Abstract

To act efficiently in our environment, we must accurately predict forthcoming events, such as the consequences of our actions. This not only requires maintenance of a predictive model, but sometimes also flexible adaptation of this model so that it stays up to date and thus useful. A core assumption is that prediction errors drive this adaptation. Such learning from prediction errors is well established, but nearly exclusively investigated in the context of reward and punishment.

However, there is only a limited number of human actions where reward constitutes the immediate feedback; in other words: there is a wide range of external events beyond reward which we can learn to predict and can use to evaluate the correctness of our actions. In fact, we can even achieve this when external feedback is absent or delayed, making use of our internal confidence to assess whether we’re right or wrong.

This complexity raises a lot of questions: does prediction-error coding for external events rely on the same neural structure as reward prediction error coding? Is the neural response to unexpected events modulated by our beliefs about their significance? And how does our estimate of the correctness of our behaviour, our confidence, influence how we process and judge what we perceive?

I will argue that the basal ganglia, typically associated with reward-prediction error processing, fulfil a similar function in coding unexpected perceptual events, driving the adaptation of predictions. I will further present fMRI & EEG data which suggests that expectations and beliefs about our environment influence adaptation as well as the prediction error signal itself. Lastly, I will discuss new results which show that we learn and adapt predictions on how confident we are in specific situations – without external feedback. This learned confidence influences neural signatures of stimulus processing and behavioural adaptation, and colours how we judge our own performance.