Algorithm Aversion: Evidence from Ridesharing Drivers

Abstract

The low rate of adoption by human users often hinders AI algorithms from achieving their intended efficiency gains. This is particularly true for algorithms that prioritize system-wide objectives because they can create misalignment of incentives and cause confusion among potential users. We provide one of the first large-scale field studies on algorithm aversion by leveraging an algorithmic recommendation rollout on a large ridesharing platform. We identify contextual experience and herding as two important factors that explain ridesharing drivers’ aversion to an algorithm that is designed to help drivers make better location choices. Specifically, we find that drivers are less likely to follow the algorithm when the algorithmic recommendation does not align with their past experience at a given location-time unit and when their peers’ actions contradict the algorithmic recommendations. We discuss the managerial implications of these findings. This paper was accepted by Catherine Tucker, Special Issue on The Human-Algorithm Connection. Funding: The research at Shanghai Jiaotong University was supported by the National Natural Science Foundation of China [Grants 72202135, 72110107001, 72231003]. S. Zhang acknowledges the support of Shanghai Pujiang Program [Grant 21PJC070], and Special Fund for Creative Research Groups [Grant 72221001]. Y. Zhu acknowledges the support of National University of Singapore [Grant WBS A-8000489-00-00]. Supplemental Material: The online appendix and data are available at https://doi.org/10.1287/mnsc.2022.02475 .