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

Three leading consciousness researchers argue in a Nature comment published 19 October that today's AI systems are not conscious, and that current approaches are unlikely to make them so. The essay is by Anil Seth of the University of Sussex, Ned Block of New York University and Giulio Tononi of the University of Wisconsin–Madison.

Their first move is to separate intelligence from consciousness. Consciousness, they write, is "what goes away with dreamless sleep or under general anaesthesia"; intelligence is about what a system can do. The human cerebellum, which holds nearly four-fifths of the brain's neurons, performs sophisticated computations without contributing to conscious awareness. People confuse the two, the authors suggest, in part because language models reflect back "our collective digitized past."

The second argument rejects computational functionalism, the view that running the right algorithm on silicon suffices for experience. The authors contend that brains are not just "computers that are made of meat."

The third reason draws on Tononi's integrated information theory: consciousness depends on a system's components being "tightly interconnected in specific ways," and deep-learning systems fall short "because of the way in which they are wired." They describe current models as "smart tools, not smart people."

Seven researchers in our Who's Who tracker circulated the piece in its first days, a sign it is landing in the safety and alignment conversation as well as consciousness studies.