AMIE
2025
Towards Conversational Diagnostic Artificial Intelligence (Articulate Medical Intelligence Explorer)
In 159 text consultations with validated patient-actors, an LLM-based conversational diagnostic AI (AMIE) improved diagnostic accuracy and rated consultation quality compared with primary care physicians.
Result
Primary endpoint: Diagnostic accuracy and performance on clinically meaningful axes (history-taking, diagnosis, management, communication, empathy) rated by specialist physicians and patient-actors
AMIE showed greater diagnostic accuracy and was superior on 30 of 32 axes rated by specialist physicians and 25 of 26 axes rated by patient-actors; the abstract reports no effect sizes.
- Subtopic
- Standalone consultation
- Design
- Randomised, double-blind crossover OSCE-style study, scenarios randomised, Canada/UK/India
- Relevance
- Important: one of several pillars.
- Source
- Tu T, et al. Nature. 2025;642(8067):442-450.doi 10.1038/s41586-025-08866-7
Discussion & critique
The first randomised, blinded head-to-head of a conversational AI against physicians across history-taking, diagnosis and empathy, a milestone, and the AMIE system was later tested in specialist care (O'Sullivan LLM (cardiology)). Heavily caveated by its own authors: synchronous text chat is unfamiliar in clinical practice, the patients were actors, and nothing was tested in real-world care.