AI outperforms doctors in Harvard trial of emergency triage diagnoses
The Guardian • • business
Key Points:
- A Harvard study found that AI systems outperformed human doctors in emergency medicine triage, diagnosing patients more accurately in high-pressure situations using electronic health records.
- The AI model, OpenAI’s o1 reasoning model, achieved a 67% accuracy rate in initial diagnoses compared to 50-55% for human doctors, with even higher accuracy when more patient details were available.
- AI also excelled in creating long-term treatment plans, scoring 89% accuracy versus 34% for human doctors, though researchers emphasize AI currently serves as a second opinion rather than a replacement for clinicians.
- Experts caution that AI lacks the ability to interpret patient distress and visual cues, and concerns remain about accountability, safety, and potential overreliance on AI in clinical decision-making.
- The study signals a shift toward a collaborative "triadic care model" involving doctors, patients, and AI systems, with AI expected to augment rather than replace physicians in the near future.
