Information provision measures for voice agent product recommendations— The effect of process explanations and process visualizations on fairness perceptions

Abstract

AbstractWhile voice agent product recommendations (VAPR) can be convenient for users, their underlying artificial intelligence (AI) components are subject to recommendation engine opacities and audio-based constraints, which limit users’ information level when conducting purchase decisions. As a result, users might feel as if they are being treated unfairly, which can lead to negative consequences for retailers. Drawing from the information processing and stimulus-organism-response theory, we investigate through two experimental between-subjects studies how process explanations and process visualizations—as additional information provision measures—affect users’ perceived fairness and behavioral responses to VAPRs. We find that process explanations have a positive effect on fairness perceptions, whereas process visualizations do not. Process explanations based on users’ profiles and their purchase behavior show the strongest effects in improving fairness perceptions. We contribute to the literature on fair and explainable AI by extending the rather algorithm-centered perspectives by considering audio-based VAPR constraints and directly linking them to users’ perceptions and responses. We inform practitioners how they can use information provision measures to avoid unjustified perceptions of unfairness and adverse behavioral responses.