The Best Doctors Don't Have Better Answers

A patient recently came into my office and sheepishly asked if we could discuss the AI review of his case. He happened to be a physician himself and had done exactly what millions of patients are now doing. He entered his symptoms into an AI platform, received a thoughtful explanation of the possible diagnoses, and came prepared with questions. But physicians are doing exactly the same thing. Platforms such as OpenEvidence and UpToDate AI have quickly become part of many clinicians’ daily workflow.

In fact, what made me smile is that I had also used AI to review aspects of their case, so now I was curious to compare notes. What surprised me was how different our notes were. Neither of us was right or wrong. We just approached the problem very differently. His list of what might be wrong with him was fairly broad: nutritional deficiencies or an autoimmune disorder. This isn’t what I was concerned about at all. By that point, I had already identified the one abnormal finding that concerned me most and was using AI to think through the most appropriate next step. This is probably my most common use for AI platforms, to determine what tests to order when a potential diagnosis lies outside my expertise as a pulmonologist.

In neither case had the AI had done a poor job. I just realized the questions we had asked of it were very, very different. My patient had asked what was most likely causing his symptoms. It was a perfectly reasonable place to start. Yet it wasn’t the question occupying my mind. By the time I turned to AI, I had already spent considerable time deciding what the problem actually was. I wasn't asking it to diagnose my patient. I was asking a much narrower question: given this particular abnormal finding, what should I do next?

That encounter stayed with me because it mirrored something I had been discovering in my own use of artificial intelligence. Like many physicians, I’ve gradually incorporated AI into my daily work. I use it to summarize papers, explore unfamiliar topics, challenge my thinking, and occasionally to help organize complex clinical problems. I remain amazed by how these systems retrieve information instantly, synthesize enormous bodies of literature, and often produce thoughtful, well-reasoned responses. More than once, I’ve found myself thinking, “That’s an excellent answer.” But I've noticed something about my own conversations with AI. I rarely stop after the first answer. The first answer almost always leads me to another question.

Suppose I ask AI why a patient is short of breath. It will generate a careful differential diagnosis with appropriate supporting evidence. But if I then ask what diagnosis would be catastrophic to overlook, the emphasis changes. If I ask what findings don’t fit the leading diagnosis, I receive a different analysis. If I ask what assumptions I’m making, or what additional history would most change the differential diagnosis, the conversation evolves yet again. Nothing about the patient has changed. The only thing that has changed is the question.

We often think expertise is measured by knowledge. Medical education certainly reinforces that belief. Students memorize anatomy, physiology, pharmacology, pathology, and thousands of clinical facts. Licensing examinations reward correct answers, and board certification reinforces the importance of mastering an ever-expanding body of information. It’s easy to conclude that experienced physicians are simply people who know more than everyone else. But after two decades in practice, I no longer think that’s what distinguishes the best clinicians.

After two decades in practice, I’ve come to think we’ve misunderstood whta expertise actually is. The best doctors don’t necessarily have better answers. They ask better questions. That realization may sound deceptively simple, but I think it lies at the heart of clinical reasoning. Experienced physicians are constantly reframing the problem in front of them. What diagnosis explains all of the findings rather than just most of them? What would I regret missing? Why doesn’t this piece of the story fit? What additional information would change my thinking? Every answer generates another question, and every question reshapes the differential diagnosis. Medicine is rarely a straight path toward certainty. More often, it is a process of gradually reducing uncertainty while always remaining open to the possibility that we are wrong.

The emerging literature is fascinating. Large language models perform extraordinarily well on standardized medical examinations and increasingly sophisticated diagnostic reasoning tasks. But they also reveal something important. Their answers depend heavily on the information they're given and the way the problem is framed. In other words, they excel at answering questions that are well posed. AI answers the questions you ask. Physicians spend years learning to recognize the question they should have asked.

I do think that AI has ultimately made me a better physician. Ironically not because it has given me better answers, but because it has encouraged me to ask more questions. When an answer seems convincing, I now find myself immediately asking what alternative explanations deserve consideration, what evidence argues against my leading diagnosis, and what assumptions I may not even realize either of us are making. AI has become a catalyst for intellectual humility. It has reminded me that confidence and correctness are not the same thing.

Increasingly, I believe expertise will not be defined by possessing more information. It will be defined by knowing how to frame the problem. Artificial intelligence has made knowledge more accessible than at any point in history. But clinical judgment is shaped by something more than information alone. It develops through years of caring for patients, seeing the same disease present in countless different ways, making difficult decisions with incomplete information, and carrying the memory of the cases that changed how we think. Whether AI will someday acquire something analogous to that experience remains to be seen. For now, it has reminded me that the questions we ask are often more important than the answers we receive.