Ambiguity and confirmatory reward learning

We tend to interpret feedback in ways that confirm our pre-existing beliefs. Such confirmatory tendencies are often viewed as cognitive flaws, but might have adaptive facets. We propose a novel experimental paradigm and Bayesian computational model to examine how confirmatory inference shapes reward learning when outcomes have ambiguous valence. In these cases, interpretation involves integrating prior beliefs with incomplete evidence, reflecting an inductive bias analogous to missing data imputation. We develop and test this theory using a reward learning task in which information about outcome valence (but not magnitude) is sometimes withheld, allowing for subjective interpretation. Our Bayesian model explains the dynamics of behavior and stated beliefs better than alternative reinforcement learning models. Moreover, stated beliefs about the positivity of ambiguous outcomes are correlated with optimism. Together, these findings demonstrate how confirmatory reward learning can emerge from inference under ambiguity, and may be individually linked to broader dispositional traits.