AI outperforms doctors in Harvard trial of emergency triage diagnoses (www.theguardian.com)

A recent Harvard study has revealed that AI systems, particularly OpenAI's o1 reasoning model, significantly outperformed human doctors in emergency triage diagnoses. In trials involving 76 patients, the AI accurately identified diagnoses 67% of the time, compared to human doctors' success rate of 50-55%. The study, published in *Science*, suggests that large language models (LLMs) have surpassed previous clinical reasoning benchmarks, especially in high-pressure situations where quick decisions are required. When provided with more detailed information, the AI's accuracy rose to 82%, showcasing its potential as an effective tool for diagnosis and treatment planning. While this progress is notable, researchers caution against the notion that AI will replace human doctors. The study focused on data communicated through text, meaning the AI did not assess non-verbal cues such as a patient's appearance or distress level. As these technologies evolve, experts envision a collaborative model merging AI with human expertise in a triadic care approach involving the doctor, the patient, and the AI. The findings underline both the promise and challenges of integrating AI into clinical practice, necessitating careful consideration of accountability and the limitations of AI in medical decision-making.