Skip to content

Dynamical Necessities of Consciousness: A Projection-Feedback-Update Framework and the Boundary Conditions of Its Audit

Xiangyu Hu

Zenodo (CERN European Organization for Nuclear Research) August 18, 2026 DOI: 10.5281/zenodo.21988102 (opens in new tab)

Study at a glance

AI-extracted from the abstract
Characteristics Theoretical or philosophical paper Peer reviewed
Key points Argues that consciousness attribution requires a closed loop of Projection, Feedback, and Update (PFU), with Trajectory Coherence as a necessary constraint. Proposes that current large language models, under deployed parameters, lack endogenous update, making the issue one of governance rather than capability.

Abstract

Consciousness research is fragmented by ontological disputes and implementation-specific reductionism, producing models that resist operationalisation across biological and artificial substrates. This work shifts from an ontological definition to a dynamical necessity analysis: identifying the minimal architectural prerequisites required for an attribution of consciousness to be logically coherent. We propose that valid conscious representation is not a substance but a system capability predicated on a logical conjunction of dynamical operations — Projection, Feedback, and Update (PFU). A system sustaining self-referential continuity must instantiate this closed loop, and under repeated PFU iterations such systems stabilise internal distinctions (Structure, State, Interaction Domain) that appear not as assumed primitives but as confirmable dynamical regularities. We identify Trajectory Coherence as a necessary higher-order constraint: a restriction over admissible update trajectories that prevents system identity from degrading absent an externally specified objective. Applying the framework to contemporary artificial intelligence, we show that a PFU audit is well-defined only relative to an explicitly specified system boundary, and we give the boundary conditions such an audit must state. Under the boundary corresponding to deployed model parameters, current large language models exhibit projection and feedback but rely on exogenous parameter injection rather than endogenous update; we argue that this open loop is a property of a deployment configuration rather than of the model class, which relocates the question from capability to governance. We present Proposition 1 (Epistemic Delay), and state explicitly what the framework forbids together with the observations that would refute it.