Towards believable and educational conversations with virtual patients

Virtual Reality (VR) technology allows the design and application of realistic but adaptive learning environments in medical education. In particular, virtual patient systems have logistical and methodological advantages compared to non-computerized interventions. However, evidence for their effectiveness is fragmented as any educational domain introduces its requirements regarding learning goals, measurements of learning outcomes, and application design. In this context, we present preliminary results of evaluating a VR training application for conducting a clinical interview to diagnose mental disorders in children and adolescents using virtual patients. The evaluation focuses on design elements related to the virtual patient’s appearance and natural language capabilities. Our results indicate that our virtual patient design is highly believable and that our dialog system is satisfying. However, conversational flow requires optimization. We discuss design directions and potential enhancements for learner-virtual patient interactions in VR and address future operations to evaluate the effectiveness of our approach.