[Paper Review] Sentient Self-Organization: Minimal dynamics and circular causality

Biswa Sengupta, Karl Friston|arXiv (Cornell University)|May 18, 2017

Neural dynamics and brain function20 references3 citations

TL;DR

This paper proposes that sentient self-organization in biological agents arises from approximate synchronization between internal and external states, mediated by a Markov blanket that prunes sensory and action dependencies. Using coupled complex Ginzburg-Landau equations, it shows that only specific sparsity and pruning of conditional dependencies—not arbitrary coupling—enable synchrony, which is essential for perception and action, thereby enabling free-energy minimization in nonequilibrium steady states.

ABSTRACT

Theoretical arguments and empirical evidence in neuroscience suggests that organisms represent or model their environment by minimizing a variational free-energy bound on the surprise associated with sensory signals from the environment. In this paper, we study phase transitions in coupled dissipative dynamical systems (complex Ginzburg-Landau equations) under a variety of coupling conditions to model the exchange of a system (agent) with its environment. We show that arbitrary coupling between sensory signals and the internal state of a system -- or those between its action and external (environmental) states -- do not guarantee synchronous dynamics between external and internal states: the spatial structure and the temporal dynamics of sensory signals and action (that comprise the system's Markov blanket) have to be pruned to produce synchrony. This synchrony is necessary for an agent to infer environmental states -- a pre-requisite for survival. Therefore, such sentient dynamics, relies primarily on approximate synchronization between the agent and its niche.

Motivation & Objective

• To establish a minimal dynamical model capable of sustaining nonequilibrium steady states (NESS) essential for biological self-organization.
• To investigate how a Markov blanket separates internal and external states while enabling synchronization through pruned conditional dependencies.
• To demonstrate that only specific coupling structures—characterized by sparsity and pruning—lead to synchrony between internal and external dynamics.
• To lay the groundwork for connecting thermodynamic free energy and variational free energy in the context of perception and action.
• To provide a foundation for understanding how agents infer environmental states via approximate synchronization, a prerequisite for survival.

Proposed method

• Uses coupled complex Ginzburg-Landau equations (CGLE) as a minimal model for dissipative, far-from-equilibrium dynamics.
• Introduces a Markov blanket to formally separate internal states from external states based on conditional independence.
• Employs a thermodynamic free energy functional (Lyapunov function) derived from the CGLE to characterize system evolution toward NESS.
• Imposes symmetric, finite-element mesh coupling between sensory and action fields to bound numerical stiffness and enable iterative solution of the eight coupled PDEs.
• Models the action-perception cycle as a feedback loop where sensory input drives internal states, and internal states generate actions that affect the environment.
• Analyzes convergence to non-equilibrium steady states using the chaotic hypothesis and the existence of an SRB measure, implying ergodicity in the long-term dynamics.

Experimental results

Research questions

• RQ1 Under what coupling conditions between sensory and internal states does synchrony emerge in a dissipative dynamical system?
• RQ2 How does the structure of conditional dependencies—specifically sparsity and pruning—enable synchronization between internal and external states?
• RQ3 What is the role of the Markov blanket in mediating the exchange of information and energy between an agent and its environment?
• RQ4 How can thermodynamic free energy and variational free energy be linked in a system that minimizes surprise through perception and action?
• RQ5 In what way does the existence of a synchronization manifold enable the agent to minimize prediction error and approximate environmental states?

Key findings

• Arbitrary coupling between sensory signals and internal states fails to produce synchrony; only pruned, structured dependencies lead to synchronization.
• Synchrony between internal and external states emerges only when the spatial and temporal structure of sensory and action signals is appropriately constrained by the Markov blanket.
• The system converges to a non-equilibrium steady state (NESS) characterized by a Sinai-Ruelle-Bowen (SRB) measure, indicating chaotic but statistically stable dynamics.
• The thermodynamic free energy functional derived from the CGLE acts as a Lyapunov function, ensuring convergence to NESS under the model’s dynamics.
• The existence of a synchronization manifold enables the agent to minimize prediction error, suggesting a mechanism for approximate Bayesian inference in perception.
• The framework provides a formal link between thermodynamic entropy production and Shannon entropy via variational free energy, a key step toward unifying physical and informational energy in self-organizing systems.