Structural Coherence and State Selection: An Information-Theoretic Continuation of Consciousness-Primary and AI Alignment Research
Zenodo (CERN European Organization for Nuclear Research) December 20, 2025 DOI: 10.5281/zenodo.17996365 (opens in new tab)
Study at a glance
AI-extracted from the abstract| Characteristics | Theoretical or philosophical paper Peer reviewed |
|---|---|
| Key points | Proposes that structural coherence, defined as the capacity to stabilise internal state representations under recursion, perturbation, and delayed feedback, can be evaluated in artificial and biological systems without resolving whether they are conscious. Argues that hallucinations, identity drift, and goal instability are better analysed as coherence failures, and that alignment and safety should be framed as problems of structural stability rather than behavioural compliance. |
Abstract
Contemporary debates on artificial consciousness (AC) are increasingly constrained by an epistemic impasse: while advanced artificial systems exhibit behavioural and functional markers associated with consciousness in biological organisms, no empirically grounded method exists to determine whether such systems possess subjective experience. Recent philosophical work has therefore argued for agnosticism regarding artificial consciousness, emphasising the limits of extrapolating biological evidence to non-biological systems. This paper proposes a complementary research direction that remains compatible with agnosticism about consciousness while enabling practical evaluation of cognitive stability in both biological and artificial systems. Building on prior work introducing Consciousness as a Primary Field (CPF) and the A-TEST benchmark for long-horizon coherence, we develop a structural account of awareness grounded in information-theoretic and dynamical principles rather than ontological claims about consciousness itself. We formalise structural coherence as the capacity of a system to stabilise internal state representations under recursion, perturbation, and delayed feedback. Drawing on state-selection dynamics and Zeno-like stabilisation mechanisms, we show how coherence can be maintained in noisy systems without invoking quantum consciousness or observer-induced collapse. This framework allows hallucinations, identity drift, and goal instability to be analysed as coherence failures rather than evidence for or against consciousness. The proposed approach reframes alignment and safety in artificial systems as problems of structural stability rather than behavioural compliance. By decoupling ethical and epistemic questions about consciousness from measurable properties of coherence, the paper offers a conservative yet actionable framework for evaluating advanced AI systems under conditions of fundamental uncertainty.