Researchers at the University at Buffalo developed an AI called SCAI (Semantic Clinical Artificial Intelligence), which went through all stages of the USMLE medical exam for physician licensure in the United States, achieving record accuracy. On the third, most complex stage, it scored 95.2%—more than most doctors and all previously existing AI.
SCAI uses a database of 13 million validated medical facts and builds logical reasoning chains that enable evidence-based clinical decisions. This is a qualitative step beyond simple information matching toward real clinical reasoning. Importantly, the system relies exclusively on verified research and official medical guidelines.
The developers emphasize that SCAI is not intended to replace doctors, but can serve as an intelligent assistant, especially useful in regions with a shortage of specialists. The project has support from the U.S. Department of Health and already received recognition as the most accurate medical AI system to date. It is available for interaction on the site: halsted.compbio.buffalo.edu/chat.
