Dr. Andrew Parsons, a physician and medical educator at the University of Virginia, writes that AI models have reached high accuracy on diagnostic benchmarks, citing a study in which OpenAI’s o1 model scored 78% on complex cases and outperformed experienced doctors on real emergency-room diagnoses. He argues, however, that deciding what to do about a diagnosis—weighing treatment options against a patient’s individual circumstances, values, and risk tolerance—remains a distinctly human skill he calls ‘management reasoning.’ Using an example of two patients with identical prostate cancer diagnoses who need different management approaches, Parsons contends current AI cannot replicate this personalized judgment.
