Modeling Multi-Party Interaction in Couples Therapy: A Multi-Agent Simulation Approach

TL;DR

This work addresses the gap in training clinicians for multi‑party dynamics in couples therapy by introducing a multimodal, multi‑agent simulation with two LLM‑powered virtual patients and a stage controller that operationalizes six recurrent stages and the demand–withdraw pattern. The authors use theory, expert interviews, and corpus analysis to ground the design and verify its relevance, then conduct a within‑subjects evaluation with 21 licensed therapists showing higher perceived realism and clearer recognition of stage‑ and cycle‑specific behaviors compared with a baseline. Key contributions include a stage‑grounded multi‑agent framework, a theory‑ and data‑driven foundation for agent behaviors and transitions, and empirical evidence of training benefits focused on in‑session dynamics such as problem raising, escalation, and de‑escalation. The system demonstrates potential to enhance therapist training by providing realistic, controllable, low‑stakes practice for managing multi‑party interactions and emotional dynamics in real time, paving the way for further integration with feedback mechanisms and multi‑session curricula.

Abstract

Couples therapy, or relationship counseling, helps partners resolve conflicts, improve satisfaction, and foster psychological growth. Traditional approaches to training couples therapists, such as textbooks and roleplay, often fail to capture the complexity and emotional nuance of real couple dynamics. We present a novel multimodal, multi-agent simulation system that models multi-party interactions in couples therapy. Informed by our systematic research, this system creates a low-stakes environment for trainee therapists to gain valuable practical experience dealing with the critical demand-withdraw communication cycle across six couple-interaction stages. In an evaluation study involving 21 US-based licensed therapists, participants blind to conditions identified the engineered agent behaviors (i.e., the stages and the demand-withdraw cycle) and rated overall realism and agent responses higher for the experimental system than the baseline. As the first known multi-agent framework for training couples therapists, our work builds the foundation for future research that fuses HCI technologies with couples therapy.

Figures (6)

• Figure 1: A three-step process for identifying design goals and strategies in couples therapy simulation.
• Figure 2: Example conversation snippets for each stage from an actual simulation session
• Figure 3: A screenshot of the interface showing full features of our system
• Figure 4: Realism outcome ratings by VP system (baseline vs experimental). Boxes show the interquartile range (25th–75th percentiles); solid line is the median; whiskers extend to 1.5×IQR; points denote outliers. Ratings were on a 1–5 Likert scale, with higher values indicating more pronounced outcomes. Medians are labeled above each box.
• Figure 5: Ratings of Training Effectiveness for the Two Conditions ($p < .001$)