Highlights: The predictive brain & how to study consciousness | Anil Seth

Core Hypothesis: Perception as Inference

• Conscious experience is not a transparent readout of external reality but a "controlled hallucination" consisting of the brain's best guess (inference) about the world and the body.
• This view, rooted in Kant's "noumenon" and Hermann von Helmholtz's 19th-century theories, is currently popularized under terms like predictive coding, predictive processing, active inference, and the Bayesian brain.
• Sensory signals are used to calibrate top-down predictions; consequently, what we experience is underpinned by internal predictions rather than bottom-up sensory data.
• Anil Seth argues that color is a primary example of this, where the brain adjusts perception of an object (e.g., a white paper) based on contextual ambient light rather than raw photonic input.
• Individual differences in perceiving context (as seen in the "dress" illusion) stem from variations in how brains weigh ambient light data during inference.

Physiological Regulation and Emotional Valence

• Unlike external senses, the brain does not generate a spatial map of internal organs; instead, it monitors physiological homeostasis (e.g., blood pressure, heart rate) to ensure survival.
• The brain cares about internal organs for "control," not for spatial identification, explaining why humans lack specific sensory awareness of organs like the spleen.
• Emotional experiences are reinterpreted as variations of physiological valence: positive valence corresponds to successful regulation, while negative valence (e.g., anxiety, disappointment) indicates a departure from the effective physiological set-point.
• Prediction serves a functional imperative to control and regulate the internal milieu, making the brain a "prediction machine" evolved for survival.

Psychedelics as Data on Construction

• Psychedelic experiences do not reveal a "truer" reality but serve as data points demonstrating that normal perception is a constructed inference.
• Pharmacologically, these substances (e.g., LSD) act on serotonin receptors, hypothesized to disrupt the predictive inference process.
• The resulting phenomenology—hallucinations and loss of voluntary control—occurs when top-down predictions overwhelm sensory data, turning "controlled hallucinations" into "uncontrolled" ones.
• Computational models are being developed at Imperial College to simulate how predictive networks break down under psychedelics, aiming to distinguish these effects from hallucinations in Parkinson's or dementia.
• The exact mechanism linking serotonin receptor inhibition to the disruption of predictive coding remains an open question requiring further granular phenomenological study.

Neural Correlates and Brain Architecture

• Consciousness is not located in a single "plug" or area; damage to lower brainstem regions (specifically midline thalamic nuclei) causes a total loss of consciousness (coma), analogous to unplugging a power source.
• Approximately three-quarters of the brain's 86 billion neurons, located in the cerebellum, are not directly implicated in consciousness despite their massive quantity and role in motor control and cognition.
• Consciousness is determined by organizational dynamics and network interactions between areas, not merely by neuron count or local functional specialization.
• Higher-order brain regions become increasingly multifunctional ("pluripotent"), requiring an analysis of how areas interact rather than isolating specific regions.

Methodological Challenges in Research

• Standard experimental paradigms like binocular rivalry and visual masking are used to isolate neural correlates of consciousness by varying experience while holding sensory input constant.
• Binocular Rivalry: Presents different images to each eye, causing conscious experience to oscillate between them while the stimulus remains static.
• Visual Masking: Prevents a target image from being consciously perceived despite sensory reception, allowing researchers to compare brain activity when a stimulus is seen versus unseen.
• A critical limitation is the "report problem": differences in neural activity often reflect the capacity for report or access rather than the conscious experience itself.
• Researchers attempt to mitigate this with "no-report" protocols, yet debates persist regarding whether these methods fully decouple the capacity for reporting from the phenomenon of consciousness.
• Seth notes that correlations are not explanations; identifying neural correlates often fails to distinguish between consciousness and associated cognitive processes like attention or access.