The brain-as-computer metaphor shaped modern AI but may be misleading the consciousness debate

The widely held assumption that advanced computation is sufficient for consciousness is directly challenged, suggesting current AI is not conscious and has no path to it on silicon.

Three prominent consciousness researchers, Anil Seth, Ned Block, and Giulio Tononi, published a commentary in Nature arguing that current AI systems are not conscious and likely cannot be on silicon hardware. They challenge the computational functionalism assumption that consciousness arises solely from information processing, regardless of the substrate. This perspective contrasts with some AI pioneers and biologists who suggest AI might already be conscious or capable of it. A related report by Patrick Butlin and 17 experts, including Yoshua Bengio, evaluated AI systems against 14 indicators of consciousness derived from neuroscientific theories. This report concluded that current AI, including large language models like GPT-4, do not strongly meet these indicators, though future systems might. The Butlin report also noted that if consciousness is tied to biological properties, its indicators might be unreliable. Anthropic's research on Claude's processing patterns, resembling conscious information processing, is also mentioned, though the company does not claim subjective experience. The Nature authors specifically contend that intelligence and consciousness are distinct, citing the human cerebellum as an example of high intelligence without consciousness. They argue that the brain-as-computer metaphor is an overextension and that the material properties of biological brains are crucial for consciousness.

The core argument posits that the "brain-as-computer" metaphor, foundational to modern AI, is a flawed overextension that obscures the biological necessity of consciousness. By separating information processing from the material substrate, computational functionalism, which underpins the possibility of conscious AI on silicon, is rendered "fatally flawed." This implies that current AI, no matter how sophisticated its algorithms or processing power, is fundamentally incapable of genuine subjective experience due to its non-biological nature. The researchers emphasize that in biological brains, function and material composition are inseparable, a crucial distinction absent in digital systems.

This scientific reframing directly counters the ethical debates surrounding AI welfare and rights, suggesting that misattributing consciousness to AI is a more significant risk than overlooking actual AI sentience. The authors advocate for a default assumption that AI systems are not conscious, framing them as "smart tools, not smart people." This position argues that pursuing "AI welfare" diverts resources from actual conscious beings and could lead to increased control issues and manipulation, particularly as AI mimics human language and behavior.