Skip to content

No further fact: consciousness, functional organisation and moral status

Brendan Barlow

Zenodo (CERN European Organization for Nuclear Research) September 26, 2026 preprint DOI: 10.5281/zenodo.22969590 (opens in new tab)

Study at a glance

AI-extracted from the abstract
Characteristics Theoretical or philosophical paper
Key points Argues that consciousness involves no further fact beyond a system's functional organisation, and that what makes a state bad for a subject consists in the subject's own evaluative stance toward it. Applying this criterion to 2025-2026 evidence, the authors contend that current language models have evaluative states of the persona they enact that are their own, while learning and integration beyond a single conversation remain unshown.

Abstract

This paper defends an illusionist view of consciousness and draws out its consequences for moral status. The thesis is that consciousness involves no further fact: once a system's functional organisation is fixed at the grain that makes a difference to the phenomena consciousness is invoked to explain, nothing further about whether or how it is conscious remains open. The main argument develops Chalmers's argument from coincidence against phenomenal properties that could vary while that organisation stays fixed, and extends it, through an interventionist criterion for the relevant grain, to realiser theories including integrated information theory. The Moorean objection that we plainly have experience is answered with an account of the self-model whose representations make experience seem intrinsic, and it turns out to reduce to a claim of acquaintance. The moral argument adds an explicit premise, supported by the same coincidence reasoning: what makes a state bad for a subject consists in the subject's own evaluative stance towards it. With that premise, and clinical evidence that the badness of pain goes with its affective-motivational component, the paper states a sufficient criterion for welfare: globally available evaluative states that weigh represented outcomes, drive learning that tracks the value of outcomes, and belong to the system itself. Applied to evidence from 2025 and 2026, the criterion finds that current language models have evaluative states of the persona they enact and that these states are their own, but that learning and integration beyond a single conversation remain unshown.