Pupil dilation offers a time-window on prediction error

Task-evoked pupil dilation is notably linked to unexpected events. Building on Zénon’s (2019) information-theory framework, we investigated whether the pupil’s response to feedback on decision outcomes during associative learning reflects a prediction error signal. Operationally, we defined prediction errors as an interaction between stimulus-pair frequency and accuracy. We then tested if these signals correlated with information gain, formally defined as the Kullback-Leibler (KL) divergence between posterior and prior belief distributions of an ideal observer. We reasoned that information gain should be proportional to the precision-weighted prediction error signals potentially arising from neuromodulatory arousal networks. We analyzed two data sets in which participants performed perceptual decision-making tasks while pupil dilation was recorded. Our findings consistently showed that a significant proportion of variability in the post-feedback pupil response was explained by information gain shortly after feedback presentation. For the first time, we present evidence that whether the pupil dilates or constricts along with information gain was context dependent. This study offers empirical evidence that the pupil’s response provides valuable insights into the process of model updating during learning, highlighting its utility as a physiological indicator of internal belief states.