Rédiger un PREreview

This paper presents a formal specification of a hierarchical predictive-processing architecture in which phenomenological constructs (termed phantoms) are modeled as emergent attractors in bidirectional message-passing between five architectural layers of cognition.

The framework proposes: (1) that information flows in both directions through the same architecture, upward from sub-perceptual sensing through consciousness, felt-self integration, and dual-process cognition, and downward from sociocultural language through ideology, habitus, and installed conscience into felt-body reality; (2) that emergent phantoms at each joint can be modeled as attractor states with definable persistence and basin-depth properties, crossing operational autonomy thresholds under specifiable conditions; (3) that cost-calibrated permeability parameters govern updating at each joint, with dimensional variation predicted to correspond to certain clinical trait dimensions; (4) that interoceptive sensitivity operates as a shorthand for a multidimensional individual-difference variable indexing how strongly deeper-layer signals influence actionable perception; and (5) that consciousness in the self-aware experiential sense may correspond to sustained mutual information across the active phantom attractor set.

The paper includes a formal computational specification, a working agent-based simulation demonstrating the framework's core dynamics, provisional testable predictions with explicit falsification conditions, and stated positioning as an extension of existing frameworks (Friston, Tononi, Baars, Dehaene, Metzinger, Seth, Damasio, Porges, Hoffman, Bourdieu). No original empirical data are presented. The paper is a theoretical synthesis offering dimensional hypotheses, not diagnostic or clinical claims. Load-bearing claims requiring empirical validation are named explicitly in the scope section.