The Predictive Brain

Predictive processing, the Bayesian brain and the free-energy principle: the influential idea that the brain constantly predicts its inputs and learns from prediction errors — and what it explains about perception, action and mental illness.

An old idea renewed

In the 19th century Hermann von Helmholtz argued that perception involves unconscious inference: the brain infers the most likely causes of sensory signals using prior knowledge. This idea has been formalised in modern theories of the predictive brain.

Predictive coding

In 1999 Rajesh Rao and Dana Ballard proposed a model of the visual cortex in which higher areas send predictions down to lower areas, and lower areas send up only the prediction errors — the parts of the input that weren't predicted. This explained some puzzling properties of visual neurons and is efficient: expected information needn't be transmitted. Predictive processing generalises this into a theory of brain function, with hierarchies of predictions from simple features up to abstract beliefs.

The Bayesian brain

Bayesian theories describe perception as combining prior expectations with sensory evidence, each weighted by its reliability (precision). When evidence is noisy — fog, darkness — priors dominate; when it's clear, evidence wins. Experiments show people often combine cues close to optimally, as Bayesian models predict.

The free-energy principle

The neuroscientist Karl Friston proposed the free-energy principle: living systems minimise a quantity ("variational free energy") that, roughly, bounds how surprised they are by their sensory inputs. They can do this by updating their model (perception) or by acting on the world to make inputs match predictions (active inference). The principle is ambitious and influential, but critics argue it is hard to test and risks explaining everything and therefore nothing specific.

What it explains

Predictive accounts help explain illusions (the hollow-mask face looks convex because the prior is strong), why self-tickling fails (the cerebellum predicts the sensation), placebo effects, and perceptual phenomena such as "filling in". In psychiatry, hallucinations and delusions have been modelled as overly strong priors or abnormal weighting of prediction errors, and autism has been modelled as giving relatively more weight to sensory evidence — hypotheses being actively tested.