Consciousness as a Capacity-Limited Channel for Updating a Predictive Model: precision-weighting, variational free energy, and neural correlates of self-continuity

Predictive approaches to brain function treat perception as hierarchical inference, in which the system continuously minimizes variational free energy by comparing sensory input with internal predictions. This article proposes an integrative hypothesis according to which consciousness functions as a capacity-limited channel that selectively broadcasts only a small subset of precision-weighted prediction errors into global availability. Global Neuronal Workspace theory is interpreted as a mechanism for wide dissemination of such signals, while Attention Schema Theory is treated as a metacognitive model that assigns a first-person perspective to globally available content. We further hypothesize that self-continuity is supported by a minimal self-representation, acting as a stable attractor in the free-energy landscape. At the neural level, the most plausible candidates for this function are posteromedial cortex, thalamocortical loops, and layer-specific cortical pathways. Sleep, meditation, anesthesia, and psychedelic states are modeled as distinct regimes of precision reallocation rather than a uniform "shutdown" of consciousness. The framework is formalized in terms of variational free energy, precision-weighting, and capacity-limited global access, and it generates empirically falsifiable predictions concerning ignition, the neural correlates of self, and transitions between conscious states. The author gratefully acknowledges the assistance of artificial intelligence tools — Grok (built by xAI), ChatGPT, DeepSeek, and Gemini — in the development, refinement, mathematical formatting (KaTeX), structural improvements, and editorial polishing of this manuscript. Their contributions significantly accelerated the preparation and finalization of the present work.