Name: Valentin Wyart (French National Institute of Health and Medical Research (Inserm), Ecole Normale Supérieure - PSL University, Paris, France)

Title: Zeroing in on human information seeking

Abstract: Seeking information is a pervasive drive in human cognition under uncertainty. However, in most laboratory settings, information seeking is either traded off against reward seeking (as in bandit tasks) or confounded with changes-of-mind about the best option (as in reversal learning tasks). In this talk, I will present two experimental paradigms that we designed to isolate information seeking from these confounding factors. In the first, we manipulated controllability in a hidden-state inference task to dissociate information seeking from changes-of-mind. Combining human magnetoencephalography (MEG) and pupillometry, we found that information seeking is associated with boosted neurophysiological signatures of attention and arousal. In the second, we compared reward seeking and information seeking in a sequential sampling task with matched sensory evidence. Computational modeling of behavior revealed specific changes in both inference and sampling policy during information seeking: more accurate belief updating, and a two-stage sampling policy in which uncertainty-guided exploration is preceded by streaks of each novel option. Electroencephalographic (EEG) recordings further showed that neural representations of information updates (but not those of reward updates) reflect a nonlinear interaction between prior beliefs and incoming evidence – a coding scheme that emerges in recurrent neural networks (RNNs) trained to minimize epistemic uncertainty. Together, these findings provide new insights into the specific signatures and drivers of human information seeking.

Website / links: @valentinwyart(.bsky.social)