Skip to content

Adaptive Virtual Patient framework: separating behavioral state transitions from language generation

Design · All designs
Evidence · 3 claims (3 for) · 1 study (1 design), q2 · 0 of 1 report an effect size · 3 claims rest on one study

Description

The AVP framework inserts an explicit behavioral-state module between the trainee's turn and the virtual patient's reply: LLM evaluators score empathy and exploration each turn, a dynamics module accumulates the scores into an ordinal disclosure level (G/M/H) using weights derived from a psychotherapy-data SEM, and an LLM generates the next utterance conditioned on that level. The article argues that "Keeping the behavioral logic separate from generation makes the VP’s adaptation inspectable, controllable, and fixed across experimental conditions in a way a prompt-only system cannot achieve." The level determines what the patient may reveal; the LLM determines how they say it.

Design Implications

Context

Requirements

  • Requires an external data-driven model of interaction dynamics (the SEM of Chen et al., 2026) to parameterize the state-update weights
  • Requires validated skill detectors scoring user behaviors each turn

Constraints

  • The dynamics module is parameterized from psychotherapy data and encodes assumptions specific to therapeutic communication; broader-domain claims are not yet validated
  • The deployed 3:1 weight is an integer approximation, not an empirically optimized parameter

Target Learners

  • clinicians and clinical trainees practicing psychotherapy micro-skills

Learning Goals

  • empathic responding
  • exploratory probing
  • recognizing contingent patient disclosure

Claims

Examples

-

Key Sources

  • Angela Chen, Siwei Jin, Catherine Bao, Canwen Wang, Robert E. Kraut, Tongshuang Wu, and Haiyi Zhu. (2026). The Empirically Grounded Adaptive Virtual Patient for Psychotherapy Training: Disclosure That Responds to Therapist Micro-Skills. https://arxiv.org/abs/2606.10051
Page metadata
Field Value
type design
id avp-state-generation-separation-framework
title Adaptive Virtual Patient framework: separating behavioral state transitions from language generation
description The AVP framework inserts an explicit behavioral-state module between the trainee's turn and the virtual patient's reply: LLM evaluators score empathy and exploration each turn, a dynamics module accumulates the score...
status review
generated by process:settle-candidates on 2026-10-10
trust unverified — no one has confirmed this page's content
sources 1 — angela-chen-2026

Frontmatter is the page's machine-readable record — see the Schema & Guide.