A groundbreaking study published in JAMA Network Open reveals that enhanced artificial intelligence (AI) can now outperform most physicians on the U.S. Medical Licensing Examination (USMLE). Using advanced training techniques, researchers developed an AI system that scored higher than human doctors and existing AI tools—marking a major milestone in medical AI.
AI in Healthcare: The Rapid Rise
Since ChatGPT’s public launch in 2022, AI has exploded into healthcare:
18% of healthcare professionals now use AI for biomedical research (Statista, 2024).
20% employ AI chatbots to answer patient questions.
Hospitals worldwide are adopting AI for:
- Patient monitoring (43%)
- Medication management (37%)
- Radiology and diagnostics (36%)
Leading institutions like the Mayo Clinic, Johns Hopkins, and Mass General Brigham are already testing AI for:
- Early detection of cancer, depression, and heart disease
- Summarizing patient records
- Predicting surgical risks
- Automating anesthesia dosing
How AI “Learns” Medicine
Large language models (LLMs) like Google’s Gemini, Meta’s Llama, and OpenAI’s ChatGPT don’t “think” like humans—they predict based on patterns in massive datasets. While impressive, standard AI has limitations:
- Relies on educated guesses rather than true reasoning.
- Can “hallucinate” incorrect answers.
- Needs constant updates to stay accurate.
The Fix? Retrieval-Augmented Generation (RAG)—a method that lets AI pull real-time data from medical journals, research, and clinical databases before answering.
The Study: Supercharging AI for Medicine
Researchers developed SCAI (“sky”), an AI enhancement that:
- Feeds AI the latest medical knowledge using RAG.
- Structures data as “semantic triples” (subject-relation-object) for deeper understanding.
- Tests performance on USMLE exams.
Results:
- SCAI-boosted AI (Llama 2 and Llama 3 models) outscored most physicians on all USMLE steps.
- Reduced errors and provided citable sources, increasing transparency.
What This Means for Doctors and Patients
AI won’t replace doctors—but doctors using AI may replace those who don’t. Key takeaways:
- AI as an assistant: Helps with diagnostics, paperwork, and staying updated.
- Better accuracy: Enhanced models like SCAI reduce mistakes.
- The future: Human-AI collaboration will likely become standard in clinics.
Next steps: Expanding SCAI to more AI models and real-world clinical trials.
Why It Matters
This isn’t just about tests—it’s about saving lives. With AI handling routine tasks, doctors can focus on complex care and patient relationships. As lead author Dr. Peter Elkin notes:
“Improving AI with targeted clinical knowledge is critical for its safe, effective use in healthcare.”
The age of AI-powered medicine has arrived—and it’s only getting smarter.
Related topic:
