173.165.36.107: variation analysis, computer go, etc. (2011-02-15 05:19) [#8302]

I am developing a tool for visualizing game trees and proving theorems about them. One domain I would like to include would be Go. To this end I am wondering if anyone is interested in developing formal mathematical definitions that can be used in logic programming programs for different concepts in Go. I have a research project to use ResearchCyc?, a mathematical logic based AI with natural language processing and understanding capabilities, to formalize concepts of Chess for use in building software that can do automated positional analysis in English. I am wanting to extend that project to Sensei's net as well. It seems to me that using an existing Semantic Mediawiki tool could be useful for developing formal mathematical definitions of the game and data structures for representing common structures. So for instance the thought of the day was something about Seke... something - forcing the opponent to take stones in order to diminish his territory. That concept I believe could easily be formalized using some classical/modal logic. In fact, it could perhaps be best to use OpenCyc? for the formalization of the definitions given they probably have developed adequate infrastructure for representing these concepts efficiently.

As far as analyzing the game trees, well there is something called the General Game Playing competition - that would be a good point to begin development of the logic of the game. I would hope that they had included a definition for some ruleset for Go already.

My interest is then in taking these concepts and doing automated theorem proving and positional analysis, that could be expressed in natural language. I think while search-based methods predominate in Chess, in Go the advantage of theorem-proving will be much greater.

In fact I am working on such a tool but that it analyzes your actual life and helps to recommend moves, and helps you to prioritize and execute different goals. This is the reason for my interest in the Go analyzer, is because Go is an example of a game that has a substantial literature, and Sensei's has already done such a great job of mapping out the structure of that - and could be used in such an effort. Go requires skills of balancing differing threats. I think that it would be nice to have a graphical tool that could view all of these and make plain the basic logic of how the threats interplay. So for instance, one could imagine computer-assisted Go where players use their AIs to help them choose the best moves, the AIs doing checking of the sanity and techniques, pointing out tactical mistakes or if they are running close to life and death with certain groups and visualizes how close, etc. The same type of system could then be used for a person in weighing their own real-life goals and priorities.

Furthermore, I think engines like GnuGo, etc, could probably be adopted to teach by pointing out what is strong or weak about a move. That way, more direct feedback could be obtained by players.