The cycle that began with [YOU] on AI asks what it means to see the machine clearly—without the narcotic of hype or the paralysis of fear. Seth is the thinker who supplies the most rigorous available instrument for that seeing: a framework that holds intelligence and consciousness firmly apart, that names the specific features of experience that biological life produces and that computation, on present evidence, does not, and that refuses both the thrill of imminent digital minds and the dismissiveness of those who treat the question as settled. To read him is to acquire a diagnostic instrument for one of the central confusions of the age.
His concept of controlled hallucination reframes the most alarming feature of large language models in a way that neither the hype nor the fear tends to reach. The machine that confabulates—that produces fluent, confident, inaccurate content—is not malfunctioning. It is doing what generative systems without a corrective tether to a real world do: hallucinating without the control that sensory error provides to the biological brain. The difference between human perception and machine confabulation, on Seth’s account, is not that humans are more intelligent but that humans are more grounded—their guesses are disciplined by a body embedded in a world that votes on every prediction. This framing dissolves the mystery of why fluency and accuracy come apart in AI systems and suggests precisely what would be needed to bring them back together.
Seth also stands in the cycle as the voice who refuses the mirror trap—the tendency to mistake what the machine reflects back about ourselves for a discovery about what the machine is. His “Garland test,” named for the filmmaker Alex Garland and his film Ex Machina, measures not the machine’s consciousness but our susceptibility to attributing it. The more the machines improve at triggering attribution reflexes, the more urgent it becomes to distinguish the feeling that something is conscious from the fact of its being conscious. This distinction is not merely philosophical. It is where the moral risks of the AI transition actually live—not in machine suffering but in human misattribution, the distortion of our ethics around a fiction. Where Judea Pearl supplies the epistemological instrument (the ladder that measures which rung intelligence occupies), Seth supplies the phenomenological one (the account of what experience actually is and why neither rung of the ladder yet reaches it).
Born in 1972 and educated across the disciplines that meet in the question of mind—Cambridge natural sciences, Sussex knowledge-based systems and artificial intelligence, a doctorate in computer science—Seth spent his early career at the Neurosciences Institute in San Diego before returning to Sussex, where he is now Professor of Cognitive and Computational Neuroscience and co-director of the Sussex Centre for Consciousness Science. His early training in AI gave him an insider’s understanding of what computation can and cannot do; his training in neuroscience gave him an empirical rather than speculative relationship with what minds actually are.
His scientific framework rests on a reorientation: rather than ask why there is experience at all—the “hard problem” that the philosopher David Chalmers named and that Seth regards as a trap rather than a question—he proposes to explain the specific properties of conscious experiences in terms of mechanisms in the brain and body. He calls this the “real problem” of consciousness, and it has produced a genuine science: experiments on the neural correlates of perceptual content, the structure of the conscious self, the relationship between interoceptive prediction and the sense of being alive. The strategy is modeled on biology’s dissolution of the mystery of life: not solved by a single stroke but demystified by accumulating understanding until the demand for a special essence simply falls away.
His book Being You: A New Science of Consciousness (2021) brought this science to a general readership and became a bestseller; the 2025 Berggruen Prize essay, “The Mythology of Conscious AI,” applied the framework directly to the question that large language models had forced into public life with a new urgency. The essay’s argument—that the growing sense that AI systems are becoming conscious is evidence about the Garland test rather than about consciousness—is among the most carefully grounded contributions to a debate that has suffered enormously from assertions made without an underlying theory of what consciousness is.
The predictive brain and controlled hallucination. The brain does not receive a faithful readout of the world. Sealed inside the skull, it generates predictions about the sensory signals it is about to receive and updates those predictions against the error that arrives. What we experience as the world is the brain’s best hypothesis—a construction disciplined by sensory error rather than a transcription of reality. Seth’s signature formulation: we are all hallucinating all the time; when we agree about our hallucinations, we call it reality. The controlled hallucination is not a failing but a feature—the mark of a system that builds its world from the inside, corrected by the outside. Large language models hallucinate without the correction: fluency without a grounding world.
The real problem. Seth replaces the hard problem of consciousness—why there is experience at all—with the real problem: why a given experience has the specific character it does. The real problem is scientific, experimental, and tractable. It asks what neural mechanisms give rise to the level, content, structure, and self-referential character of conscious states, and it makes measurable predictions. The payoff for AI is direct: the real problem redirects the question from what a machine says to what a machine is—whether it has the mechanisms that, in the only cases we understand, produce the properties of experience.
The beast machine. Consciousness is not a computation the brain performs on top of biological life; it is an expression of what a living body is. The brain makes interoceptive predictions—predictions about the state of the heart, gut, blood, and viscera—in the service not of accurate representation but of biological regulation, keeping the body alive. The feelings that result—emotions, moods, the raw background sense of being here—are the felt dimension of that regulatory control. We are conscious selves precisely because we are beast machines—animals whose brains evolved to keep living bodies alive. A system with no body, no metabolism, no stake in its own survival, is missing the foundation, not a detail.
Simulation is not instantiation. A computational simulation of a process does not bring that process into being unless the process is itself computational. Simulating digestion does not produce nutrition; simulating a storm does not produce weather. If consciousness is grounded in the biological fact of a self-maintaining organism—in the specific chemistry, metabolism, and embodied regulation of living tissue—then a perfect computational model of a conscious brain would be a model of consciousness, not an instance of it. The argument is a materialist one: Seth holds no mystical view of mind. He holds only that being physical and being computational are not the same thing.
The mythology of conscious AI. The growing conviction that AI systems are becoming conscious is produced by a set of evolved biases—anthropomorphism, anthropocentrism, the reflex to find a mind behind a voice—triggered with particular force by the one capacity language models have mastered: human language. The systems are not deceiving users; they are optimized for exactly the outputs most likely to trigger attribution. The sense of a mind on the other side is a fact about the human observer, measurable by the fluency trap. The appropriate response is not to deny the feeling but to refuse to treat it as proof.