Preventing algorithm aversion: People are willing to use algorithms with a learning label

Abstract

As algorithms often outperform humans in prediction, algorithm aversion is economically harmful. To enhance algorithm utilization, we suggest emphasizing their learning capabilities, i.e., their increasing predictive precision over time, through the explicit addition of a «learning» label. We conducted five incentivized studies in which 1,167 participants may prefer algorithms or take up algorithmic advice in a financial or healthcare related task. Our results suggest that people use algorithms with a learning label to a greater extent than algorithms without such a label. As the accuracy of advice improves beyond a threshold, the use of algorithms with a learning label increases more than algorithms without a label. Thus, we show that a salient learning attribute can positively affect algorithm use in both the financial and health domain.