Building virtual patients for training mental health professionals
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
We present a framework for developing “virtual patients” to augment training for Mental Health Professionals (MHPs) with a process that is more scalable and systematic than current practice which relies on human role-play for the training and evaluation of patient interaction. We show how to combine large language models, retrieval-augmented personification (a novel variant of retrieval-augmented generation), and custom code-based logic to create a psychology engine that simulates realistic patient responses by emulating several key psychological mechanisms: short- and long-term memory, varying levels of conscious awareness about topics (as well as modulation of such awareness), and dynamic mood states where attitudes toward topics of conversation evolve over the course of the dialogue. We also describe algorithms for creating realistic patients with coherent symptom profiles and backstories. Taken together, these tools produce a realistic training partner for an MHP, enabling both training-at-scale as well as automated evaluation of specific skill sets. We discuss how our psychology engine framework makes training qualified MHPs more efficient and scalable, facilitates the continuing education needed as potential new treatments such as psychedelics emerge from clinical trials.
