Can You Train Better Priors?

Have you ever had the experience of realising, months into a new situation, that a whole class of moments you used to find difficult had quietly stopped being difficult — without you consciously working on them? That quiet shift is one of the honest signatures of priors updating. This piece is about how that mechanism works and what actually moves it.

The short answer to the question is yes. Priors are trainable. Because they are built out of experience, repeated experiences of a different kind rewrite them (Clark, 2013; Feldman Barrett, 2017) [Supported Finding]. This is not a hopeful metaphor. It is what the research consistently suggests, and it is the mechanism underneath much of what effective therapy, deliberate practice, and ordinary maturation are doing.

What actually moves a prior is worth being specific about. Priors update on prediction error — the difference between what the nervous system expected to happen and what actually happened. A single large prediction error rarely rewrites a well-established prior. A pattern of consistent prediction errors, held over time, does. This is why single insights, however striking, tend not to change behaviour by themselves. The insight arrived. The prior did not yet have enough new data to update.

This is also why deliberate practice works. When you rehearse a new response to a familiar situation, you are supplying the nervous system with prediction errors — moments in which the old prediction expected one outcome and a different outcome actually occurred. Each such moment is a small vote against the old prior. Enough votes, over enough time, and the prior updates.

A few honest observations about the timeline. Priors that were laid down early in life, over many years of consistent experience, tend to be more resistant to updating than priors formed more recently or in fewer instances. This is not because early priors are magically privileged. It is because they carry more accumulated evidence, and displacing them requires more accumulated new evidence. The difference between a prior formed at age six and one formed at age twenty-six is often measured in the amount of counter-evidence needed to move it.

What this suggests practically is patience with the mechanism. Change that seems slow is often the correct pace for the underlying update. A person practicing a new emotional response for two weeks may not see a change because two weeks is not enough new evidence to move the old prior. Two years of consistent practice is a different quantity of evidence, and the update, when it comes, tends to arrive not as a dramatic breakthrough but as a quiet realisation that the old moment has stopped happening.

One useful specific practice is to notice, in the moment, when a prediction feels wrong to the situation you are actually in. That noticing itself is a form of prediction error being flagged. Over time, repeated flagging accumulates the evidence the nervous system uses to update.

The most useful thing to take from this is that priors are not personality. They are working models built from experience, and experience keeps arriving. What you keep experiencing quietly becomes what you expect.

For the underlying mechanism, see Prediction Error — The Currency of Learning. For where priors sit in the frame, Priors — The Brain's Working Assumptions is the companion piece.