The Doctor Who Never Slept

For nearly two years, the first conversation about my health has not always begun with a human doctor. It has begun with artificial intelligence. I show it results, organize symptoms, compare values, and prepare questions. Then, when necessary, I go to the doctor with a clearer history and less time wasted trying to remember what happened first.

I know the sentence sounds more dramatic than the practice. I have not handed my health over to a machine. I use it to do something a brief appointment rarely allows: look at the whole film. One test is a photograph. A sequence of studies, habits, symptoms, and changes begins to resemble a story.

That is where I find the first advantage. Artificial intelligence can read in seconds what is scattered across laboratories, clinics, and specialists. It does not arrive tired at the last appointment of the day or need to reconstruct two years of history in fifteen minutes. It can point to a trend and help me formulate the right question. That does not make it a doctor; it makes it a tireless reader.

A trial published in JAMA Network Open in 2024 produced a striking result. Fifty physicians worked through complex cases using conventional resources or with additional access to a language model. Their median scores were similar: 74 and 76 percent. In an exploratory analysis, the model alone scored 92 percent. That result does not justify replacing a consulting room with a screen. The authors themselves warn against interpreting it as support for autonomous diagnosis. It does show that we have not yet learned how to combine two very different abilities well.

A machine can process more information than any professional could retain. A doctor can examine someone, recognize nuance, understand context, and take responsibility for a decision. When either capacity tries to take the place of the other, we lose. When they work together, a different kind of medicine becomes possible.

I fully understood this through a family close to me. They had spent three years moving through treatments, specialists, and incomplete diagnoses for an exceptionally rare disease. With their authorization and with care for the information, I gathered scans, studies, and reports. Instead of consulting one system, I used three models from different companies and asked them to challenge one another: one proposed a hypothesis, another had to refute it, and the third looked for missing evidence.

The conversation lasted for hours. They reached a concrete explanation and showed which facts supported it and which still needed to be checked. Later, a specialist at a clinic dedicated to that disease reached the same diagnosis. Artificial intelligence had not seen the patient, performed a physical examination, or closed the case. It had done something narrower and, at that moment, decisive: found a possibility hidden inside a mountain of information.

I did not think the earlier doctors had failed for lack of ability. A disease that almost never appears can remain outside an entire career's experience. Medical literature, by contrast, preserves cases from many countries and many decades. For a machine, what is exceptional in one city may be part of a worldwide archive.

That difference matters because rare diseases often impose a long diagnostic odyssey. The World Economic Forum has cited waits of six to eight years and more than one incorrect diagnosis in over forty percent of cases. Against that journey, a tool that widens the search does not compete with the doctor; it allows the doctor to see farther.

The second change is already on a wrist or a finger. A watch or ring can record pulse, sleep, activity, temperature, and other signals for months. The value is not in a single number but in the curve. Something can remain within a general range while still moving away from what was normal for that person.

In 2023, Kimmie Watkins's watch woke her because her heart rate had remained at 178 beats per minute for more than ten minutes. At the hospital, doctors found a serious pulmonary embolism. Years earlier, another alert led a teenager in Florida to seek care for a high resting heart rate; there, doctors found kidney failure that had not yet produced obvious symptoms.

Individual stories are powerful, but there is also evidence at scale. The Apple Heart Study enrolled 419,297 participants. Only a small proportion received irregular pulse notifications; among those who had another notification while wearing an electrocardiogram patch, the positive predictive value for atrial fibrillation was 84 percent. The watch did not deliver a final diagnosis. It indicated that something deserved a closer look.

This is what I am working on now: using rings to support the health of older adults. I am interested in a discreet device that can observe patterns without turning daily life into a hospital stay. If something changes persistently, it can raise an alert. The person keeps their routine, and the family gains an early signal without receiving a dashboard that invites them to monitor every minute.

The promise, however, brings three problems that should not be minimized. The first is privacy. Health data is intimate and can cause harm in the wrong hands. It should belong to the patient, travel encrypted, remain portable, and never become a tool for discrimination in employment or insurance.

The second problem is error. Artificial intelligence can be wrong and sound impeccably certain. That is why no important conclusion should depend on a single answer. We need to compare sources, ask the system to expose its doubts, and, above all, validate with a professional anything that could change a medical decision.

The third is anxiety. Measuring everything does not mean we should look at everything. A well-designed system should not wake us for every variation. It should recognize persistent patterns and explain why they deserve attention. Technology should reduce unnecessary worry, not manufacture a new form of hypochondria.

There is also a problem that comes before all three: our medical history is fragmented. The laboratory keeps one part, the clinic another, each specialist writes in a separate system, and the patient becomes a courier carrying studies, passwords, and files. Every appointment begins by rebuilding a life from scattered documents.

A personal health record reverses that logic. The data follows the person rather than the institution. Artificial intelligence can help organize it, summarize it, and turn it into questions. But that requires choosing tools whose privacy policies we understand. Uploading medical records to any service without knowing how it stores them would simply trade one problem for another.

We do not need to wait until that future is complete to improve a medical appointment. We can gather our studies, put dates in order, list medications, and record symptoms. A secure tool can help us find gaps or prepare a summary. The rule remains simple: every important finding must be discussed and validated with a flesh-and-blood doctor.

What I imagine ahead is not the disappearance of the human doctor. It is a doctor who receives a readable history instead of a disorderly folder. Less administrative work, less repetition, and more time for difficult cases, judgment, and the conversation no database can replace.

I like the paradox: the more we automate the mechanical part of medicine, the more human the encounter can become. Between one appointment and the next, a machine can watch trends at three in the morning. It does not sleep, lose focus, or grow tired. But when something truly matters, we need someone who looks at the person, not only the numbers.

That is why the doctor who never slept is not here to occupy the consulting room. It is here to accompany us to the door with better data, a fuller history, and more precise questions. Medicine can begin before something hurts without ending up in the hands of a machine. It can begin there and become, for that very reason, a little more human.

This note recounts personal experiences and discusses published research. It is not a diagnosis or medical advice.

Christian Ruggeri

Study cited: JAMA Network Open.