Algorithm Aversion and the Aversion to Counter-Normative Decision Procedures

Abstract According to Norm Theory, decisions that turn out badly result in greater levels of regret if they stemmed from a non-normative decision. Algorithm Aversion (AA) holds that people penalize errors made by algorithms more than errors made by humans. Often, though, studies of AA have explored contexts in which utilizing algorithmic decision-making is unconventional, confounding these two psychological forces. Across five studies, we show that much of what appears as AA can instead be explained by an aversion to counter-normative decision procedures. We find that algorithms are excessively penalized to the extent that using an algorithm to make a forecast is uncommon for that particular domain. In fact, when algorithms are the common decision procedure, we reverse AA and observe a preference for algorithms. Using these insights, we can decompose apparent AA into a combination of an aversion to unconventional decision procedures and a residual aversion to algorithms themselves. Overwhelmingly, the larger effect seems to be an aversion to uncommon decision procedures. This investigation offers insight into the mechanism driving AA, explains why people are sometimes averse to algorithms and other times favor them, and suggests a strategy for increasing the utilization of algorithms to improve wellbeing.