Automated elementary geometry theorem discovery via inductive diagram manipulation

Abstract

I created and analyzed an interactive computer system capable of exploring geometry concepts through inductive investigation. My system begins with a limited set of knowledge about basic geometry and enables a user interacting with the system to teach the system additional geometry concepts and theorems by suggesting investigations the system should explore to see if it "notices anything interesting." The system uses random sampling and physical simulations to emulate the more humanlike processes of manipulating diagrams "in the mind's eye." It then uses symbolic pattern matching and a propagator-based truth maintenance system to appropriately generalize findings and propose newly discovered theorems. These theorems could be rigorously proved using external proof assistants, but are also used by the system to assist in its explorations of new, higher-level concepts. Through a series of simple investigations similar to an introductory course in geometry, the system has been able to propose and learn a few dozen standard geometry theorems.