Artificial Intelligence & designing tsumego

[Atorrante]

Is it possible for AI to design (new) tsumego puzzles?
And if so, has this already been tried?
What are the results and experiences?

[jlt]

I created a life and death solver and generator

A while ago I started creating an app for playing tsumego puzzles, and some of its shortcomings could be solved by automatically generating the puzzles, as well as using a solver for providing in-app “intelligence”. So a few months ago I started creating a solver and generator for life and death puzzles, and it’s finally got to a point where it works end–to-end. Anyway, here’s the link: https://github.com/cameron-martin/tsumego-solver Please try it out, ask any questions and file a bug if you…

(I didn’t try it.)

[Atorrante]

@jlt

(I didn’t try it.)

I did try the app but it is not available anymore (or at least not when I tried).

And that github page is abracadabra for me.

[shinuito]

Generating Tsumego from games on OGS

There won’t be a unique solution for many of the puzzles you find in Go, unless it’s about capturing or living with a certain group, and that its status decides the game. It’d be a fairly limited set of puzzles in that instance. In a different thread (The Conquest of Go - A Videogame featuring Go Gameplay - #223 by shinuito) there was a small discussion about trying to make puzzles from games. AI sensei has a tool that picks the big enough mistakes from a game and makes them a puzzle, but at th…

Within the conquest of go thread

The Conquest of Go - A Videogame featuring Go Gameplay

Defining what is ‘local’ automatically is very hard, and kind of goes back to why AIs were so bad at go for forever: evaluating a position is hard to describe in an algorithm. In any case I disagree with AI sensei’s approach in putting that much focus on the AI’s top choice. It’s the only thing it can do, not having a live connection to an engine, but I personally think it destroys much of the openness that makes go so amazing. There usually are many good moves. However, this is an interesting…

which links to for example

https://neuralnetgoproblems.com/

I believe these are ai generated problems from games.

Are you wondering about just random tsumego? Like have the computer throw random stones in a corner of the board until it finds something with unique (ish) solutions?

[Atorrante]

@shinuito

Are you wondering about just random tsumego?

That can be an option. But what first came to mind was that AI could create tsumegos itself just like it can solve problems, develop a strategy, etc.
Explain the concept of (false) eyes, liberties, capturing races, etc. (if not already done so) and formulate the goal: create a problem with a given number of black and white stones (not necessarily the same number for black and white) with possible outcomes: capture or seki.

[Groin]

Long ago i heard about some successful experimentation with programs which could generate and solve some tsumegos.
That was much before AI and i can not tell you more, hope you find information on those old program(s?).

At that time the computing go community was publishing computergo, that could be a source of information.

Tsumego was then somewhat easier as programing the full game, an interesting first step at least.

For today i only heard of AI used on famous difficult tsumego to check the existing solutions, if my memory still working, a thread on L19 maybe?

[richyfourtytwo]

A version of Katago was even explicitly trained to work one just one tsumego: Jane Street Tech Blog - Deep-Learning the Hardest Go Problem in the World