Humans for technique, chatbots for understanding

I recently had an opportunity to present my thoughts on medical AI at a scholarly conference hosted by Oxford University.  I started with a situation that I faced as a patient.  After my doctor noticed some irregular heartbeats, I wore a monitor for a few weeks. It showed that I was having multiple runs of ventricular tachycardia.  The results were terrifying.  I wondered: Am I going to die?

I consulted many cardiologists. They offered me options—pacemaker, implantable defibrillator, medication—without much clarity about which was right for me. One expert put it plainly: "You are in an evidence-free zone." An electrophysiology study found no identifiable source for the arrhythmias but the doctors ablated some scar tissue from a prior MI anyway. A telemetric monitor was implanted. The runs of v-tach resolved.

In addition to talking to my many specialists, I also consulted ChatGPT. The LLM offered an advantage – I could ask it to frame the information about my situation and prognosis in an optimistic way or a pessimistic way. I could design my own nudges.

The optimistic AI doctor told me that many patients with my situation live for decades with treatment. The pessimist reminded me that ventricular scarring and documented v-tach each independently increase mortality risk, and that my ten-year survival odds were substantially lower than average. Both were clear, careful, and honest. I could ask follow-up questions. I could request citations. I could explore my own situation at my own pace, on my own terms.

The AI was more helpful than the human doctors. I say that not simply to be provocative but because understanding why it was better is a crucial question in medicine right now.

What Makes Doctors Special?

Traditional answers to this question usually highlight the doctors’ specialized knowledge, lengthy training, fiduciary commitment to the patient's interests, and their skill in responding to each patient's particular situation rather than to statistical averages. These are genuine goods. But each of them has been eroding for decades.

Yes, doctors have specialized knowledge, but the more specialized they are, the less they can see the big picture and help patients understand the range of choices and trade-offs that they face. Doctors’ lengthy training focuses mostly on the acquisition of factual knowledge, a skill in which he settings in which health care is delivered today are impersonal.  Doctors who have years of experience may have trouble keeping up to date with the latest developments.  Evidence-based medicine, for all its genuine achievements, is built on populations, not individuals, and has led to treatment protocols that render both doctors and patients interchangeable—nodes in a system optimized for reproducibility.

Given all that, the concern that AI is impersonal just doesn’t resonated. So is the rest of the health care system.  AI merely reflects what physician-scientists have created over decades. Ironically, it often does so in ways that are more humanistic. AI doesn’t face the constraints of the productivity-rewarding work environments that shape medical practice.

Walter Benjamin's Challenge

To think about what is happening, I have found it useful to reach for Walter Benjamin. Writing in Berlin in 1935, Benjamin was worried about the ways that cinema was being used to glorify the Nazis.  He started thinking about the ways that mechanical means of creating visual art changed the meaning of art.  He argued that an original work of art possesses an aura—its unique, unrepeatable presence in time and space, its embeddedness in tradition and history. Mechanical reproduction—photography, cinema—strips that aura away. There is no original. The image can be copied infinitely, distributed everywhere, seen by anyone.

Benjamin was watching, in real time, as cinema became a weapon. Leni Riefenstahl was manufacturing artificial aura around a political figure through sheer technical spectacle. Charlie Chaplin was using the same medium to mock and deflate it. The technology was neutral; what mattered was what it was put to work for.

Today, with AI, medicine is undergoing a similar transformation. As I found out, the expertise of an experienced clinician can now be recreated, in seconds, on my phone.  Furthermore, I could “edit” the portrayal of the physician to meet my own informational needs. I could ask for more detail or less, for studies backing up claims, and even for different designs of the choice architecture.

The physician's authority was once inseparable from physical presence—sight, touch, sound, intuition. It lived in one person, in was accessible only in one room. It could not be downloaded.

Now it can—or something very like it can. A machine learning model can encode the pattern of that attending's intuition as a risk score. Accurate, scalable, impersonal, reproducible.

Two Futures

I ended my Oxford talk with a simple contrast. One future is algorithmic medicine: the patient is a dataset, the clinician is a node, the encounter is a workflow, the algorithm decides, and no one is quite responsible. The other is relational medicine: the patient has a story, the clinician is a steward of meaning, the encounter is irreplaceable, and AI handles the pattern while the human holds the care.

I was unsettled by my own experience as a patient.  My doctors were excellent. They did everything according to the book, they followed the evidence, they offered me choices. With my problems, I could have been dead. Instead, I feel great.  Still, the particular excellence of the doctors was in providing skilled, high-tech, interventions.  Machines cannot (yet) do that. But when it came to understanding what was going on, the chatbots were better than the humans. 

If my experience is generalizable, then it is clear what the future holds.  Physicians will be the technicians of medicine, while AI will be both the brains and the heart. AI will gather the data, counsel the patient, present options, and discuss prognosis.  Doctors will insert the stents.

This post is drawn from a talk I gave at Oxford, based on my paper "The Lost Aura of the Physician in the Age of Artificial Intelligence," published in JAMA in March 2026.