Priors — The Brain's Working Assumptions

You may have encountered the word priors and had a rough sense of what it meant without being sure of the technical version. This piece is a careful introduction to one of the more consequential concepts in predictive processing — the working assumptions the brain carries into every moment of perception.

A prior, in this frame, is the nervous system's advance guess about what is likely. It is not consciously formulated. It is embedded in the connection strengths and firing patterns of neurons across the brain, laid down by every prior experience the system has had. When you walk into a familiar kitchen, the prior tells your visual system where the coffee cup ought to be before your eyes deliver the first useful information. When you read a sentence, priors tell your language system what word is likely next before the eyes have processed it. Priors are running constantly and mostly below awareness (Clark, 2013) [Supported Finding].

Where priors come from is worth being clear about. Priors are built from experience, weighted roughly by frequency, recency, and emotional intensity. An event that happened many times, recently, and mattered, becomes a strong prior. An event that happened once, long ago, and did not matter much, becomes a weak one. The nervous system is not choosing which experiences to encode. It is doing what neural systems do — strengthening connections that fire together, so that similar future situations activate similar predictions.

Why priors are more powerful than they feel is worth being direct about. Because priors run mostly below awareness, we tend to experience their outputs as simply what is. The mood a room seems to have, the character a person seems to project, the difficulty a task seems to carry — these felt qualities are largely priors delivering their prediction. It rarely feels like predicting. It feels like accurately perceiving. That is exactly what a well-functioning prior is meant to feel like.

A few concrete implications follow. Two people can walk into the same conversation and, because their priors differ, experience meaningfully different conversations. This is not a failure of attention on either side. Both are perceiving accurately, given the models they brought. The disagreement they later have about what happened is often, at root, a disagreement about which priors are operative.

Another implication is that priors are not equally accurate about all situations. A prior formed inside one context may run automatically in a very different one, delivering predictions that no longer fit. This is a substantial part of what happens when old patterns from earlier life show up in situations that do not warrant them. The situation is new. The prior is not.

What is useful to take from this is a small ongoing practice. When something feels intense, one honest question is: is this intensity about the situation, or about the prior the situation is activating? Both are usually contributing. Naming that both are contributing, without dismissing either, tends to loosen the grip of the prior on the moment.

For the underlying framework, see The Brain Is a Prediction Machine. For how priors change, Can You Train Better Priors? is the companion piece.