Seth, Block, Tononi: AI isn't conscious, perhaps never

Anil Seth of the University of Sussex, Ned Block of New York University and Giulio Tononi of the University of Wisconsin–Madison argue in a Nature essay published 19 October 2026 that today's AI systems are not conscious, and may never be.

The first move pries apart two ideas that get routinely bundled. 'Consciousness is about being and feeling; intelligence… is all about doing, or performing functions,' the authors write, defining consciousness as subjective experience, 'what goes away with dreamless sleep or under general anaesthesia.' They point to the cerebellum, which contains 'nearly four-fifths of the neurons in the human brain' yet contributes nothing directly to conscious experience, as evidence that highly sophisticated computation can proceed outside awareness. Language models mislead us partly, they argue, because they are 'mirrors that reflect an image of our collective digitized past,' trained on text that was itself written by minds.

The second reason targets computational functionalism, the view that any substrate running the right program will do. The authors contend that 'brains are not… computers that are made of meat,' suggesting biological properties beyond algorithmic processing may be essential to experience.

The third reason reaches into integrated information theory, the framework Tononi has long championed. Consciousness, the authors argue, requires a system's components 'to be tightly interconnected in specific ways.' Brain regions that support experience meet that bar; the cerebellum does not, and neither, they contend, do the computers running deep learning, 'because of the way in which they are wired.'

The essay also pushes back on the governance assumption that under-attributing AI consciousness is the bigger danger. For Seth, Block and Tononi, the false positive carries its own costs, and 'uncertainty is not the same as knowing nothing.' Seven researchers from our Who's Who roster had posted the Nature link by the end of publication day, a sign it is already landing as a reference text in the field.