To err is human, not algorithmic – Robust reactions to erring algorithms
Laetitia A. Renier, Marianne Schmid Mast, Anely Bekbergenova
Abstract
When seeing algorithms err, we trust them less and decrease using them compared to after seeing humans err; this is called algorithm aversion. This paper builds on the algorithm aversion literature and the third-party reactions to mistreatment model to investigate a wider array of reactions to erring algorithms. Using an experimental design deployed with a vignette-based online study, we investigate gut reactions, justice cognitions, and behavioral intentions toward erring algorithms (compared to erring humans). Our results show that when the error was committed by an algorithm (vs. a human), gut reactions were harsher (i.e., less acceptance and more negative feelings), justice cognitions weaker (i.e., less blame, less forgiveness, and less accountability), and behavioral intentions stronger. These results remain independent of factors such as the maturity of the algorithms (better than or same as human performance), the severity of the error (high or low), and the domain of use (recruitment or finance). We discuss how these results complement the current literature thanks to a robust and more nuanced pattern of reactions to erring algorithms.