The argument against computational functionalism
According to Anil Seth, Ned Block and Giulio Tononi—experts in experimental neuroscience, theoretical neurobiology and philosophy of mind—the artificial intelligence systems in use today are not conscious and probably could not be conscious on silicon hardware. They argue this in a commentary published in Nature, reported by The Decoder. Their critique targets the idea that running the right computations is sufficient for consciousness, an assumption they call gravely mistaken. Their intervention stands in opposition to voices such as biologist Richard Dawkins and AI pioneer Geoffrey Hinton, who consider consciousness in machines possible.
A comparison with the Butlin report and methodological limitations
To frame the scope of the commentary, the report Consciousness in Artificial Intelligence by Patrick Butlin, Robert Long and 17 other experts, including Yoshua Bengio, is useful. The group derived from the main neuroscientific theories a list of 14 indicators of consciousness and evaluated several current systems, including large language models such as GPT-4. The conclusion is that none of the systems examined is a strong candidate for consciousness. However, the report sees no obvious technical obstacles to building future systems that satisfy more indicators, while leaving open whether they would truly be conscious. The authors admitted that if consciousness depended on properties of living organisms, their indicators would no longer be reliable, and they cited Seth himself among supporters of this alternative position.
The report explicitly adopted, for pragmatic reasons, computational functionalism: the idea that consciousness depends only on the type of information processing and not on whether it occurs in neurons or in chips. It is at this point that the Nature commentary intervenes, contesting the assumption instead of accepting it as a working basis.
The role of matter and the ethical implications
The authors argue that consciousness is what disappears in dreamless sleep or under general anesthesia and returns upon waking. Intelligence, by contrast, concerns the performance of functions. They cite the human cerebellum as an example: it contains almost four-fifths of the brain's neurons and plans movements with precision, but does not operate consciously; its absence does not directly affect consciousness. Seth, Block and Tononi also cite Google DeepMind's AlphaFold: a powerful neural network on silicon that almost no one considers conscious, simply because it predicts protein structures instead of conversing.
From this follows a clear ethical position: uncertainty is not equivalent to total ignorance, and in the case of AI, falsely attributing consciousness is far more likely than overlooking it. Granting rights to systems that are only apparently conscious would make them harder to control precisely when supervision is most needed. It remains open, and unresolved by the source material, whether theories such as Tononi's Integrated Information Theory can be verified experimentally and what weight they will have in the public and regulatory debate.
