AI systems outperform doctors in emergency triage diagnoses
Updated
Updated · The Guardian · Apr 30
AI systems outperform doctors in emergency triage diagnoses
10 articles · Updated · The Guardian · Apr 30
In a Harvard study of 76 Boston emergency patients, OpenAI's o1 scored 67% accuracy against doctors' 50%-55%, with results published in Science.
With more detailed records, AI accuracy rose to 82% versus experts' 70%-79%, and it also beat 46 doctors on five treatment-planning case studies, scoring 89% to 34%.
Researchers said AI is better seen as a second-opinion tool because non-text cues were not tested, while experts warned about safety, bias, liability and doctors over-relying on AI.
If AI can spot rare diseases hours early, why does it still fail at basic clinical reasoning?
Who is liable when a diagnostic AI, smarter than a doctor, makes a fatal mistake?