Search papers, labs, and topics across Lattice.
The study introduces the Adaptive Virtual Patient (AVP), a simulated patient that dynamically adjusts its level of disclosure based on the therapist's micro-skills during psychotherapy training. By utilizing a structural equation model derived from nearly 2,000 hours of real-world therapy transcripts, the AVP effectively quantifies the relationship between therapist empathy, exploration, and patient openness. Evaluation results show that the AVP's adaptive disclosure significantly improves in response to therapist skills, outperforming traditional prompt-only systems and confirming the importance of exploration in driving patient engagement.
The Adaptive Virtual Patient adapts its disclosure behavior in real-time, revealing that therapist exploration is key to enhancing patient engagement during training.
Simulated patients offer a scalable way to train psychotherapy micro-skills such as empathic responding and exploratory probing, but current systems either follow fixed scripts or rely on LLMs that drift unpredictably over long sessions. We present the Adaptive Virtual Patient (AVP), which adapts its disclosure behavior -- from guarded, through moderate openness, to full disclosure -- in response to trainee skill. The AVP is grounded in a structural equation model fit to nearly 2{,}000 hours of real-world psychotherapy transcripts, which quantifies how therapist empathy and exploration shift a patient's openness over time. An LLM generates the AVP's utterances conditioned on a disclosure level that the dynamics module updates each turn. In an evaluation with 20 clinicians and trainees over 80 sessions (1{,}033 turns), the AVP's disclosure rises in response to therapist empathy and exploration, while a prompt-only baseline stays flat; ablations confirm that the empirically motivated parameterization outperforms alternatives, with exploration carrying most of the adaptive signal.
