Series: The Patient, the Physician, and the Machine
Part 2 of 3
Every generation of medicine has a moment when the profession is forced to confront a difficult truth about itself.
Sometimes that truth is scientific, when we realize a treatment we believed in doesn’t work. Sometimes it’s cultural, when a hierarchy we inherited is not serving patients as well as we thought. And sometimes it is technological: a tool arrives that does something we believed was uniquely ours.
AI is creating that moment for physicians.
I recently wrote about the trust paradox in healthcare: patients are uneasy about AI, but they are already using it to fill the gaps our system leaves behind. That is the patient side of the story. Now we need to talk about the physician side.
The current debate around AI in medicine sometimes defaults to a familiar phrase: physician supervision. On the surface, that sounds reasonable. Medicine is high-stakes, AI can be wrong, and patients can be harmed. Of course physicians should be involved.
But we should be careful here. “Physician supervision” may sound safe, but it can also become a container for imprecise thinking. It borrows from models we already know, especially physician collaboration and supervision in care teams. But AI is not a resident or an NP with a defined scope of practice and a license.
AI is a different kind of tool, and if we use the wrong mental model, we may build the wrong kind of oversight around it.
When the Machine Challenges the Status Quo
There are already areas where AI performs extraordinarily well. A randomized clinical trial in JAMA Network Open found that physicians using a large language model did not significantly outperform physicians using conventional resources, while the LLM alone scored higher than both physician groups on diagnostic reasoning performance.
That doesn’t mean AI should be practicing medicine independently, certainly. But it does challenge the assumption that putting a physician above the machine automatically makes the system safer or smarter. Our current hierarchy may feel comforting, but the purpose of medicine is to take better care of patients, not to make clinicians feel in control.
This is where humility matters.
I don’t mean the performative version, the kind we put in mission statements and then forget when the going gets rough. I mean the real kind. The uncomfortable kind that forces us to admit that a tool may see something we missed, retrieve something we forgot, or generate a differential diagnosis broader than the one we had in our head.
A big part of humility is understanding that you may not always be the smartest person in the room.
Doctors make decisions under pressure, often in chaotic environments. We carry enormous cognitive load. We get interrupted. We get tired. We have biases, and we anchor too early. Sometimes we settle on the diagnosis that feels most available, even when another possibility deserves more attention. A good tool should be welcomed into those moments.
Of course, AI can make mistakes. It can be plainly wrong and sound certain when it should be cautious. Anyone who says otherwise is selling something. But physicians make mistakes too. A standard that requires AI to be perfect while humans are allowed to remain human misses the point.
A better standard is simpler: does the tool help us deliver better care? If the answer is yes, then our job is to learn how to use it well.
What Doctors Can Bring to the Partnership
The best future is not one where doctors rubber-stamp whatever the machine says. That would be dangerous and lazy. There is another path, one that asks physicians to understand the strengths and limitations of AI with the same seriousness we bring to any other clinical tool.
Physicians can help define where a tool can operate, what evidence earns it greater autonomy, when it should escalate, and what patterns of failure should narrow or stop its use.
Responsibility has to show up in the design and monitoring of the system, not simply in a signature at the end of a workflow. The goal is to build a form of oversight that produces the best care, even if it doesn’t resemble the hierarchy we’re used to.
When physicians raise concerns about patient safety, bias, privacy, or accountability, those concerns deserve serious attention. AI only belongs in medicine if we’re disciplined about those issues. But preserving physician authority for its own sake is not a patient-centered strategy — patients should not have to wait for the profession to feel emotionally ready for a tool that may help them.
The patient should always remain our North Star. If AI helps someone understand their condition, that matters. If it helps a clinician consider a diagnosis they might have missed, that matters. If it expands access in a system where too many people wait too long for answers, that matters. And if it introduces new risks, those matter too.
But risk can’t become an excuse for resistance, because healthcare is already full of risk. Delayed care, missed diagnoses, rushed visits, and patients sitting at home with unanswered questions all carry consequences.
The physician of the future will be valuable because they can bring judgment, context, ethics, good communication, and trust to an increasingly complex care environment.
AI may become an extraordinary clinical partner. It may also cause harm if used poorly. We have to hold both realities at the same time.
Our responsibility is to build a better model of care around that tension: one where physicians are humble enough to be helped, technology is powerful but wisely deployed, and the patient remains the reason any of this matters.
— Rafid Fadul, MD
