Predictive Processing × Memory — Cross-Links

Have you ever wondered why memory feels more like reconstruction than replay — why the same event can be remembered slightly differently each time you recall it? You may recognise the pattern. Predictive processing and memory research have quietly been converging on the answer, and this short piece is a bridge to the memory science collection.

The traditional folk model treated memory as storage — an event happens, it gets recorded, and later it gets replayed. That model was durable but incomplete. Decades of research have shown that memory is more accurately described as reconstruction — the brain rebuilds an approximation of the original event each time it is retrieved, using a combination of stored traces and current context (Schacter, 2012) [Supported Finding].

What predictive processing adds to this picture is the mechanism. In the predictive frame, memory retrieval is itself a prediction — the nervous system generating a best-guess forecast of what the remembered event contained, based on the traces it has stored and everything else it currently knows. The prediction is then compared, as best it can be, against whatever additional cues are available. What consciousness gets is the resulting reconstruction.

This explains several observations that pure storage models struggle with. It explains why memory is malleable — because it is being reconstructed each time, current context and current concerns leak into the reconstruction. It explains why witnesses to the same event can produce different accounts — different priors produce different reconstructions from the same underlying trace. And it explains why deliberately recalling a memory can, in some conditions, subtly change what is stored about it, a phenomenon called memory reconsolidation.

The reconsolidation finding is worth naming because of what it suggests for therapy. If memory is not stored once and then retrieved intact, but rather reconstructed each time and briefly re-encoded, then the specific conditions under which a memory is recalled can shape what it will be next time. This is the mechanism behind several evidence-based approaches to trauma — the memory is retrieved in the safety of a therapeutic context, updated with new emotional information, and re-encoded in a slightly different form.

Where the two fields most productively meet is on the question of what learning actually is. Predictive processing suggests learning is the updating of priors from prediction error. Memory research suggests learning is the modification of stored traces. On the current picture, both descriptions point at roughly the same underlying process — the changes to neural connection strengths that let the nervous system generate more accurate predictions next time.

For a reader who has been thinking in predictive processing terms, memory research is the natural companion collection. The prior you carry into any moment is, in a real sense, the accumulated result of every prior memory the system has laid down. Understanding how those memories were built, and how they change, deepens the predictive picture.

For the foundation, see The Brain Is a Prediction Machine. For where priors themselves live, Priors — The Brain's Working Assumptions is the companion piece.