Guardian article
AI outperforms doctors in Harvard trial of emergency triage diagnoses
https://www.theguardian.com/technology/2026/apr/30/ai-outperforms-doctors-in-harvard-trial-of-emergency-triage-diagnoses
Partially translated with GPT-5.5 Thinking.
A Harvard study found that an AI system outperformed human doctors in the high-pressure situation of triaging emergency room patients, making more accurate diagnoses when people first arrived at the hospital.
The researchers reported in a paper published in Science that large language models (LLMs) "exceeded most benchmarks in the field of clinical reasoning."
https://www.science.org/doi/10.1126/science.adz4433
One experiment involved 76 patients who arrived at the emergency room of a Boston hospital. Both an AI and two human doctors were given the same standard electronic health record (EHR) to read. This record typically included vital sign data, demographic information, and a few sentences from nurses briefly describing why the patient came to the hospital. The AI made accurate or very close diagnoses in 67% of cases, outperforming the human doctors whose accuracy rates were between 50% and 55%.
The experiment highlighted the strengths of AI, particularly in triage situations where quick decisions must be made based on limited information. When more detailed information was provided, the AI's diagnostic accuracy (specifically o1, an inference model from OpenAI) rose to 82%. Specialists' accuracy ranged from 70% to 79%, but this difference was not statistically significant.
The AI also outperformed a larger number of human doctors in tasks such as suggesting antibiotic regimens and planning end-of-life procedures, which involve developing long-term treatment plans. The AI and 46 doctors were asked to review five clinical case studies, and the computer generated significantly better plans than the human doctors who relied on existing resources like search engines. The AI scored 89%, while the human doctors scored only 34%.
However, the research team emphasized that emergency room doctors are not about to be replaced. This study only compared the performance of humans and AI based on patient data that could be conveyed in text. Factors such as a patient's level of pain or outward appearance, which AI cannot currently perceive, were not included in the test. This means that the AI functioned more like a consulting clinician who reviews records rather than directly examining patients in person.
Arjun Manrai, co-lead author of the study and director of the AI lab at Harvard Medical School, explained: "We don't think our findings mean that AI will replace doctors." He added, "Rather, we believe they show that we are at a pivotal moment for a very significant technological shift in medicine."
Adam Rodman, MD, a physician at Beth Israel Deaconess Medical Center (where the study was conducted) and another co-lead author, called large language models "one of the most impactful technologies in decades." He predicted that over the next 10 years, AI will not replace doctors but rather become part of a new "triad model" of care involving doctors, patients, and AI systems.
[The rest of the content is omitted]
