Mind, Matter, and Meaning

Anil Seth

Anil Seth studies consciousness through neuroscience, computational modeling, and experiments on perception and self-experience. His work connects predictive processing with the body’s internal regulation. The phrase controlled hallucination summarizes his account of perceptual construction; the beast machine proposal connects conscious selfhood with the processes that sustain a living organism.

Major Works and Research Themes

“Interoceptive Inference, Emotion, and the Embodied Self” (2013) proposes a predictive account of bodily feeling and self-experience. The 2017 TED talk “Your Brain Hallucinates Your Conscious Reality” presents the framework for a broad audience. Being You: A New Science of Consciousness (2021) integrates perception, measures of conscious level, and the biological basis of selfhood.

“Conscious artificial intelligence and biological naturalism,” first published online in 2025, develops his case that consciousness depends on living organization and questions whether computation alone suffices. It considers what more brain-like or life-like artificial systems might require.

These themes involve distinct questions: whether someone is conscious, which contents they experience, and how they experience themselves. A theory of perceptual content is not automatically a full account of conscious level or subjecthood.

Perception as Predictive Inference

Sensory signals underdetermine their external causes. A similar pattern at the retina can arise from different combinations of illumination, surface properties, and viewpoint. Predictive processing explains perception through interaction between expectations about those causes and sensory evidence that corrects them.

In predictive-coding models, generative predictions interact with prediction errors across processing levels. The relative weight assigned to signals depends on their estimated reliability, often called precision. “Inference” here need not involve conscious reasoning or an inner observer: it describes a proposed computational organization of perceptual processing.

An ambiguous sound becoming clearer after one learns what was said illustrates how prior information can change experience. Such demonstrations motivate the role of expectations, but do not uniquely establish every detail of a predictive-coding architecture.

Controlled Hallucination

Seth uses the phrase to emphasize that perception depends on active prediction rather than passively copying sensory input. His stronger claim identifies conscious perceptual contents with the joint content of top-down perceptual predictions: the brain’s best guesses about the causes of sensory signals. This goes beyond saying that expectations sometimes influence perception. In ordinary perception, sensory contact constrains the predictions. Hallucinations involve different relationships between prediction and constraint.

The slogan does not mean that every perceived object is nonexistent or that successful perception and clinical hallucination are identical in every respect. A prediction can concern a real external object and be accurate. However, calling all perception hallucination can obscure those differences, and philosophical critics question what follows from the constructive mechanism to the nature of perceptual contact.

Predictive processing itself does not logically require a mental screen standing between the subject and the world. Representational, enactive, and relational interpretations disagree about that issue while accepting various roles for prediction. Seth’s framing is one account within this debate, not a consensus metaphysical conclusion of neuroscience.

Interoception and the Beast Machine

Interoception concerns the body’s internal physiological condition. Seth proposes that emotions and aspects of embodied selfhood depend on predictive models of internal signals, integrated with vision, touch, proprioception, and other information.

The rubber-hand illusion provides an example of bodily self-experience’s flexibility. When a visible artificial hand and a hidden real hand receive suitably coordinated stimulation, a participant can experience the artificial hand as belonging to their body. Interoceptive and multisensory approaches investigate how reliability, expectation, and bodily signals affect this attribution.

The beast-machine account emphasizes regulation, not just representation of the body as an external object. Keeping physiological variables within viable bounds can explain why bodily feeling has an affective character and why self-experience is tied to an organism’s continued existence.

In his later biological naturalism, Seth argues that consciousness depends on our nature as living organisms, rather than on computation alone. He judges real artificial consciousness unlikely along current AI trajectories, while regarding more brain-like or life-like systems as more plausible candidates. These are his substantive claims, not experimentally established necessities. They do not amount to a restriction to carbon as such, and critics can challenge whether biological organization is required or merely one route to consciousness.

The Real Problem of Consciousness

Seth proposes explaining why particular experiences have the properties they do by building systematic links between phenomenology and mechanisms. He calls this the real problem, distinguishing it from both cataloguing functional capacities and answering why physical processes produce experience at all.

The strategy allows empirical progress without first settling every metaphysical question. Critics maintain that explaining discriminations, reports, or even experiential structure may leave the hard problem untouched. Seth’s response is methodological: detailed explanatory bridges may change what remains puzzling, rather than merely postpone investigation until philosophy is settled.

Evidence, Objections, and Open Questions

Expectations, bodily signals, and sensory reliability demonstrably affect perception. What remains disputed is how broadly predictive architectures explain those effects and what distinctive predictions separate them from competing models. Flexibility becomes a weakness if a model can accommodate every outcome only by changing its assumptions after the fact.

Questions also remain about which predictive processes are conscious, how conscious level relates to content, and whether bodily regulation is necessary for consciousness or especially important for biological selfhood. Seth’s account supplies hypotheses for investigating these questions; it should not be presented as either a complete solution or merely a provocative slogan.

Further Reading

  • Seth, Anil K. “Interoceptive Inference, Emotion, and the Embodied Self.” Trends in Cognitive Sciences 17, no. 11 (2013): 565–573.
  • Seth, Anil. “Your Brain Hallucinates Your Conscious Reality.” TED talk, 2017.
  • Seth, Anil. Being You: A New Science of Consciousness. Faber & Faber, 2021.
  • Seth, Anil K. “Conscious artificial intelligence and biological naturalism.” Behavioral and Brain Sciences. First published online April 21, 2025. doi:10.1017/S0140525X25000032.