Abstract:

Effective counselor and therapist training relies on supervised practice in clinical interviewing, yet many programs face resource limitations in providing structured, high-fidelity training opportunities. Advances in large language models (LLMs) and voice-enabled AI offer scalable alternatives through interactive, AI-driven simulations. However, previous research has found that pre-LLM chatbots have been largely ineffective in providing emotional responses or maintaining natural dialogue, often forcing trainees to adhere to rigid scripts.

This study addresses these limitations by developing silicon clients, AI-driven simulated patients created using psychological vignettes of individuals presenting with Major Depressive Disorder (MDD). These profiles were integrated into ChatGPT-4o’s advanced voice chat feature, chosen for its ability to detect emotional tone and nonverbal cues, enabling more realistic, dynamic conversational exchanges between practitioner and client. Graduate trainees conducted two 15-minute structured clinical interviews with the AI, with an experimental group receiving AI-generated feedback between sessions. Performance was assessed through self-reports from trainees, expert clinician evaluations of interview transcripts, and an analysis of AI-generated feedback. The self-reports capture perceived improvements in clinical confidence, while expert evaluations assess rapport-building, structured questioning, and active listening skills. Clinicians also evaluate the AI’s effectiveness as a simulated patient, examining its credibility, adaptability, and ability to provide useful feedback. This multi-method approach ensures a comprehensive assessment of AI’s potential to supplement traditional counselor training.

Our findings contribute to the growing field of AI-driven simulation-based training, with implications for healthcare education, telehealth training, and human-AI interaction design. By integrating generative AI into clinical training pipelines, institutions can provide realistic, on-demand practice opportunities that complement existing supervisory methods. Future work will explore adaptive AI models capable of personalized feedback, real-time scenario adjustments to further enhance training efficacy, and AI’s ability to display other psychological disorders.

Keywords: AI;EDUCATION;FEEDBACK;HEALTH CARE;MEDICAL MODELING AND SIMULATION;PERSONALIZED TRAINING;TRAINING

Menu

Michael Devotta