Uncertainty-Based Reasoning Constrains Human Information Demand

Effective decisions require not only choosing between alternative options but selecting evidence that is informative for those options. Humans often deviate from strategies that maximize information gain, but the mechanisms of these deviations are not understood. We tested adult participants in a task in which they chose which question to inspect to prepare for a test. Participants received two alternative questions and could identify the more informative question based on its uncertainty (Unc) and probability of appearing at test (PTest). Despite the simplicity and transparency of the task, participants systematically deviated from the normative strategy. Computational modeling showed that most participants underweighted Unc and PTest and relied more strongly on PTest vs Unc. Strikingly, a substantial minority systematically preferred theleast-informative question, suggesting that they incorrectly reasoned about how uncertainty impacted informativeness. Moreover, in the absence of external feedback, sub-optimal strategies were amplified over time. The findings highlight the importance of abstract reasoning, task understanding, and self-reinforcing learning in constraining human information demand.