Paper Number

ICIS2025-1096

Paper Type

Short

Abstract

Virtual characters powered by generative Artificial Intelligence (AI) offer companionship to capture users’ interest through role-playing experiences. However, maintaining user engagement is challenging, as interactions often become predictable over time. To address this, our study explores how AI companions can evoke user surprise, a psychological reaction triggered by deviations from expectations that can renew user interest. Drawing upon interpersonal circumplex theory, we propose that AI companions' responses, either affiliation-oriented or dominance-oriented, play critical roles in eliciting surprise. Additionally, we theorize that the extent of personality disclosure by AI companions moderates the relationship between these response styles and user surprise. Using a preliminary dataset from a subreddit dedicated to AI characters, we empirically validate our hypotheses. Empirical findings suggest that affiliation-oriented and dominance-oriented responses significantly induce surprise, especially when personality disclosure is limited. This research highlights effective conversational strategies for sustaining user engagement in user-AI interactions, with future directions discussed.

Comments

15-Interaction

Download

Share

COinS
 
Dec 14th, 12:00 AM

When Virtual Characters Surprise: Exploring Conversational Experience with AI Companions

Virtual characters powered by generative Artificial Intelligence (AI) offer companionship to capture users’ interest through role-playing experiences. However, maintaining user engagement is challenging, as interactions often become predictable over time. To address this, our study explores how AI companions can evoke user surprise, a psychological reaction triggered by deviations from expectations that can renew user interest. Drawing upon interpersonal circumplex theory, we propose that AI companions' responses, either affiliation-oriented or dominance-oriented, play critical roles in eliciting surprise. Additionally, we theorize that the extent of personality disclosure by AI companions moderates the relationship between these response styles and user surprise. Using a preliminary dataset from a subreddit dedicated to AI characters, we empirically validate our hypotheses. Empirical findings suggest that affiliation-oriented and dominance-oriented responses significantly induce surprise, especially when personality disclosure is limited. This research highlights effective conversational strategies for sustaining user engagement in user-AI interactions, with future directions discussed.

When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.