Dr. Shahar Madjar, Journal columnist

It’s a moonless night. The roads are wet and slippery. A car accident along the highway results in two drivers being severely injured. They are rushed to a trauma center in a nearby hospital. The doctors in the emergency department are called into action. They take vitals, assess the injuries and order bloodwork and imaging studies. Within seconds to minutes, they will have to examine the patients, make critical diagnoses and decide on the best course of action.

Upstairs on the general medical floor, the pace seems slower, but it is no less urgent. A 32-year-old woman presents with a 10-year history of unexplained abdominal pain that has become intolerable. During her hospitalization, she develops psychiatric symptoms. Her diagnosis is a mystery. Doctors scratch their heads in search of an explanation, yet the answer remains elusive.

These and other challenging cases–in the emergency department, on the hospital floors, at the clinic–seem to await a magic wand. Doctors are left reaching for the exact diagnosis, a proper treatment plan, and a solution that will diminish pain, resolve suffering and save lives.

What if such a solution already exists?

A groundbreaking study published in Science in April 2026, by Peter G. Brodeur and colleagues, tested whether artificial intelligence can serve as this magic wand.

The researchers didn’t just give AI a simple multiple-choice test; they threw it into the deep end of unstructured, chaotic, real-world medical data. First, they tested it on the cases described in the legendary New England Journal of Medicine Clinicopathological Conferences–notoriously complex, real-life diagnostic mysteries that baffle the brightest human minds. Then, they took it straight into the chaos of the Emergency Department (ED), feeding it the fragmented, messy initial notes of patients freshly admitted to triage.

The results were staggering. When tackling the elusive NEJM diagnostic mysteries, the model placed the correct diagnosis in its differential list 78.3% of the time. When the researchers looked at its broader list of “potentially helpful or close” diagnoses, its accuracy skyrocketed to a near-perfect 97.9%. But perhaps its most stunning triumph happened in the frantic environment of the ED. Faced with sparse, incomplete initial triage records, the AI identified the correct diagnosis 67.1% of the time, substantially outperforming two expert human physicians who scored 55.3% and 50.0%. When rated on a validated 10-point scale for clinical reasoning quality, the model achieved a perfect score in 78 out of 80 instances, soundly defeating resident and attending physicians alike.

What does this mean for the day-to-day work of a doctor? In the immediate future, it means physicians are gaining the ultimate cognitive co-pilot. Imagine a grueling shift where, instead of drowning in electronic health records or missing a rare symptom due to exhaustion, a doctor has a brilliant safety net constantly running in the background, whispering, “Based on this subtle lab trend, have you considered these three alternatives?” It will streamline paperwork, sharpen diagnostic accuracy, and cut through the administrative fog that leaves so many healers burnt out.

Will doctors become redundant? If medicine were nothing more than data processing, an advanced AI–acting as a brilliant but disembodied ‘brain in a jar’–would easily take the crown. But a clinical text prompt is not a living patient.

I see this technological transition as being similar to the advent of aviation autopilot or the rise of self-driving cars. An autopilot system can calculate wind resistance and fuel efficiency perfectly, and a self-driving car can effortlessly cruise a mapped-out highway. Yet, we would never board a plane without a pilot in the cockpit, nor would we trust a driverless car to navigate a chaotic, unmapped blizzard without a steering wheel. The AI excels when the road is clear and the data is neat, but real human health is messy, unpredictable, and full of sudden, complex detours.

Ultimately, there is a vast, unbridgeable chasm between possessing encyclopedic knowledge and practicing the art of medicine. An AI can hold the entire history of medical literature in its digital memory, but it cannot step into a patient’s room and truly see them.

Who will be there to listen to the patients, answer their questions and demonstrate genuine empathy? Who will notice the patient’s tears or hold their trembling hand in the quiet terror of a devastating diagnosis? Who will sit down with a grieving family, look them in the eye, and carefully reframe the painful story of health lost. Who will breathe hope into the mind of a patient faced with a harsh diagnosis and poor prospects?

An AI always seeks a definitive, mathematical output–a digital cure. But a good doctor understands the human spirit and that sometimes, healing requires time, patience, and a guided acceptance. The machine can handle the data, but the soul of medicine will always belong to the human at the bedside.

EDITOR’S NOTE: Dr. Shahar Madjar is a urologist and the author of “Is Life Too Long? Essays about Life, Death and Other Trivial Matters.” Contact him at smadjar@yahoo.com.