A groundbreaking study by researchers at Harvard Medical School has found that artificial intelligence (AI) systems can outperform human doctors in diagnosing critically ill patients in emergency situations. The findings, published in the journal Science (journal), are being described by independent experts as a major advancement in clinical reasoning powered by AI.
The study evaluated 76 patients admitted to the emergency department of a Boston hospital. Both AI models and physicians were given identical patient data, including vital signs, brief clinical descriptions, and nurses’ notes. The AI correctly diagnosed conditions in about 67% of cases, compared to 50–55% accuracy among doctors.
Researchers noted that AI performed especially well when decisions had to be made quickly with limited information. When more detailed data was provided, advanced models such as OpenAI o1 reasoning model achieved up to 82% diagnostic accuracy, surpassing doctors, whose accuracy ranged between 70% and 79%.
AI also demonstrated a significant edge in long-term treatment planning. In areas such as antibiotic management and end-of-life care decisions, the AI scored 89%, while physicians scored only 34%.
Despite these results, researchers emphasized that AI is not ready to replace doctors. The study was limited to written clinical data and did not account for physical examinations or visual cues—areas where human physicians excel.
Lead author Arjun Manrai, head of the AI lab at Harvard Medical School, said the findings do not signal the replacement of doctors but rather a transformation in healthcare delivery. Co-author Dr. Adam Rodman of Beth Israel Deaconess Medical Center echoed this view, stating that AI will likely function as a collaborative tool alongside doctors in the coming decade.
One case highlighted in the study involved a patient with a life-threatening pulmonary condition. While doctors initially misinterpreted the situation, the AI identified a critical link to the patient’s history of lupus, leading to the correct diagnosis.
However, concerns remain about the widespread use of AI in clinical settings. A recent survey found that nearly one in five doctors in the United States already use AI for diagnosis. In the United Kingdom, 16% of doctors use it daily, and 15% weekly. The biggest concern among practitioners is accountability in case of errors.
Experts like Professor Ewen Harrison from the University of Edinburgh noted that AI is increasingly serving as a valuable “second opinion” tool. Meanwhile, Dr. Wei Jing of the University of Sheffield warned that overreliance on AI could reduce doctors’ independent critical thinking.
Researchers concluded that while AI shows immense promise, it should not be used as a substitute for professional medical advice. Human judgment remains central, particularly in life-and-death decisions.
