The role of prediction error signals in updating internal models

Flexible behaviour allows us to adapt responses to a dynamic environment1. The framework of predictive processing provides a simple yet powerful way of describing flexible behaviour. Animals detect discrepancies between predicted and observed events, a 'prediction error', which may be used to update the internal model of the world2. The prefrontal cortex represents the state or context in which the animal acts3-6, and may use these representations to compare predictions and observations7,8. However, there are outstanding questions over the nature of the prediction error signal that drives update, and its origin.

Strong evidence implicates the anterior cingulate cortex (ACC) in computing prediction error signals9,10. Prior work from the host lab in mice has found that silencing the ACC during the period of state updating prevents transitions between different learned contexts, identifying single cells whose activity was associated with prediction errors preceding state update11. Thus, the ACC likely plays a causal role in using prediction error signals to drive state update. However, it is important to distinguish prediction error signals from simple surprise signals which do not update an animal's internal model. An example of a surprise signal which has no bearing on an animal's behaviour is found in 'oddball' paradigms, which can induce a 'mismatch negativity' in EEG signals. However, this distinction between informative and uninformative surprise has rarely been made, with mixed results in human imaging studies12,13. By manipulating the reliability of a surprise triggering an update in the task context, a prediction-error may continuously change from highly informative to uninformative of model update. Comparing these conditions would provide a more comprehensive model of prediction error signalling, surprise processing and internal model update.

It also remains to be seen how the ACC fits into the wider network of regions implicated in this behaviour. While...