Seth, Block and Tononi argue in Nature that today's AI is not conscious and probably cannot be on silicon

Anil Seth, Ned Block and Giulio Tononi argue in a commentary for Nature that today's AI systems are not conscious and probably cannot be on silicon hardware. They reject the idea that running the right computations is sufficient for consciousness, which challenges a central assumption in the debate, according to a report by The Decoder (published Nov 8, 2026).

The three authors have spent decades researching consciousness from the perspectives of experimental neuroscience, theoretical neurobiology and philosophy of mind. Their commentary responds to biologist Richard Dawkins and AI pioneer Geoffrey Hinton, who consider AI systems possibly already conscious. The authors also point to a study Anthropic published this year, in which the company reported internal processing patterns in Claude that resemble properties of consciously accessible information processing. Anthropic does not see this as evidence of subjective experience.

The Butlin report as backdrop

The commentary follows the report "Consciousness in Artificial Intelligence" by Patrick Butlin, Robert Long and 17 other experts, including Yoshua Bengio. The team derived a checklist of 14 indicators from leading neuroscientific theories. The more indicators a system meets, the stronger the case that it could be conscious.

The report evaluated several current systems, including large language models such as GPT-4, and concluded that none is a strong candidate for consciousness. For language models, it found only weak signs that they meet the indicators of one of the most prominent theories. It also saw no obvious technical barriers to building future systems that meet more indicators, while leaving open whether such systems would be conscious.

The report assumed computational functionalism, the view that consciousness depends only on the type of information processing, not on whether it runs on nerve cells or chips. Its authors adopted this "for pragmatic reasons" and acknowledged the indicators would no longer be reliable if consciousness is tied to properties of living organisms. They cited Seth as a proponent of that opposing view.

The three arguments

Intelligence is not consciousness. The authors define consciousness as what disappears in dreamless sleep or under general anesthesia, and cite Thomas Nagel: "Conscious experiences feel like something." Intelligence means carrying out functions. Their evidence is the cerebellum, which holds nearly four-fifths of all neurons in the brain and plans movements with precision, speed and learning, none of it consciously. They say consciousness is not directly affected if the cerebellum is missing or removed. They write that "LLMs are mirrors that reflect an image of our collective digitized past." Alison Gopnik of UC Berkeley made a similar argument, saying AI systems mainly reflect their training data. The authors also note that AlphaFold from Google DeepMind is a powerful neural network on silicon, yet hardly anyone considers it conscious.

Brains are not computers made of flesh. The authors call computational functionalism "fatally flawed." They trace it to Alan Turing's 1936 work and the 1943 artificial neuron model of Warren McCulloch and Walter Pitts, and call the resulting brain-as-computer metaphor "deeply problematic." In biological brains, they argue, what the tissue does cannot be separated from what it is made of. Any evidence that the brain's material matters for its function counts against computational functionalism, and by extension against consciousness in today's AI.

Leading theories rule out today's AI. Tononi's Integrated Information Theory (IIT) holds that consciousness requires a physical system whose components are tightly interconnected in specific ways. According to the theory, computers running deep-learning systems cannot meet these conditions because of how they are wired, however powerful they are. The Butlin report excluded IIT because it is incompatible with computational functionalism. Biological naturalism, separately, ties consciousness to material properties of living brains and bodies. The authors cite electrochemical signaling, electromagnetic fields that bind activity across the cerebral cortex, metabolism, and the capacity of living systems to renew themselves.

Behavioral tests and false attribution

The source reports that no universally accepted, definitive test for consciousness exists. The Butlin authors rejected behavioral tests because AI systems could be trained to mimic human behavior while working differently inside. They warned that language models would likely lead many people to believe they are talking to a conscious counterpart, citing former Google engineer Blake Lemoine's belief that LaMDA was sentient.

Some researchers call for a precautionary principle: if there is a realistic chance of consciousness, AI systems should be protected from possible suffering. The commentary's section on ethics is headed "The bigger danger is false attribution."

What this means

The disagreement is less about today's systems than about the premise used to evaluate them. The Butlin report and the Nature authors agree that no current system is a strong candidate. They split on whether future systems could qualify, because the report's checklist depends on computational functionalism, which the authors reject. The Butlin authors state that their indicators lose reliability if consciousness is tied to living organisms.

For anyone interpreting interpretability results, including Anthropic's reported processing patterns in Claude, the commentary's position is that resemblance to features of conscious processing does not establish subjective experience. Anthropic itself draws the same line.