What happened: An eleven-chapter survey argues that we do not know whether near-future AI systems are conscious and will not know before large numbers of disputable cases exist. It canvasses ten candidate essential features of consciousness, then the arguments that might settle matters without them — behavioural tests, the Chinese Room, mimicry, global workspace and higher-order theories, integrated information, local recurrence, unlimited associative learning, and biological substrate — and finds none decisive. Chapter Four turns the author’s own work on introspective unreliability against essential-feature claims, closing with what he calls the gravest problem: a biased sample of experiencers cannot warrant claims about all experiencers. Chapter Nine records that “there is no consensus on a general theory of consciousness even for the human case” (p. 84), and Chapter Ten extends substrate liberalism to hypothetical aliens on grounds of Copernican mediocrity before declining to extend it to silicon. The closing chapters describe a “Social Semi-Solution”: social consensus arriving ahead of, and independently of, scientific warrant.
Diagnostic reading: Two things about where this lands are worth stating precisely. The limit reached is the practical one, not the principled one — the central argument’s own premise is scoped, “we cannot, in the near-term future, know through scientific inquiry which among these ten features is in fact essential” (p. 19), and Chapter Nine’s response to the theories’ failure is to conjoin their favoured features and read off relative likelihoods, since “the darkness is not pitch” (p. 85). That is criteria-aggregation offered by the person who has just shown that no criterion is decisive, and it is the epistemic-humility terminus ARG-02 v1.7 prices: a fact affirmed to exist with the discriminator declared unproducible for now. It is the strongest available specimen of that document’s standing claim, supplied by a writer with every reason to have found a fourth path. The second point concerns the parity. He runs an anti-privilege argument — Earthlings would be “too suspiciously special” if similarly sophisticated life elsewhere lacked experiencing — but between humans and hypothetical aliens, never between humans and AI, and no objection to machine consciousness is pressed back onto the human case anywhere in the manuscript. Chapter Four’s sampling argument runs the other way, explicitly granting the human case and attacking the generalisation outward. Where he stops, he states his reason rather than assuming it: AI systems are “designed or selected to mimic the behavior of conscious entities”, so a neutral observer might reasonably discount their behavioural evidence (p. 95). That is ARG-01 v1.5’s fourth outcome working as the formulation says it should, and it puts him with Birch’s gaming problem and Yasukawa in the criterion-bearing provenance family rather than with ARG-03’s arrivals at the verification limit. By ARG-01’s own stopping rule the premise is stated and not discharged: no evidence is marshalled for it and no step from mimicry to consciousness-relevance is made explicit, in a clause carrying the weight of the book’s one exception.
Four dates, one body of work: arXiv:2510.09858 first posted October 2025; this is v4, dated 30 March 2026, the version read; the Cambridge Element appeared online 7 August 2026 and in print on 27 August. Dated here to the version read, per design brief §16.3. Page references are to the manuscript's own pagination.