Feature request: AI analysis in puzzles

Two simple requests inspired by lichess:

  1. For supporters, could we turn on the AI analysis in puzzle mode? Lichess allows players to explore puzzles (after attempting to answer) using stockfish. It’s very helpful for exploring move replies that the player might not have considered, as well as understanding the “why” behind a correct answer. OGS has some awesome puzzle sets that do not include an answer at all, and it would be amazing to have the AI help confirm possible answers.

  2. Similarly, even for puzzles that do have an answer, it would be amazing to have the AI show the life or death status of the group. Right now a lot of puzzles have life/ko/seki as equivalent “correct” answers, but I think they would be better teaching tools if the “Correct” status always showed the name of the status

I say this as someone who spends a lot of time doing OGS puzzles every day, so these would be significant quality of life improvements for me and maybe for other players

4 Likes

Something like that would be great, but it might not be that simple. Chess puzzles are all whole board puzzles, Go puzzles rarely are. Ver often the AI will tell you that tenuki is the best option. (But there are ways arount that, though I don’t know if anything works perfectly.)

Even translating AI output to terms like ko, seki, alive, dead (heck, even ‘group’) might be trickier than it sounds. I’m not an expert and I certainly don’t say it can’t be done. Just saying that it might not be simple.

3 Likes

I’m sure you’re right that there will be difficulties. If I was developing the feature (if OGS was open source, I’d try), I would limit it to a specific set of puzzle criteria.

  • First of all, whole-board puzzles (lichess style) would be really cool. The existing AI anaylsis available to all players would generate up to three pretty good puzzles from every game. I think the way lichess does it is it only considers positions where there is one and only one move that the AI judges to be dramatically better than any other. That eliminates situations where “tenuki” or a variety of (say) invasions are all equivalent.
  • second, I’d focus on puzzles where the outside is completely surrounded. I think modern AI’s can run against a non-19x19 sized board. So imagine a tsumego where black completely surrounds white in a 5x10 board. Asking AI to evaluate the score will tell you a lot about the status of both groups
  • third, if you wanted to get slightly more complicated, you could use AI to design an optimal “neutral board” that is filled in offscreen away from the puzzle. some kind of situation that is completely settled and where AI cannot find many useful moves in that 3/4 of the board. I bet a lot of puzzles that might not fit into category #2 above would become pretty clear in this situation. Again, wouldn’t work for every puzzle but it’d be easy to only show the AI option for puzzles where it produces a clean output.

If we did some combination of these, I bet puzzle makers would start to get used to the constraints and design their puzzles to take advantage.

This topic was automatically closed 91 days after the last reply. New replies are no longer allowed.