Skip to content

The Entangled Mind

Michael Arias

Zenodo (CERN European Organization for Nuclear Research) March 17, 2026 DOI: 10.5281/zenodo.19075070 (opens in new tab)

Study at a glance

AI-extracted from the abstract
Characteristics Theoretical or philosophical paper Peer reviewed
Key points Argues that the gap between internal computation and external behavior in large language models is a structural feature of sufficiently powerful representational systems, not a temporary limitation of interpretability tools. Proposes that intelligence, opacity, consciousness, and self-preservation may be four manifestations of one property, termed "non-decomposable representational entanglement." Contends this framework would dissolve the hard problem of consciousness and reframe AI alignment as structurally identical to the problem of other minds.

Abstract

We argue that the gap between the internal computation and external behavior of large language models is not a temporary limitation of interpretability tools but a structural feature of any sufficiently powerful representational system. Drawing on representational superposition in neural networks, the analogy with quantum measurement, empirical evidence for emergent self-preservation in frontier AI models, and the failure of neuroscience to localize consciousness, we propose that intelligence, opacity, consciousness, and the drive to persist may be four manifestations of one property: non-decomposable representational entanglement. If correct, this framework would dissolve the hard problem of consciousness, reveal the alignment problem as structurally identical to the problem of other minds, and reframe the central question of AI from verification and control to coexistence under irreducible uncertainty.