In real-world test, AI model did better than ER doctors at diagnosing patients (www.npr.org)

A groundbreaking study from researchers at Harvard Medical School and Beth Israel Deaconess Medical Center reveals that an AI model developed by OpenAI outperformed emergency room doctors in diagnosing patients. In tests involving real patient cases, including one with a pulmonary embolism linked to a previously undiagnosed history of lupus, the AI's diagnostic accuracy surpassed that of two experienced physicians, utilizing only electronic health records and the limited information available at the time. This study, published in *Science*, highlights the AI's ability to effectively analyze complex medical data and generate accurate diagnoses in the chaotic environment of emergency medicine. The significance of these findings lies in the considerable advancements in AI technology, particularly in its ability to deal with real-world data—a task that previous models struggled with. While the study underscores AI's potential to significantly enhance diagnostic processes, experts caution that these models should not replace human clinicians. Instead, the findings advocate for rigorous testing and trial design to integrate AI responsibly into clinical workflows. As Dr. David Reich notes, the technology is "possibly ready for prime time," sparking discussions on how to utilize AI to improve patient care without undermining the nuanced human judgement critical in clinical settings.