Design and Preliminary Evaluation of a Generative AI Role-Play System for Practicing Cognitive Empathy

Open Access
Article
Conference Proceedings
Authors: Tsubasa Kishiume, Taiyo Kojima, Toshikazu Kato, Toru Nakata, Takashi Sakamoto
Abstract

When another person's psychological state is not fully verbalized, understanding that person requires active questioning and interpretation. We designed a generative AI role-play system in which the AI portrays a junior colleague who does not readily disclose their feelings. The system combines a predefined persona, rules governing psychological information disclosure, and three dialogue tasks: exploration, interpreting, and paraphrasing. In a preliminary evaluation with 16 participants, illustrative dialogues from two conditions showed both task-aligned communication and instances in which advice or directive statements replaced the intended dialogue behaviors. The mean change in cognitive empathy (CE) was +0.438 in the cognitive-empathy condition and +0.094 in the factual–situational condition. However, baseline scores differed between conditions, and the baseline-adjusted analysis did not show a difference in the same direction. This study presents a concrete design for practicing perspective-taking through dialogue and identifies response-contingent support for users' dialogue behaviors as a direction for future system development.

Keywords: Generative AI, Cognitive Empathy, Role-Playing, Dialogue Training, Human–AI Interaction

DOI: 10.54941/ahfe1008384

Cite this paper
Downloads
0
Visits
1
Download PDF

More from this volume

← Detection-Aware Generation of Fatigue-Conditioned Ocular and Head Behaviors for Virtual DriversImproving accessible mobility for visually impaired travelers: An IPA analysis of the sightseeing verbal map in Shanghai →
View all articles in Human Factors in Design, Engineering, and Computing →