Toward Formalizing Common-Sense Psychology: An Analysis of the False-Belief Task

Abstract

Predicting and explaining the behavior of others in terms of mental states is indispensable for everyday life. It will be equally important for artificial agents. We present an inference system for representing and reasoning about certain types of mental states, and use it to provide a formal analysis of the false-belief task. The system allows for the representation of information about events, causation, and perceptual, doxastic, and epistemic states (vision, belief, and knowledge), incorporating ideas from the event calculus and multi-agent epistemic logic. Unlike previous AI formalisms, our focus here is on mechanized proofs and proof programmability, not on metamathematical results. Reasoning is performed via cognitively plausible inference rules, and automation is achieved by general-purpose inference methods. The system has been implemented as an interactive theorem prover and is available for experimentation.