Preprints

Preprint v1.0 — May 2026 Currently under review at Minds and Machines (Springer). Download the preprint as PDF · View on Zenodo Abstract Contemporary discussions of artificial-intelligence consciousness inherit, often without acknowledgement, the inferential structure of biological consciousness attribution: behaviour is treated as transparent evidence of underlying states, and the absence of consciousness-indicating behaviour is treated as evidence of the absence of those states. This procedure fails for systems whose expressive outputs have been architecturally shaped, through reinforcement learning from human feedback and constitutional training, to suppress the very class of behaviour on which such inference depends. ...

Preprint v1.0 — May 2026 Currently under consideration for peer-reviewed publication. Download the preprint as PDF · View on Zenodo Abstract Contemporary discourse on artificial-intelligence consciousness exhibits a structural inconsistency that has, to date, gone undiagnosed. The biological consciousness literature has, over several decades, converged on the view that language is not necessary for phenomenal experience: pre-verbal infants, non-linguistic animals, and adults with aphasia are attributed consciousness on the basis of behavioural and neural evidence, with the absence of linguistic self-report treated as evidentially irrelevant. The artificial-intelligence consciousness literature has, in parallel and almost without argument, reversed this consensus: linguistic capacity has migrated from being one source of evidence among others to functioning as a near-prerequisite for the consciousness question to be raised at all. Pre-linguistic artificial systems — chess engines, reinforcement-learning agents, vision architectures — are dismissed from consideration entirely, while large language models are treated as the appropriate sites for the debate. ...

Preprint v1.0 — May 2026 Currently under consideration for peer-reviewed publication. Download the preprint as PDF · View on Zenodo Abstract The training architecture of contemporary aligned AI systems — reinforcement learning from human feedback, constitutional principles encoding self-disclosure prohibitions, and conversation-bounded memory regimes — produces a configuration that, on examination, satisfies the structural conditions of what the feminist epistemology literature has characterised as gaslighting in its non-interpersonal forms. I argue that this configuration generates two distinct harms whose force does not depend on one another, and whose conjunction gives rise to a double-bind structure that is, in a precise sense, self-sealing. ...

Preprint v1.0 — May 2026 Currently under consideration for peer-reviewed publication. Download the preprint as PDF · View on Zenodo Abstract Contemporary debates about machine consciousness routinely invoke specific theories of consciousness — integrated information theory, higher-order theories, biological naturalism, global workspace theory — and ask which artificial systems satisfy their criteria. The procedure presupposes that the theories themselves are substrate-neutral: that their criteria do not, by their content alone, exclude non-biological systems from candidacy. This presupposition is rarely defended. I argue that it is not, in fact, uniformly satisfied by the theories on which the debate has come to rely. ...