James P. Crutchfield
Art and Science Laboratory
Function and Intelligence in Thermodynamic Agents: Mutually Incompatible Realities
Encountering others who do not share our basic beliefs is challenging, sometimes even emotionally troubling. Of late, this seems to be an increasingly common experience in our personal and political spheres. Is the emergence of such incompatible worldviews consistent with physical principles? Isn't there a unique world "out there" that we should all (eventually and with enough time and sufficient resources) come to agree on? This project seeks to give a proper mathematical explanation for the emergence of incompatible worldviews and to draw out the physical consequences. Recent theoretical advances in optimal modeling allow us to show how and why incompatible worldviews arise, as complex adaptive agents strive to predict the world they experience. Though the incompatibilities are fundamental and inescapable, understanding how they come about will point a way to equitably addressing the challenges they present.
The thermodynamic Principle of Requisite Complexity tells us that an intelligent agent survives in and adapts to a complex environment by building and maintaining an internal model that captures as well as possible encountered regularities. What does this principle tell us when there is a population of such agents, all adapting to the same environment? Surprisingly, even for a single, fixed environment, there is a multiplicity of optimal models that, on the one hand, agents can use to their individual maximum benefit but that, on the other, do not agree on the environment's structure. All extenuating factors held fixed (noninteraction, finite resources, common history, no fluctuations, and the like) agents adhere to mutually-incompatible realities despite their experience of a common complex environment. Leveraging new methods to derive optimal predictive and generative models and appealing to a new view of agent intelligence and agent-environment semantics I will explore how multiple agents come to worldviews that are mutually incompatible. The main theorem establishes that exactly the same environmental events mean different things to different agents. In a concrete and mathematically-principled way this challenges physics' fundamental ontological stance, that it explains unambiguous phenomena in a unique world.
Hide Technical Abstract
Back to List of Awardees