When Algorithms Err: Differential Impact of Early vs. Late Errors on Users' Reliance on Algorithms

Abstract

Errors are a natural part of the development and use of predictive algorithms, but they could discourage people from relying on algorithms even when doing so could lead to better decisions. In this paper, we conduct two experiments to demonstrate that people's reliance on a predictive algorithm following a substantial error depends on when the error occurs and how the algorithm is used in decision making. We find that, when the prediction tasks are fully delegated to an algorithm, the impact of an error on reliance is different if the error occurs early versus late. While an early error results in substantial and persistent reliance reduction, a late error affects reliance only temporarily. However, when users have more control over how to use the algorithm's predictions, the risk associated with early errors decreases. Our work advances the understanding of algorithm aversion and informs the practical design of algorithmic decision-making systems.

Keywords: Algorithm, decision support, prediction, timing of error, reliance