Automated Verification of Epistemic Properties for General Game Playing

Automatically deriving properties of new games is one of the fundamental challenges for general game-playing systems, whose task is to learn to play any previously unknown game solely by being given the rules of that game. A recently developed method uses Answer Set Programming for verifying finitely-bounded temporal invariance properties against a given game description by structural induction. Addressing the new challenge posed by the recent extension of the general Game Description Language to include games with imperfect information and randomness, we extend this method to epistemic properties about games. We formally prove this extension to be correct, and we report on experiments that show its practical applicability.