The Neuro-Computational Origin of Disposition: Unconsciousness as Lifelong Prior Overfitting and Consciousness as Active Inference

A formal computational framework redefines the unconscious as an active, overfitted generative model of an individual's world, rather than a passive reservoir. Lifelong minimization of variational free energy leads to over-specialization of internal model parameters to the specific statistical regularities of one's sensory history. This overfitting creates a stable, low-entropy informational manifold that determines an individual's probabilistic disposition to perceive and act. Consciousness is characterized as the recursive, energy-consuming process of active inference, where the brain dynamically minimizes prediction error between this overfitted prior model and real-time sensory data. The framework synthesizes theoretical neuroscience, artificial intelligence, and non-equilibrium thermodynamics to derive a mathematical model for a Consciousness Potential.

Study at a glance

Abstract

Abstract A central challenge in neuroscience is to understand how an organism’s unique history of interactions with its environment shapes its behavioral dispositions and gives rise to the phenomenal experience of consciousness. Here, we propose a formal computational framework that redefines the unconscious not as a passive reservoir, but as an active, overfitted generative model of an individual’s world. We demonstrate that the lifelong process of minimizing variational free energy—analogous to empirical risk minimization in machine learning—inevitably leads to the over-specialization of internal model parameters θ to the specific statistical regularities of an individual’s sensory history. This overfitting creates a stable, low-entropy informational manifold that determines an individual’s probabilistic “destiny”, or disposition to perceive and act. Within this framework, consciousness, including social forms thereof, is characterized as the recursive, energy-consuming process of active inference, where the brain dynamically minimizes the prediction error between this overfitted prior model and real-time sensory data. By synthesizing concepts from theoretical neuroscience, artificial intelligence, and non-equilibrium thermodynamics, we derive a mathematical model for a “Consciousness Potential” and propose that the brain operates as an entropy-reducing computational system governed by a fundamental information geometry. Our framework provides a unified mathematical language for describing the interplay between experience, disposition, and conscious awareness, offering testable predictions for neuroimaging and AI research.