Predictive processing
Predictive processing frames the brain as a hierarchy of prediction machines. It is one of the most influential bridges between neuroscience and AI today.
Key takeaways
- The brain is modelled as constantly generating predictions and updating on prediction errors.
- Karl Friston's free-energy principle is the most prominent formal version.
- Self-supervised learning in modern AI shares the same core intuition.
Core idea
Rather than passively responding to inputs, cortex is thought to generate predictions about expected inputs and adjust internal models to minimise prediction error. Perception, action, and learning all fall out of this single principle in many formulations.
Why it matters for AI
Self-supervised pretraining — the engine behind modern large language and vision models — also learns by predicting missing or future signal. Predictive processing offers a unified theoretical lens for why this works.
Frequently asked questions
Is predictive processing proven?
+
It is a powerful framework with strong supporting evidence at multiple levels, but 'proven' is too strong — alternative formulations remain active in the literature.