Structured dynamics in the algorithmic agent
Giulio Ruffini,Francesca Castaldo,Edmundo Lopez-Sola,Roser Sanchez-Todo,Jakub Vohryzek
DOI: https://doi.org/10.1101/2023.12.12.571311
2024-10-17
Abstract:In the Kolmogorov Theory of Consciousness, agents utilize inferred compressive models to track coarse-grained data produced by simplified world models, capturing regularities that structure subjective experience and guide action planning. Here, we study the dynamical aspects of this framework by examining how the requirement of tracking natural data drives the structural and dynamical properties of the agent. We first formalize the notion of {\em generative model} using the language of symmetry from group theory, specifically employing Lie groups to describe the continuous transformations observed in natural data. Then, by using a neural network model to represent the agent system, we demonstrate that data tracking forces the agent to mirror the symmetry properties of the generative world model. This constrains both the constitutive parameters of the agent (structure) and its dynamical repertoire, imposing a hierarchical structure that reflects the organization of the world, aligning with the {\em manifold hypothesis}. Our findings connect the perspectives of algorithmic information theory (model, Kolmogorov complexity, simplicity), symmetry, and dynamics (group theory, reduced manifolds), providing new insights for identifying neural correlates of agenthood and structured experience in natural agents as well as enhancing the design of artificial intelligence and computational brain models.
Neuroscience