Two distinct and separable processes underlie individual differences in algorithm adherence: Differences in predictions and differences in trust thresholds.

Abstract

Algorithms play an increasingly ubiquitous and vitally important role in modern society. However, recent findings suggest substantial individual variability in the degree to which people make use of such algorithmic systems, with some users preferring the advice of algorithms whereas others selectively avoid algorithmic systems. The mechanisms that give rise to these individual differences are currently poorly understood. Previous studies have suggested two possible effects that may underlie this variability: users may differ in their predictions of the efficacy of algorithmic systems, and/or in the relative thresholds they hold to place trust in these systems. Based on a novel judgment task with a large number of within-subject repetitions, here we report evidence that both mechanisms exert an effect on experimental participant's degree of algorithm adherence, but, importantly, that these two mechanisms are independent from each-other. Furthermore, participants are more likely to place their trust in an algorithmically managed fund if their first exposure to the task was with an algorithmic manager. These findings open the door for future research into the mechanisms driving individual differences in algorithm adherence, and allow for novel interventions to increase adherence to algorithms.