COMPUTATIONAL MOTIVES AND NEURAL SIGNATURES DISTINGUISH EPISTEMIC INFORMATION SEEKING FROM REWARD SEEKING

Everyday decisions aim not only to earn rewards but also to learn about the world. Across three studies (702 participants), we examined how people gather epistemic information stripped of rewarding value, and compared their strategy with reward seeking in matched conditions. Computational modeling of human behavior revealed a two-stage information-seeking policy, where participants begin by repeatedly sampling each novel option in turn (which we call ‘streaking’) before engaging in uncertainty-guided exploration. While artificial neural networks trained to optimize inference accuracy acquired uncertainty-guided exploration but not early streaking, this two-stage policy improves human inference accuracy under noisy belief updating. During reward seeking, participants switched options after receiving evidence against their current beliefs. Using electroencephalography (EEG) and pupillometry, we found that these reward-guided switches were preceded by strong suppression of alpha-band EEG activity over frontal and centro-parietal regions, and followed by increased pupil-linked arousal, consistent with the adaptive gain theory. By contrast, in the epistemic condition, participants made their first switch only after receiving large belief-confirmatory outcomes that supported their current hypothesis about an option. Only during this early ‘hypothesis-testing’ phase did we observe switch-related alpha suppression and pupil-linked arousal increase. These signatures were absent in the later phase, when switching served to further reduce uncertainty about available options. Together, these findings provide a new perspective on human information seeking. Rather than reflecting a single exploration mechanism governed by adaptive gain, epistemic information seeking unfolds as a structured two-stage process with specific computational motives and neural signatures.