Naive Utility Calculus underlies the reproduction of disparities in social groups

On the road to a more fair and just world, we must recognize ubiquitous disparities in our society, but awareness alone is not enough: Observed disparities between groups often get wrongly attributed to inherent traits (e.g., African Americans are disproportionately arrested because they are more prone to crime), creating a self-perpetuating feedback loop. As shown in a past study (Meng & Xu, 2020), such reasoning can result from the Naive Utility Calculus (Jara-Ettinger et al., 2016): If an agent knows a target trait's "hit rate'' in every group and avoids unnecessary sampling, it is rational to infer that groups sampled from more often have higher hit rates. The previous study used non-social categories (robot chickens) as stimuli, which raises the question of whether the results generalize to the social domain. In the current study, we replicated past findings using novel social groups (aliens): Overall, people were more likely to check groups examined more often by the agent but when observed hit rates did not support the agent's sampling behavior, people incorporated both information sources to infer group hit rates. This work brought NUC-based models one step closer towards tackling disparities in the real world consisted of social groups