CAMI: A Counselor Agent Supporting Motivational Interviewing through State Inference and Topic Exploration

TL;DR

CAMI introduces a STAR-based counselor agent for Motivational Interviewing, integrating client state inference, topic exploration via a hierarchical topic tree, and MI-strategy-guided response generation with ranking. Evaluations on simulated clients show CAMI outperforms several baselines in MI competency, exploration quality, and client experience, with ablations confirming the critical roles of state inference and topic exploration. Expert assessments further corroborate CAMI's relative strength while highlighting remaining gaps to expert human counselors. The work highlights practical paths for scalable, MI-aligned digital counseling and suggests broader applicability to other psychotherapies like CBT, alongside ethical considerations for deployment.

Abstract

Conversational counselor agents have become essential tools for addressing the rising demand for scalable and accessible mental health support. This paper introduces CAMI, a novel automated counselor agent grounded in Motivational Interviewing (MI) -- a client-centered counseling approach designed to address ambivalence and facilitate behavior change. CAMI employs a novel STAR framework, consisting of client's state inference, motivation topic exploration, and response generation modules, leveraging large language models (LLMs). These components work together to evoke change talk, aligning with MI principles and improving counseling outcomes for clients from diverse backgrounds. We evaluate CAMI's performance through both automated and manual evaluations, utilizing simulated clients to assess MI skill competency, client's state inference accuracy, topic exploration proficiency, and overall counseling success. Results show that CAMI not only outperforms several state-of-the-art methods but also shows more realistic counselor-like behavior. Additionally, our ablation study underscores the critical roles of state inference and topic exploration in achieving this performance.

Figures (3)

• Figure 1: The STAR Framework illustrates how an LLM-based counselor agent and a client simulator can be created in an MI-based counseling session. The counselor agent infers the client's state, explores topics that motivate change talk, and utilizes MI techniques to generate appropriate responses. Check marks represent the selected actions, strategies, or states. The client agent may be simulated by adapting the STAR framework, incorporating modules for state transition, dynamic engagement, action selection, and response generation, allowing the client agent to closely align with the client's profile and adapt its engagement level with the counselor accordingly. The "Precon." and "Cont." indicate Precontemplation and Contemplation, respectively.
• Figure 2: The topics tree constructed in our work consists of 5 Super-Class topics (i.e., Health, Economy, Relationship, Law, and Education), 14 Coarse-Grained Topics, and 59 Fine-Grained Topics.
• Figure 3: Topic Exploration Path by the Counselor in the Example.