Zenith Grant Awardee

Susanne Still

University of Hawaii at Manoa

Project Title

Thermodynamics of Agency

Project Summary

How does physical reality reveal itself to the conscious mind? In trying to understand the fundamental nature of reality, it is by no means a given that we can have a fully objective description which is independent of the observer. Furthermore, we realize that observers are agents that change the reality for which they construct descriptions, in the process of learning about it. This brings up deep and controversial questions: Is physical theory making an intrinsically interactive process, a communication between the universe and the creatures that inhabit it? Is consciousness an emergent phenomenon of this process, or rather its original creative cause? To tackle questions such as these, a better understanding of agency in the physical world is required. Agents are embedded in physical reality, and, as such, subject to fundamental physical limits. Therefore, we can ask: Does pushing these limits result in general rules for agent behavior? We will derive rules for information acquisition and behavioral strategies from optimizing bounds on energy efficiency, and then ask how these rules come to bear on theory making. Specifically, do these rules support the reconstruction of quantum theory in an operational approach?

Technical Abstract

Can agency in the physical world be understood via underlying motivations and goals? Interpreting agent behavior as optimizing specific, context-dependent goals is deeply problematic, because it may require the construction of arbitrarily complicated objective functions. True understanding of agency must encompass the ability to derive complex behavior from simple principles. Those principles are unknown. However, we know that agents are embedded in physical reality, and, as such, subject to fundamental physical limits. Therefore, we can ask: Does pushing these limits result in general rules for agent behavior? This proposal focuses on thermodynamic limits. Recent progress has shown that certain inference rules can be derived from optimizing fundamental bounds on thermodynamic efficiency. Here, we will develop thermodynamic foundations for interactive agent behavior, which apply to non-equilibrium systems. We will then derive rules of information acquisition and behavioral strategies from optimizing bounds on efficiency. Finally, we will ask how these rules come to bear on theory making. Specifically: 1) Do these rules allow for dynamic learnability, a novel concept that this proposal defines and addresses? 2) Do these rules support the reconstruction of quantum theory, replacing (some of) the axiomatic rules of information acquisition used in an operational approach?