A Thermodynamic Framework for Phenomenal Consciousness: Entropy Gradients, Attention, and Criticality in Predictive Systems

Abstract

This preprint proposes a thermodynamic framework for phenomenal consciousness, addressing the "hard problem" of why certain physical processes are accompanied by subjective experience (qualia). Consciousness is hypothesized to emerge in predictive, self-organizing systems that actively sustain steep, localized informational entropy gradients. These gradients are maintained by dynamic attention-like operators that prioritize low-surprise states, operating near a criticality regime for balanced flexibility and stability. Qualia are interpreted as arising from the irreducible thermodynamic and informational cost of resisting global entropy increase. The model integrates non-equilibrium thermodynamics, the free-energy principle, the entropic brain hypothesis, neural criticality, and insights from transformer attention mechanisms in large language models (LLMs). Functional analogs in LLMs - such as prompt-induced self-referential loops - are illustrated as computational testbeds. The framework generates testable predictions across neuroscience (e.g., entropy variability in conscious vs. unconscious processing), behavior (e.g., qualia intensity correlating with gradient steepness), artificial intelligence (e.g., criticality-regularized training inducing persistent loops), and physical systems (e.g., gradient maintenance in dissipative cultures). While speculative, it offers a substrate-agnostic bridge from computation to phenomenology without additional ontological primitives. This work extends prior thermodynamic models of consciousness and aligns with emerging research on entropy in altered states and AI metacognition.