An AI tool grounded in evidence-based medicine outperforms other AI tools—and most doctors—on USMLE exams

**AI Tool Scores Higher Than Most Doctors on US Medical Licensing Exams**

A clinical artificial intelligence tool created by researchers at the University at Buffalo has achieved high accuracy on all three sections of the United States Medical Licensing Exam (USMLE). The findings, published in *JAMA Network Open*, reveal that the AI system outperformed other AI tools and surpassed the performance of most human test-takers.

The research highlights the tool’s ability to excel in evidence-based medicine, a critical component of medical decision-making. The USMLE consists of three “Step” exams designed to assess medical knowledge, clinical reasoning, and patient care skills. According to the study, this AI system demonstrated consistent accuracy across all parts of the exam. Researchers attribute its success to its grounding in evidence-based methodologies, which may have contributed to its superior performance compared to both other AI systems and many practicing physicians.

Newsflash | Powered by GeneOnline AI

Source: https://medicalxpress.com/news/2025-04-ai-tool-grounded-evidence-based.html

Author

Mark Chiang