Costed Mode Switching under Resource-Constrained Active Inference: A Coherence-Hardened Formal Interface for Broadcast Recruitment

Abstract We propose a compact interface between Global Neuronal Workspace / Global Workspace Theory (GNW/GWT), Integrated Information Theory (IIT), and the Free Energy Principle (FEP) by treating conscious ignition as a mode selection problem under resource constraints. The system recruits global coordination when the expected benefit of integrated inference exceeds the coordination tax required to couple distributed processors. This note hardens the formalism by (i) introducing an explicit mode variable $m_t \in \{F, I\}$, (ii) defining expected free energy at the mode level via restricted policy classes, and (iii) stating admissibility conditions required to identify quadratic control cost with path-measure KL divergence (Girsanov regime). Plain Language Summary This paper proposes that consciousness behaves like an expensive "Boardroom Meeting" in a large corporation, rather than a continuously active light. Most of the time, the brain operates in a modular, low-cost "Autopilot" mode (Mode F). However, when the brain encounters data that is "non-separable"—meaning different brain modules (like vision and sound) cannot explain it separately—it triggers a "Boardroom Meeting" (Mode I). We model this as an economic decision: the brain only "ignites" consciousness when the value of solving the confusion outweighs the metabolic tax of connecting the whole brain. This framework offers a rigorous way to predict when consciousness will occur based on the "cost of connection" versus the "cost of error."