Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Virtual Agents for Alcohol Use Counseling: Exploring LLM-Powered Motivational Interviewing

Ian Steenstra, Farnaz Nouraei, Mehdi Arjmand, Timothy W. Bickmore

TL;DR

This work investigates the viability of large language model (LLM)–driven virtual agents to conduct motivational interviewing (MI) for alcohol-use counseling. By implementing a prompt-engineered MI dialog system atop GPT-4 and pairing it with a web-based humanoid avatar (Dr. Anderson), the study evaluates linguistic soundness, safety, MI competency, and expert-perceived effectiveness through three empirical studies. Results show the LLM-powered agent can match or exceed human counselors on MI adherence, complex reflections, and client-facing outcomes, while offering advantages in empathy and accessibility, though challenges remain in psychoeducation and planning for sustained change. The findings support the potential of LLMs as scalable, initial-access interventions that can complement traditional care, with clear design and safety considerations for responsible deployment in mental-health services.

Abstract

We introduce a novel application of large language models (LLMs) in developing a virtual counselor capable of conducting motivational interviewing (MI) for alcohol use counseling. Access to effective counseling remains limited, particularly for substance abuse, and virtual agents offer a promising solution by leveraging LLM capabilities to simulate nuanced communication techniques inherent in MI. Our approach combines prompt engineering and integration into a user-friendly virtual platform to facilitate realistic, empathetic interactions. We evaluate the effectiveness of our virtual agent through a series of studies focusing on replicating MI techniques and human counselor dialog. Initial findings suggest that our LLM-powered virtual agent matches human counselors' empathetic and adaptive conversational skills, presenting a significant step forward in virtual health counseling and providing insights into the design and implementation of LLM-based therapeutic interactions.

Virtual Agents for Alcohol Use Counseling: Exploring LLM-Powered Motivational Interviewing

TL;DR

This work investigates the viability of large language model (LLM)–driven virtual agents to conduct motivational interviewing (MI) for alcohol-use counseling. By implementing a prompt-engineered MI dialog system atop GPT-4 and pairing it with a web-based humanoid avatar (Dr. Anderson), the study evaluates linguistic soundness, safety, MI competency, and expert-perceived effectiveness through three empirical studies. Results show the LLM-powered agent can match or exceed human counselors on MI adherence, complex reflections, and client-facing outcomes, while offering advantages in empathy and accessibility, though challenges remain in psychoeducation and planning for sustained change. The findings support the potential of LLMs as scalable, initial-access interventions that can complement traditional care, with clear design and safety considerations for responsible deployment in mental-health services.

Abstract

We introduce a novel application of large language models (LLMs) in developing a virtual counselor capable of conducting motivational interviewing (MI) for alcohol use counseling. Access to effective counseling remains limited, particularly for substance abuse, and virtual agents offer a promising solution by leveraging LLM capabilities to simulate nuanced communication techniques inherent in MI. Our approach combines prompt engineering and integration into a user-friendly virtual platform to facilitate realistic, empathetic interactions. We evaluate the effectiveness of our virtual agent through a series of studies focusing on replicating MI techniques and human counselor dialog. Initial findings suggest that our LLM-powered virtual agent matches human counselors' empathetic and adaptive conversational skills, presenting a significant step forward in virtual health counseling and providing insights into the design and implementation of LLM-based therapeutic interactions.
Paper Structure (15 sections, 1 figure, 3 tables)

This paper contains 15 sections, 1 figure, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Virtual Agents for Substance Use Counseling
  4. Virtual Agents for Motivational Interviewing
  5. Modeling Counselor Dialog using Large Language Models
  6. An LLM-powered Virtual Agent for MI Counseling
  7. LLM Dialog System for MI Counseling
  8. LLM-Powered Virtual Agent
  9. Empirical Evaluations
  10. Study 1: Comparison of LLM and Human Counseling Moves on Linguistic Soundness and Safety
  11. Study 2: Comparison of LLM and Human Counseling Moves on MI Competency
  12. Study 3: Expert Evaluation of LLM Virtual Agent Counselor
  13. Overall Discussion
  14. Limitations
  15. Future Work

Figures (1)

  • Figure 1: MI Competence Comparison of Composite Measures (LLM-Powered Virtual Agent vs. Basic Therapeutic Threshold)