Seth, Block, Tononi lay out why AI likely isn't conscious

The cerebellum contains nearly four-fifths of the neurons in the human brain. It does not produce conscious experience. That anatomical fact anchors the opening move of a Nature commentary from Anil Seth of the University of Sussex, Ned Block of NYU, and Giulio Tononi of the University of Wisconsin-Madison, published 19 October 2026, that argues current AI systems are not conscious and probably will not become conscious along their current trajectory of development.

The core separation runs through the piece. 'Consciousness is about being and feeling; intelligence… is all about doing, or performing functions,' the authors write. Consciousness, in their definition, is 'what goes away with dreamless sleep or under general anaesthesia, and what returns after waking up or while dreaming.' Intelligence, by that split, is a different problem that silicon can clearly handle.

The second reason targets computational functionalism, the assumption that running the right program on any substrate is enough for experience. 'Brains are not… computers that are made of meat,' the authors contend. The third reason reaches into integrated information theory, Tononi's framework, concluding that deep-learning systems fall short 'because of the way in which they are wired.' Consciousness, they argue, requires a system's components 'to be tightly interconnected in specific ways' that today's architectures do not satisfy.

Running underneath is a warning about readers. Large language models exploit 'anthropocentrism, viewing the world through a human lens, and anthropomorphism, attributing human characteristics to things that merely resemble people.' LLMs, in the authors' phrasing, are 'mirrors that reflect an image of our collective digitized past.' In the AI case, false positives, crediting consciousness that isn't there, are 'much more likely' than false negatives.

The piece is commentary, not an empirical study, and it does not offer a test that would settle the question for any specific model. Seven of the researchers we track shared it.