JMIR Publications

ABSTRACT

Background:

Obesity is a common, serious, and costly chronic disease, which poses a significant public health problem in the United States and Canada. Current clinical practice guidelines recommend providers to augment longitudinal care of people living with obesity with consistent messaging to support the development of self-efficacy and motivation, which is directly in line with the goals of motivational interviewing (MI), a client-centered yet directive counseling style. However, the expensive and time-consuming training of healthcare providers in MI and the lack of trained counselors are major barriers to the widespread adoption of MI in clinical practice. AI counselors accessible via the Internet to deliver MI-based weight loss counseling can help circumvent these barriers and complement human counselors in the healthcare delivery continuum.

Objective:

The primary objective is to test the feasibility of conducting unscripted counseling with Neural Agent for Obesity Motivational Interviewing (NAOMI), a deep learning-based Web application for MI-style weight loss counseling. The secondary objective is to evaluate the acceptability, usability, and ability of NAOMI to shift motivational precursors in a study with overweight and obese patients of primary health clinics.

Methods:

NAOMI will be developed based on recent advances in deep learning in 4 stages. In the first two stages, NAOMI will be implemented using an open-source foundation large language model (LLM) and: 1) few-shot learning based on a prompt with task-specific instructions and 2) domain adaptation strategy based on fine-tuning LLM using a large corpus of psychotherapy and MI session transcripts. In the remaining two stages, we will refine the best of these two approaches. Each version of NAOMI will be evaluated using a mixed-methods approach, in which 10 adults (18-65 years), who meet the criteria for overweight or obesity, interact with NAOMI and provide feedback. NAOMI’s fidelity to the MI framework will be assessed using the MI Treatment Integrity (MITI) scale. Participants’ perceptions of AI conversational agents in general and NAOMI specifically will be assessed via pre- and post-interaction questionnaires. The motivational precursors, including participants’ confidence, importance, and readiness for changing their lifestyle will be measured pre- and post-interaction, and at a one-week follow-up. Participants will also rate NAOMI’s usability and empathetic skills post-interaction via questionnaire-based assessments along with providing feedback about their experiences with NAOMI and suggestions for its improvement in qualitative interviews. Framework matrix analysis will be used for the analysis of qualitative interviews.

Results:

The first version of NAOMI has been developed. The investigative team is currently developing a Web site for the study visit. Participant recruitment will commence in May 2024. Data collection activities are expected to conclude in March 2025.

Conclusions:

If proven effective, neural counseling agents can have a broad, transformative effect on MI and other psychotherapeutic modalities.