Information, Physics and the Representing Mind

Abstract

A primary function of mind is to form and manipulate representations to identify and choose survival-enhancing behaviors. Representations are themselves physical systems that can be manipulated to reason about, predict, or plan actions involving the objects they designate. The field of knowledge representation and reasoning (KRR) turns representation upon itself to study how representations are formed and used by biological and computer systems. Some of the most versatile and successful KRR methods have been imported from computational physics. Features of a problem are mapped onto dimensions of an imaginary physical system in which solution quality is inversely related to energy. Simulating the fictitious physical system on a digital computer yields a low-energy, and hence high-quality, solution to the original problem. This paper suggests a rethinking of the traditional metaphor of cognition as execution of algorithms on a digital computer.  It may be both more fruitful and more accurate to conceive of representation as mapping problem features to an energy surface, learning as identifying representations that map good solutions to low free energy, and problem solving as efficient search for low free energy states. This conception of cognition is in natural accord with Stapp's theory of efficacious conscious choice.