Seth, Block, and Tononi argue in Nature that today’s AI is not conscious

Anil K. Seth, Ned Block, and Giulio Tononi argue in a Nature comment, published October 19, 2026, that today’s AI systems are not conscious and are unlikely to become conscious on the path the field is on. The piece is “Three reasons why AI is not conscious and perhaps never will be”, in Nature volume 658, pages 866–868. It is a comment: an argument from three researchers, not a new dataset.

Nature’s public page includes the opening and the first reason in full, then stops the second reason behind a sign-in wall. This account uses only the text on that open page. It does not reconstruct the rest of the essay from other people’s summaries.

Seth is a professor of cognitive and computational neuroscience at the University of Sussex. Block is a professor of philosophy at New York University. Tononi directs the Wisconsin Institute for Sleep and Consciousness and is a professor of psychiatry at the University of Wisconsin–Madison. Nature prints those affiliations.

The claim, in their words on the open page

The authors define consciousness as subjective experience: “what goes away with dreamless sleep or under general anaesthesia, and what returns after waking up or while dreaming.” They quote Thomas Nagel’s 1974 line that there is “something that it is like to be” a conscious organism. Intelligence, they write, is about doing. A system can be intelligent and still not be conscious.

They say the stakes are practical. If systems were conscious, questions about distress and about “AI welfare” would follow. If systems only seem conscious, treating them as subjects could pull attention away from governing them. Their judgment is that false alarms are much more likely than missed cases, so a precautionary grant of protections would be a mistake. That is their weighing of the risk, not a measurement.

They write that Richard Dawkins and Geoffrey Hinton have argued AI might already be conscious, and that Anthropic this year published research “claiming to have identified computational signatures of a form of consciousness” in Claude. Both points are the comment’s citations. This desk did not re-read those sources for this story. The comment’s conclusion does not depend on accepting either claim.

The first reason, and the start of the second

The first reason is the split between intelligence and experience. They use the cerebellum as the example: it adjusts movement in a precise way, it holds nearly four-fifths of the brain’s neurons, and, they write, people are unaware of its computations. Removing it, they say, does not remove consciousness directly. They argue that fluent chat invites anthropomorphism, because the models are trained on text about thoughts and feelings. AlphaFold, they note, is also a neural network on silicon, and almost nobody asks whether it is conscious, because it predicts protein structures instead of talking.

The second reason, as far as the open page goes, rejects computational functionalism: the assumption that the right algorithm is enough for consciousness, whether it runs in tissue or in silicon. They call that view “fatally flawed because brains are not — or at least not just — computers that are made of meat.” The page then asks the reader to sign in. The third reason is not on the open page, so it is not summarized here.

Nature’s competing-interests note says Seth is an adviser to Conscium Ltd and AllJoined Inc, a co-founder and paid adviser of a stealth-mode AI company, and is contracted to write lectures for Wondrium. Block is supported by the Templeton World Charity Foundation and the Longview Foundation. Tononi is supported by that Templeton foundation and by Tiny Blue Dot, and is a board member with a financial interest in Intrinsic Powers Inc. Those are disclosures, not findings about the argument.

The comment is one position in a live scientific argument. Readers who want the full three reasons need the complete Nature text. What the open page supports is the authors’ conclusion and the distinction they draw between performing a task and having an experience.

This report is not legal, financial, or health advice. See the disclaimer.