Researchers build AI patient avatars for real-time ACT therapy training feedback

Training psychotherapists in evidence-based interventions like Acceptance and Commitment Therapy (ACT) requires repeated practice with meaningful feedback, yet opportunities are limited by ethical and resource constraints. To address this, researchers developed a system that uses large language models to drive embodied virtual patient avatars. These avatars simulate behavior based on profiles derived from real therapy sessions and configurable clinical scenarios. A separate automated evaluator provides turn-by-turn feedback on therapist responses using established ACT fidelity criteria. The system is designed not to replace supervision, but to support deliberate practice through experimentation and reflection in a low-risk environment.

Expert evaluation with practicing psychologists confirmed high realism in patient behavior and showed that immediate turn-by-turn feedback increased therapists' awareness of intervention choices. Quantitative evaluation across 49 therapy transcripts identified GPT-4o-mini as the optimal feedback model, achieving the lowest mean absolute error (MAE = 6.12) in replicating human supervisor ACT fidelity ratings with statistically significant agreement. This work demonstrates that fidelity-aware simulated patients can serve as a scalable complement to traditional psychotherapy training, potentially expanding access to high-quality supervision and accelerating therapist skill development.