On 19 October 2026, Nature published a commentary by the neuroscientist Anil Seth of the University of Sussex, the philosopher Ned Block of New York University, and the neuroscientist Giulio Tononi of the University of Wisconsin–Madison. The title of the piece is blunt:
“Three reasons why AI is not conscious and perhaps never will be”.
The authors argue that modern AI systems do not possess subjective experience and, under current approaches to development, are unlikely to acquire it.
The central idea of the work is to separate two concepts that public discussion almost always conflates.
Intelligence, by the authors' definition, is what a system can do: perform functions, solve tasks, control movement.
Consciousness is something else entirely: it is what it is like to be that system, in the spirit of the philosopher Thomas Nagel's formulation.
The difference can be tested on one's own brain. The cerebellum contains almost four fifths of all the neurons in the human brain and performs precise, rapidly adapting motor computations, yet we do not feel its work. The authors recall that removal or absence of the cerebellum does not directly affect consciousness. Consciousness, they say, is what disappears during dreamless sleep and under general anesthesia.
Previously, the idea of conscious AI was considered marginal, but now it is discussed seriously: the biologist Richard Dawkins and Geoffrey Hinton have publicly allowed for such a possibility, and Anthropic earlier, in 2026, published a study that spoke of computational signs of consciousness in the Claude model.
The work drew criticism, including from Anil Seth. The authors of the commentary respond to this backdrop through the asymmetry of errors: a decision must be made by weighing the risk of mistakenly denying consciousness where it exists against the risk of mistakenly attributing it where it does not. For current AI, by their assessment, the second error is far more likely.
The second argument concerns the substrate—what the system is made of. The authors reject computational functionalism, according to which it is enough to run the right program on any medium for experience to arise. Computation, they acknowledge, is sufficient for intelligence but not necessarily for consciousness. The brain, in their formulation, cannot be reduced to a computer made of meat: nervous tissue has no separation into “software” and “hardware”; its structure and physical processes are inseparable from what it does.
Therefore silicon chips, in the authors' view, do not provide the organization that consciousness requires, unlike the neural networks of the brain.
The third argument rests on specific scientific theories, above all integrated information theory (IIT), which Tononi has developed for many years. According to it, consciousness arises from the dense connectivity of a system's components, and typical deep neural networks with a layered feed-forward architecture lack such connectivity. It follows that even with full functional equivalence to a human, there may be nothing behind a machine's behavior that constitutes phenomenal experience.
Besides IIT, the authors also rely on biological naturalism, which ties consciousness to the biological processes of the brain.
It is also worth understanding the limits of the work itself. This is not an experiment but a commentary: the authors analyze already accumulated neuroscientific and philosophical data and existing positions. Their conclusions rest on the acceptance of certain theories of consciousness, and alternative interpretations remain open.
Separately, the authors warn: if AI really were conscious, this would raise questions about suffering, rights, and well-being; if it merely seems conscious, then granting it rights could complicate the management and control of such systems, while attention and resources would be diverted from beings whose capacity to suffer is beyond doubt.
The distinction between intelligence and consciousness is conveniently illustrated by comparing a calculator and a human. A calculator performs operations flawlessly, but it has no inner world, and no increase in power will turn arithmetic into an experience of the result. A human, by contrast, does not merely calculate but feels what came out.
The authors essentially classify modern AI as a tool rather than a subject, and even the most complex models, by this logic, are closer to a calculator than to a human.
The work calls into question not only the technical prospects of “conscious AI” but also broader assumptions: whether the mind can be separated from its biological basis, and what this means for ethics and for understanding human experience. If the mind is transferable to any medium, the question of machine consciousness becomes a question of engineering. If it is not, then modern AI remains a powerful tool, and moral questions about suffering should be addressed to those who are truly capable of suffering.
The authors propose that the scientific community proceed by default from the assumption that AI is not conscious. The position can be disputed, but it is precisely this position that now sets the reference point for the discussion.
