Zenith Grant Awardee

James Crutchfield

University of California, Davis

Co-Investigators

John R. Mahoney, University of California, Davis

Project Title

Information Thermodynamics of the Observer

Project Summary

Imagine an “intelligent” observer who leverages information to convert disorganized thermal energy (heat) into useful work. This observer has a long history, going back to James Clerk Maxwell’s 19th century paradox of a “Demon” who sorts molecules into fast and slow ones, yielding a temperature difference that can perform some useful task. Since then, it was believed that Maxwellian Demons could not be implemented. Recent theoretical and experimental advances in nanoscale thermodynamics, though, have opened the door to rethinking how smart systems, smart materials, and even living systems can harvest useful energy. Our goal is to provide a conceptual framework, and key theoretical tools, describing how observers build models of and then control their environments. It focuses on the resources required by such an observer, including energy, information, and computation. We will show that, with a proper model of the correlations and regularities in its environment, an observer can extract useful work. We posit that an observer be defined in terms of its functions: sensing, storing information from, predicting, controlling, synchronizing to, and simulating its environment. This perspective allows any physical system to be analyzed in terms of a spectrum of “functional agents”. This will put the physics of information on equal footing with our familiar physics of energy, over a focused and concrete conception of the observer, and introduce meaningful questions about the nature of structure, function, and agency in the physical universe.

Technical Abstract

An observer’s perception of randomness versus complexity in its environment depends directly on the resources available for inference. These resources include both computational resources: sensory data, memory, and time available for model estimation and decision; and thermodynamic resources: energy in thermal and work reservoirs. Surprisingly, even an environment with a finite structural complexity can appear to an observer to have infinite correlation due to measurement distortion. The consequences for representing beneficial regularities are clear: the resources required for accurate inference can be unbounded. This results in an irreducible uncertainty in predicting classical stochastic dynamical systems. Similar consequences hold when one considers what in an environment can be controlled by an observer. And, they are made even more subtle and surprising if the observer and environment are governed by quantum mechanics. The project’s goal is to draw out the informational and thermodynamical resource trade-offs for observers—physically embedded inference and control agents. The result will be a new view of the information thermodynamics of dynamical, adaptive observers: complexity and unpredictability are distinct and complementary properties with direct energetic costs. It will provide a deeper understanding of how the informational, computational, and structural aspects of a physical observer are supported by its thermodynamics and energetics.