In Nature, Seth, Block and Tononi argue AI isn't conscious

Three of the field's heavyweights have put their names to a flat claim: today's AI systems are not conscious, and perhaps never will be. Writing in Nature, cognitive neuroscientist Anil Seth of Sussex, philosopher Ned Block of NYU, and Giulio Tononi, who runs the Wisconsin Institute for Sleep and Consciousness, lay out three reasons.

They start with a definition flat enough to argue from. "What we mean by consciousness is simple enough: it is what goes away with dreamless sleep or under general anaesthesia, and what returns after waking up or while dreaming."

The first reason is that intelligence and consciousness are different things. The authors note that "absence or removal of the cerebellum, which contains nearly four-fifths of the neurons in the human brain, does not affect consciousness directly." A structure running fast, adaptive motor computation, holding most of the brain's neurons, contributing nothing to subjective experience. If that organ can be intelligent and unconscious, the authors argue, so can a language model.

On the ambient feeling that chatbots are stirring, they are blunt. "LLMs are mirrors that reflect an image of our collective digitized past," they write. "We gaze into them and all too readily confuse the reflections with our own conscious minds."

The second reason targets computational functionalism, the view that any substrate running the right algorithm is sufficient for experience. Brains, the authors write, are "not… computers that are made of meat."

The third reason is about wiring. Integrated information theory, Tononi's framework, holds that a conscious system's components must be interconnected in a particular way. The cerebellum fails that test, and so, the authors say, do today's deep-learning systems. Seven of the researchers in our Who's Who directory posted the link the day it ran.