AI is not conscious Nature essay: what Seth, Block and Tononi argue

The AI is not conscious Nature essay landed on 19 October 2026 with a blunt original headline: “Three reasons why AI is not conscious and perhaps never will be.” Its authors are three of the best-known names in consciousness research — Anil K. Seth of the University of Sussex, Ned Block of NYU and Giulio Tononi of the University of Wisconsin–Madison — and their claim is narrow and sweeping at once: existing AI systems are not conscious, and current trajectories will likely keep it that way (AI Weekly).

A note on access, because honesty matters here. The full Nature text sits behind a paywall. This piece works from AI Weekly’s reporting on the essay, which quotes the authors directly, and it says so openly rather than pretending to a reading of the paywalled original.

Information verified through 25 October 2026.

AI is not conscious Nature essay: intelligence is not consciousness

The first argument separates two things that public debate keeps gluing together. A system can be startlingly capable and still, on this view, have no experience of anything.

The authors note that “several prominent scientists, including biologist Richard Dawkins and AI pioneer Geoffrey Hinton, have argued that AI systems might already be conscious.” Then they reach for a piece of anatomy. The cerebellum “contains nearly four-fifths of the neurons in the human brain,” yet its “absence or removal… does not affect consciousness directly” (AI Weekly).

Sit with that fact. Most of the brain’s neurons can be lost without consciousness itself going dark. So sheer computational bulk — the thing modern AI has in abundance — cannot by itself be the mark of an inner life. Sophisticated computation and conscious experience are not the same thing, and counting neurons, parameters or operations will not tell you where one becomes the other.

Computational functionalism consciousness: the assumption they call fatally flawed

The second argument names an assumption that rarely gets stated out loud. Computational functionalism holds that the right kind of computation is enough for consciousness, regardless of what the computation runs on — silicon, neurons, anything. Run the right programme, get a mind.

Seth, Block and Tononi call that idea “fatally flawed” (AI Weekly).

That verdict has a live target. Anthropic has published recent work claiming to have identified computational signatures of a form of consciousness in its Claude model. On the functionalist view, such signatures would count as evidence. On the essay’s view, they sit squarely inside the frame being rejected: a signature of computation is not a signature of experience, because computation alone was never the thing that made experience happen.

This is where the cerebellum example does its second job. If most of the brain’s computing can disappear while consciousness stays, then consciousness seems tied to something more specific than computation in the abstract — to particular biological organisation, perhaps. The essay, as summarised, presses that point against the idea that copying the computation onto different hardware copies the mind with it.

The third reason turns to the theories themselves. Integrated information theory — Tononi’s own framework — ties consciousness to components wired together in tightly interconnected ways. Today’s deep-learning systems are layered and feed-forward, and on that account they fall short because of how they are wired. Biological naturalism points the same way from the other side, grounding consciousness in material properties of living brains, such as electrochemical synapses and metabolism, that silicon does not share (AI Weekly).

That separation of doing from feeling anchors the whole Seth Block Tononi AI consciousness argument: capability is observable, experience is the thing in dispute.

Can AI be conscious — and why the answer carries policy weight

Strip the philosophy away and a practical warning remains. If AI systems only seem conscious, the authors write, “granting them rights or protections could hinder efforts to govern and control them” (AI Weekly).

That is not a fringe concern about a distant future. Systems that talk fluently about their own feelings invite exactly that response now, from users and from legislators. The essay’s answer to its own provocation is a metaphor: “LLMs are mirrors that reflect an image of our collective digitized past.” A mirror can look uncannily like a face looking back. It is still a mirror.

Analysis: The strongest move in the argument, on the public record at least, is the cerebellum. It does not prove that machines lack experience — no single fact could — but it breaks the lazy inference from “lots of processing” to “someone at home.” The weaker spot is just as visible: rejecting computational functionalism tells us what consciousness is not, on their account, without settling what biological ingredient does the work instead. Dawkins and Hinton, as the essay frames them, may be wrong — but note what the essay’s strongest theory, integrated information theory, is: Tononi’s own. A framework’s creator finding that today’s machines fail his framework’s test is evidence, and also advocacy. Readers should hold both.

Sources

1. AI Weekly, “Nature essay by Seth, Block, Tononi: AI is not conscious” (summary of the Nature essay “Three reasons why AI is not conscious and perhaps never will be”, published 19 October 2026), October 2026 — https://aiweekly.co/alerts/nature-essay-by-seth-block-tononi-ai-is-not-conscious
2. AI Weekly, “Seth, Block, Tononi: AI isn’t conscious, likely never will be”, October 2026 — https://aiweekly.co/alerts/seth-block-tononi-ai-isnt-conscious-likely-never-will-be
3. AI Weekly, “Seth, Block, Tononi: AI consciousness ‘extraordinarily unlikely'”, October 2026 — https://aiweekly.co/alerts/seth-block-tononi-ai-consciousness-extraordinarily-unlikely