nature.com web signal

Seth, Block, Tononi: AI isn't conscious, and may never be

TL;DR

  • Three senior consciousness researchers argue in Nature that current AI systems are not conscious and are unlikely to become so along today's development paths.
  • Their first point: the human cerebellum holds nearly four-fifths of brain neurons yet does not contribute to conscious experience.
  • They warn that false positives, mistaking the appearance of consciousness for the real thing, are far more likely than false negatives, so a precautionary approach would be misguided.

Three consciousness researchers used the pages of Nature to argue that current AI systems are not conscious, and probably will not become so on today's trajectories. The authors, Anil Seth, Ned Block and Giulio Tononi, write that 'we think that existing AI systems are not conscious, and that, along current trajectories of development, they are not likely to become conscious.'

Their first argument separates intelligence from consciousness. They define consciousness as 'what goes away with dreamless sleep or under general anaesthesia, and what returns after waking up or while dreaming,' and point to the cerebellum as a case where sophisticated computation yields no conscious experience at all. It contains, they note, 'nearly four-fifths of the neurons in the human brain.'

Their second argument rejects computational functionalism, the idea that any sufficiently rich algorithm running on silicon could in principle be conscious. They call that view fatally flawed, writing that brains are not just 'computers that are made of meat.' The third reason draws on Tononi's integrated information theory: consciousness depends on components being 'tightly interconnected in specific ways,' and deep-learning systems fall short 'because of the way in which they are wired.'

The authors are blunt about the stakes. LLMs, they write, are 'mirrors that reflect an image of our collective digitized past,' and 'we gaze into them and all too readily confuse the reflections with our own conscious minds.' On the question of extending AI welfare protections as a precaution, they argue that 'false positives are much more likely to occur than false negatives,' so 'a precautionary approach would be misguided.' Seven researchers we track in our Who's Who directory circulated the comment in the days after it ran.

Shared on Bluesky by 7 AI experts (top 5 by trust)