Seth, Block, Tononi argue silicon AI won't be conscious

Anil Seth, Ned Block and Giulio Tononi argue in a Nature comment that consciousness in current AI 'is extraordinarily unlikely, at least in the digital, silicon-based systems' running today's models. The three authors come from different theoretical traditions and still arrive at the same verdict.

Their first move pulls intelligence and consciousness apart. 'Consciousness is about being and feeling; intelligence — very broadly — is all about doing, or performing functions,' they write, defining consciousness as 'what goes away with dreamless sleep or under general anaesthesia.' The human cerebellum, which holds nearly 80% of brain neurons, performs sophisticated processing 'wholly outside awareness,' the authors note, treating that as evidence that intelligent computation and conscious experience come apart.

The second reason targets the view that any substrate running the right program would do. 'Brains are not — or at least not just — computers that are made of meat,' they write. They name anthropomorphism and anthropocentrism as the biases pulling people toward the opposite conclusion, and describe large language models as 'mirrors that reflect an image of our collective digitized past.'

The third reason leans on integrated information theory, the framework long associated with Tononi, which holds that consciousness depends on physical components being wired together in very specific, tightly interconnected ways absent from the layered, feed-forward architectures of today's deep-learning systems. Seven researchers from our Who's Who tracker had already shared the piece by the time this write-up went out, which fits how much the AI-welfare debate has moved from thought experiment into live policy.

The authors conclude that a precautionary approach to AI welfare 'would be misguided,' warning that attributing consciousness to systems that do not have it could pull resources away from beings that demonstrably do.