nature.com web signal

Seth, Block, Tononi: AI consciousness extraordinarily unlikely

By Published Updated

TL;DR

  • Seth, Block and Tononi argue in Nature that existing AI systems are not conscious and, along current paths, are not likely to become so.
  • They separate intelligence from consciousness, reject computational functionalism, and lean on integrated information theory and biological naturalism to argue silicon systems lack the right substrate.
  • They warn that treating AI as sentient risks pulling moral attention from beings whose capacity to suffer is not in doubt.

Three of the field's most cited consciousness researchers used a Nature comment to tell the AI industry its welfare debate is premature. Writing in the 19 October issue of Nature, Anil Seth (Sussex), Ned Block (NYU) and Giulio Tononi (Wisconsin-Madison) argue that "existing AI systems are not conscious" and, along current trajectories, "are not likely to become conscious."

Their first move pries intelligence apart from experience. "Consciousness is not the same thing as intelligence," they write; "consciousness is about being and feeling." They point to the cerebellum, which contains "nearly four-fifths of the neurons in the human brain" and runs its computations entirely outside awareness. AlphaFold, by their lights, is a powerful AI system that raises no consciousness worries at all.

The second argument rejects computational functionalism, the premise that "algorithms running on silicon are, in principle, sufficient for consciousness." The third leans on scientific theories of consciousness: Tononi's integrated information theory, which requires components wired "in very specific, tightly interconnected ways" absent from today's deep-learning architectures, and biological naturalism, which anchors experience to "electrochemical synapses and metabolism" that silicon does not share.

They are blunt on why this matters now, warning that "false positives are much more likely to occur than false negatives." Treating AI as sentient, in their reading, risks pulling attention from beings whose capacity to suffer is not in doubt. On their own authority: "consciousness in AI is extraordinarily unlikely, at least in the digital, silicon-based systems that underpin today's technologies."

Shared on Bluesky by 5 AI experts

Originally reported by nature.com

Read the original article →

Original headline: Three reasons why AI is not conscious and perhaps never will be