Using Large Language Models to Embed Relational Cues in the Dialogue of Collaborating Digital Twins
Year: 2025
Type: article
Abstract: Embodied Conversational Agents (ECAs) serve as digital twins (DTs), visually and behaviorally mirroring human counterparts in various roles, including healthcare coaching. While existing research primarily focuses on single-coach ECAs, our work explores the benefits of multi-coach virtual health sessions, where users engage with specialized diet, physical, and cognitive coaches simultaneously. ECAs require verbal relational cues—such as empowerment, affirmation, and empathy—to foster user engagement and adherence. Our study integrates Generative AI to automate the embedding of these cues into coaching dialogues, ensuring the advice remains unchanged while enhancing delivery. We employ ChatGPT to generate empathetic and collaborative dialogues, comparing their effectiveness against manually crafted alternatives. Using three participant cohorts, we analyze user perception of the helpfulness of AI-generated versus human-generated relational cues. Additionally, we investigate whether AI-generated dialogues preserve the original advice’s semantics and whether human or automated validation better evaluates their lexical meaning. Our findings contribute to the automation of digital health coaching. Comparing ChatGPT- and human-generated dialogues for helpfulness, users rated human dialogues as more helpful, particularly for working alliance and affirmation cues, whereas AI-generated dialogues were equally effective for empowerment. By refining relational cues in AI-generated dialogues, this research paves the way for automated virtual health coaching solutions.
Source: Systems
Authors: Sana Salman, Deborah Richards
Institution: Macquarie University
Language: English
Topic: Speech and dialogue systems
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,896
OpenAlex ID: W4410121838