Medical Education and AI Standardized Patients

AI standardized patients, learner-centered simulation, and benchmarked practice environments for medical education.

Medical education and human-agent interaction
AI standardized patients EasyMED SPBench Medical learning Co-design
EasyMED AI-SP Paper Co-design Paper All Projects
Medical education and automated patient evaluation

This project asks how AI can become a reliable practice partner for medical learners. The goal is not only to make a model talk like a patient, but to make standardized-patient training scalable, controllable, repeatable, and evaluable.

Background and Motivation

Standardized patient training is expensive to scale

Human standardized patients are valuable, but scheduling, case coverage, training consistency, and feedback quality are hard to maintain across many learners.

LLMs can simulate conversation, but education needs more

A useful AI patient must preserve clinical persona, reveal information gradually, respond to learner behavior, and support structured feedback.

Medical learners need safe repetition

Students need repeated practice in history taking, communication, empathy, and clinical reasoning before facing high-stakes clinical settings.

Core Ideas

Case
Build controllable patient cases

Define patient background, symptoms, hidden findings, emotional stance, and disclosure rules so the AI patient behaves consistently across sessions.

Dialogue
Make the patient interactive, not scripted

The simulated patient should respond to learner questions, resist over-disclosure, and expose missing history-taking skills.

Rubric
Connect interaction to evaluation

Practice conversations are paired with rubrics and benchmarks so the system can measure information coverage, reasoning quality, communication, and safety.

Design
Co-design with medical learners

The CHI-facing work studies how students experience AI patients, what feels different from human SPs, and how interfaces should support learning.

Typical Work

EasyMED
Human or LLM as Standardized Patients?

Introduces EasyMED and SPBench, comparing AI standardized patients with human standardized patients in medical education scenarios.

Paper
CHI
It Talks Like a Patient, But Feels Different

Co-designs AI standardized patients with medical learners and surfaces design requirements around realism, agency, feedback, and trust.

Paper
Workflow
Doctor-centric workflow-aligned tasks and benchmarks

Connects medical education and AI evaluation to the real workflows doctors use for documentation, reasoning, and decision support.

Paper

Display Figures

Automated medical evaluation signal
Medical education scenarios need both realistic conversations and structured evaluation signals.
Medical benchmark overview
The AI-SP line connects to the broader medical benchmark stack, including knowledge, multimodal, and workflow-aligned evaluation.

Resource Map

EasyMED

Repository for AI standardized patient simulation and related medical education resources.

Repository
Medical Evaluation Benchmarks

Companion project for medical QA, multimodal medical AI, diagnosis, live clinical benchmarks, and doctor workflows.

Project page
HuatuoGPT Ecosystem

The medical model family that supplies the language, vision, reasoning, and deployment backbone for medical AI scenarios.

Project page
Back to Projects