perhaps in general, not useful for amateurs, simply because I cannot understand it. For AI, if one wins by 1 point in the foreseeable future with no mistake in endgame, then one’s win rate would be more than 90% or it’s a guaranteed win. But amateurs have tons of mistakes in endgame that will result in point fluctuations. In earlier stages it’s even more of a mystery. AI would think my move caused 30% drop in win rate and suggests me play somewhere else. But that somewhere else is totally unrelated to current battle and simply is a tenuki because it has bigger value than the current “small” battle. Even after playing out the suggested move tree, I cannot understand why the new area would be of greater value if I forfeit the current battleground.
I think the biggest thing is AI assumes both players play “correctly” and have many steps of calculation in consideration. While I cannot not make mistake every few steps. Even 9-dan pro’s games have huge win rate fluctuation according to AI but they don’t realize it. I am at a very very low level therefore the analysis seems completely irrelevant to me.
I am a level very similar to yours and often find many useful suggestions in AI analysis.
The “top three” are often misleading, but full analysis is easier to understand.
And also there is an amazing forum here where we can ask the advice of better players.
So, if you find some blue move completely meangless for you, just ask!
Actually sometimes AI tend to stress on forcing moves that are not very useful (opponent is forced to respond, so the threat is wasted). But in general I must often admit that my moves were slower or smaller or just silly respect to those of the AI.
Isn’t it interesting that a tenuki is better than continuing a small battle? I’ve been trying to think more about what the biggest move is at any given time and although I don’t understand why AI plays a particular move but I think it’s useful to know that the move you played would have been better spent somewhere completely different.
If you want to take AI review to a different level, there are actually reviewing tools which can estimate the valueof each individual move (i.e. score estimated in points) . It is instantly clear how big the consequences are of playing unnecessary exchanges, bad shapes, … And points are certainly more objective than a winrate in many cases.
spoiler I am the developer of ZBaduk, which is an effort to make high-end hardware available to the entire go world. ZBaduk is a website which can run on any modern webbrowser (also mobile), and it makes it unnecessary to buy expensive computers, because it runs the bots on dedicated GPU servers in the cloud. It actually runs 2 bots, let’s them suggest moves to each other, and finally merges their results.
Those bots are open-source bots which you can download for yourself from Github. It uses both LeeLa Zero and KataGo.
Anyway, if you like to try out ZBaduk, I can offer you a 15-days trial, which you can activate by entering voucher code EGC2019ZB or by clicking this link. After that trial all functionality stays active, however there are interruptions every x minutes.
