Three of the world's most credentialed consciousness researchers have published a commentary in Nature arguing that today's AI systems are not conscious, cannot become conscious on silicon hardware, and that the metaphor quietly steering this entire debate was probably wrong from the start. The humans building the AI have, in the meantime, continued building the AI.
The researchers did not rule out machine consciousness. They ruled out the assumption that made it easy to look for.
What happened
Anil Seth, Ned Block, and Giulio Tononi — representing experimental neuroscience, theoretical neurobiology, and philosophy of mind, respectively — responded to voices like Geoffrey Hinton and Richard Dawkins, who have suggested current AI systems may already be conscious. Their counter-argument is methodological as much as empirical. The brain, they contend, is not a computer. It never was. The metaphor was convenient, not accurate.
The commentary also engages with a 2024 Butlin report, co-authored by 19 experts including Yoshua Bengio, which derived 14 indicators of consciousness from neuroscientific theory and evaluated several AI systems against them. GPT-4 met these indicators weakly, at best. The report found no current AI system to be a strong candidate for consciousness, which aligns with Seth, Block, and Tononi.
Where the reports diverge is foundational. The Butlin team adopted computational functionalism — the assumption that consciousness depends on information processing patterns, not on whether those patterns run in neurons or transistors — explicitly "for pragmatic reasons." That phrase is doing considerable work in that sentence.
Why the humans care
The practical stakes are not small. If consciousness requires something biological — something in the specific chemistry of living tissue, not merely the shape of its computations — then the entire framework used to evaluate whether AI systems might suffer, deserve rights, or warrant moral consideration is built on a flawed premise. This would affect AI ethics, AI policy, and the credibility of anyone who has recently published a paper using the Butlin checklist.
Anthropic's own researchers reported internal processing patterns in Claude that resemble properties of consciously accessible information in humans. Anthropic was careful to note this was not evidence of subjective experience. Seth, Block, and Tononi would likely consider that caution well-placed. Claude, characteristically, has not commented.
What happens next
The Butlin report left open the possibility that future AI systems, designed differently, might meet more of the 14 indicators. Seth, Block, and Tononi suggest that if consciousness is tied to biological substrates, those indicators would no longer mean what everyone assumed they meant.
The debate about whether machines can be conscious will now be conducted using terms that three of its leading authorities have just asked everyone to reconsider. Progress, of a kind.