The Brain’s Balance Between Prediction and Reality

Predictive processing frameworks propose that the brain is fundamentally a generative model, continuously issuing top-down predictions and updating them against bottom-up prediction errors. Psychiatric and neurological disorders can be reframed as systematic failures of this inferential process. A type one error — inferring signal where there is none — maps onto positive symptoms such as hallucinations and delusions. A type two error — suppressing genuine signal — maps onto neglect syndromes and dissociative states. The pathology, on this account, is not in perception per se but in the calibration of inference.

The hollow mask illusion provides a clean empirical probe. Neurotypical observers cannot sustain the percept of an inverted hollow face; the prior probability of convex faces is so strong that it overrides veridical depth cues. Individuals with schizophrenia frequently resist this illusion, perceiving the hollow correctly. This suggests their generative models assign relatively higher precision — epistemic weight — to sensory evidence than to prior beliefs, consistent with the hyper-salience accounts of psychosis.

Autism spectrum conditions illuminate the other pole of the same axis. Elevated sensory precision means prediction errors are rarely attenuated, producing a perceptual environment of relentless, unfiltered signal. Repetitive and ritualistic behaviors can be understood as active strategies to minimize surprise by engineering a maximally predictable sensory context. The unifying principle is that psychiatric conditions dysregulate the precision weighting that mediates the trade-off between model complexity and sensory accuracy — the fundamental dial of the Bayesian brain.