Paper Type
Complete
Paper Number
PACIS2025-1945
Description
AI aversion is widely observed in evaluative contexts. Drawing on role theory, this study explores the impact of AI’s communication styles when delivering subjective and negative evaluation information on user acceptance. We conducted two experiments in the context of facial diagnosis. We discover when AI uses an indirect (versus direct) communication style to convey negative information about one’s facial features, user acceptance will be enhanced; whereas such effect is non-significant when the information is conveyed by a human. This effect is due to users’ distinctive role perceptions of AI and humans. Specifically, users tend to regard AI as an assistant (low-power role) that meets specified needs, while considering humans as experts (high-power role) who offer advice. We also find users’ role perception of AI can be altered through interventions and the effect of communication styles will change accordingly. Our findings provide practical implications for optimizing AI’s natural language interaction design.
Recommended Citation
Li, Mengxin; Zou, Lili Wenli; Yi, Cheng; and Wang, Harry, "Indirect Communication in Human-AI Interaction: Experiments with an AI Facial Diagnosis System" (2025). PACIS 2025 Proceedings. 18.
https://aisel.aisnet.org/pacis2025/general_topic/general_topic/18
Indirect Communication in Human-AI Interaction: Experiments with an AI Facial Diagnosis System
AI aversion is widely observed in evaluative contexts. Drawing on role theory, this study explores the impact of AI’s communication styles when delivering subjective and negative evaluation information on user acceptance. We conducted two experiments in the context of facial diagnosis. We discover when AI uses an indirect (versus direct) communication style to convey negative information about one’s facial features, user acceptance will be enhanced; whereas such effect is non-significant when the information is conveyed by a human. This effect is due to users’ distinctive role perceptions of AI and humans. Specifically, users tend to regard AI as an assistant (low-power role) that meets specified needs, while considering humans as experts (high-power role) who offer advice. We also find users’ role perception of AI can be altered through interventions and the effect of communication styles will change accordingly. Our findings provide practical implications for optimizing AI’s natural language interaction design.
Comments
General