AI and the World's Hard Problems · Medicine & drug discovery · Nature (computational biology) · 2026-10-23
From simulation to pedagogy: multi-model and randomized evaluation of structured AI standardized patients
Standardized patients for clinical communication training are costly, hard to scale, and rely on trained actors. The team built LLM-driven AI standardized patients that adjust information disclosure to learner skill, and expert assessment of 350 simulated consultations from five LLMs found learner skill explained more performance variation than model choice. In a pilot randomized study of 58 participants, OSCE gains with AI training did not differ significantly from those with human standardized patients.