Source: The Guardian

A Harvard-led trial reported that AI systems outperformed human doctors in emergency-medicine triage by diagnosing more accurately in high-pressure first-contact situations. One experiment at a Boston hospital involving 76 emergency patients gave an AI and two physicians the same electronic health record; the AI produced an exact or very close diagnosis in 67% of cases versus 50–55% for the doctors. The advantage was clearest when rapid decisions were required with minimal information; the AI — identified as OpenAI’s o1 reasoning model — reached 82% accuracy when more detail was available, compared with 70–79% for expert humans (the latter difference was not statistically significant). The study’s authors also reported an evaluation where the AI and 46 doctors reviewed five clinical case studies: the AI scored 89% for treatment planning against 34% for clinicians using conventional resources.

Researchers and commentators emphasized key limitations and cautions. The trials compared AI and humans using textual patient data from electronic health records, not visual cues or assessments of patient distress, so the AI acted more like a second-opinion tool rather than a full clinical replacement. Lead author Arjun Manrai said the findings mark a profound technological change, and co-author Adam Rodman predicted a future “triadic care model” of doctor, patient and AI. The paper includes a case in which the AI correctly identified lupus-related lung inflammation missed by human clinicians. The article also cites survey data on clinician AI use (nearly one in five US physicians use AI to assist diagnosis) and records expert warnings about accountability gaps, potential clinician over-reliance, uneven performance across patient groups, and the absence of evidence that AI is yet safe for routine clinical deployment.