ClientBot: A conversational agent that supports skills practice and feedback for Motivational Interviewing for AUD

点击翻译按钮获取中文摘要

PROJECT SUMMARY/ABSTRACT Millions of Americans are in need of evidence-based counseling, such as motivational interviewing (MI), for alcohol use disorders (AUDs) each year. To develop competence in an evidence-based practice like MI, trainees require ample opportunities for practice and immediate, performance-based feedback on the skills that they are learning. However, this is challenging if not impossible to offer at scale -- to the large number of providers in need of training. Opportunities for practice typically rely on roleplays with other trainees with limited experience, and feedback requires either direct supervision from an expert trainer or behavioral coding from a trained coding team; these are costly, limited, and time consuming. AI-based technology can meet this need, generating many opportunities for practice, and providing regular, actionable feedback. Many practice opportunities coupled with rapid, performance-based feedback can enhance and expand training in evidence-based counseling for AUDs in a scalable and cost-efficient manner. Lyssn.io?, Inc., (“Lyssn”) is a start-up developing AI-based technologies to support training, supervision, and quality assurance of evidence-based counseling. Our goal is to develop innovative health technology solutions that are objective, scalable, and cost efficient. ?Lyssn’s? team includes expertise in natural language processing, machine learning, user-centered design, software engineering, and clinical expertise in evidence-based counseling. Previous research demonstrated the basic utility of a prototype conversational agent (ClientBot) for training counselors. Currently, ClientBot simulates a general mental health client who can engage in open-ended interaction with trainees and provides immediate, performance-based feedback to trainees using machine learning. The current Fast-Track SBIR proposal partners ?Lyssn? with Prevention Research Institute (PRI), who has a long track-record of training counselors in evidence-based approaches for AUD and currently trains approximately 1,250 counselors per year. Phase I will adapt ClientBot to an AUD training context, including understanding PRI training workflows, assessing usability, and accuracy of machine learning based MI feedback. Phase II will conduct a field-based usability trial and a randomized training trial (N = 200 PRI trainees) to evaluate the effectiveness of ClientBot on learning of MI skills compared to a wait-list and PRI training-as-usual. Analyses will also examine the hypothesized mechanisms of behavior change underlying ClientBot’s MI skills training. The successful execution of this project will break the reliance on role plays with peers and human judgment for training and performance-based feedback and support commercialization of a ClientBot product for training of AUD counselors in evidence-based practices.