Bayesian Reasoning the Social Domain

Abstract

People are simultaneously attuned to group-based statistics and to norms concerning fairness. While group-based statistics support Bayesian principles, norms concerning fairness underlie egalitarian values. This dissertation examines the judgments people make when Bayesian principles and egalitarian values are both at stake. The findings of three papers reveal that people employ group-based statistics in their social judgments even when they believe it is inappropriate to do so. In Paper 1, people eschewed obvious and relevant statistics in their explicit judgments but hewed with these statistics in their implicit judgments – both before and after individuating information was learned. In Paper 2, people shared less money with a third party who made a judgment that was consistent with established statistics. These people even incurred financial costs on themselves to punish this third party. However, these very same people used statistics as a Bayesian statistician would to render the same judgment that they found morally and intellectually repugnant when offered by someone else. In Paper 3, people continued to rely on statistics even when Bayesian reasoning dictated that these statistics should be ignored. This resulted in not only a statistical error, but also a moral error according to people’s own standards of conduct. These findings suggest that people’s unwitting reliance on statistics may be a barrier to the fairness they desire. This dissertation concludes with a discussion of how psychological science can contribute to newly emerging but pressing issues about the rise of artificial intelligence in what were once exclusively human domains.