The Role of Doctors in an AI World: What a Century of Automation Predictions Gets Wrong About Medicine

Gurpreet Dhaliwal, MD, spoke at the 2026 TDC Group Executive Advisory Board Meeting in Napa, California, which gathered top healthcare executives, academic researchers, and clinical leaders to discuss “Healthcare AI in 2026: Emerging Issues as the Field Matures.”

Dr. Dhaliwal opened his presentation by describing himself as a fan of AI but emphasized that he is an even bigger fan of the human brain. “I have spent a career studying how physicians think in order to teach the next generation to do so even better,” he noted.

When asked the question on everyone’s mind right now—will we still need doctors in a world of AI?—his answer is a resounding yes.

To explain why, it helps to look less at what AI can do today and more at what history tells us happens when powerful new technology meets an established profession. That history is far more instructive than the AI prophecies currently dominating the headlines.

The Persistent Myth of the Disappearing Doctor

In 2011, Vinod Khosla, the venture capitalist and Sun Microsystems co-founder whose track record for spotting transformative technology is genuinely extraordinary, predicted that AI could replace 80 percent of what doctors do within roughly five years. That did not happen. He doubled down with a paper proposing the “20 percent doctor,” arguing that within five years, only the hand-holding, emotive parts of medicine would remain human while the cognitive work could be fully coded into software. This also did not happen. Later, he predicted the FDA was on the cusp of approving an app to replace primary care physicians—arguably the hardest job in all of medicine. More recently, he described no longer consulting doctors at all, texting imaging findings of a fractured wrist to ChatGPT instead of his surgeon.

Elon Musk has gone further, suggesting medical school is now “pointless” because his company’s Optimus robots will eventually operate more skillfully than any human surgeon, giving everyone access to what amounts to the world’s best surgical care.

These are bold claims from people who have made extraordinarily good bets before. But when you actually consult the people who study what happens to labor markets when new technology arrives, the picture looks very different. Four years into the current AI boom, there is still no credible evidence of the widespread labor disruption that was forecast.

The consistent finding across decades of technological change is that a handful of industries see real disruption, but for most professions, technology arrives slowly and changes the texture of the work gradually, rather than eliminating it outright.

What the Doorman Teaches Us

To see why this occurs, consider a more familiar example than diagnostic medicine: the hotel doorman.

The technology to replace a doorman with an automatic sensor has existed since the 1950s, migrated to supermarkets by the 1960s, and could easily have replaced doormen at hotels and apartment buildings by the 1980s. The sensor is cheaper, has near-perfect uptime, needs no employee benefits, and performs the stated function—opening the door—essentially flawlessly. So why do doormen still exist?

The most common mistake is to model a doorman’s job as “opens the door” which misses everything else the role actually does: providing security, offering wayfinding and local knowledge, creating a sense of hospitality, and signaling the prestige that lets a hotel charge premium rates.

Replace the doorman with a sensor and you may save money on staffing while quietly eroding the value of the entire property.

The same mechanistic mistake shows up whenever people forecast the end of a profession. We define the job by its most visible, most easily codified output, and assume that is the whole job.

In 2017, Stanford researchers trained a deep learning system on roughly 130,000 skin images and found it matched—and in some cases exceeded—board certified dermatologists at distinguishing benign lesions from melanoma. It was a genuinely remarkable result. What took a dermatologist a decade of medical school and residency to learn, the algorithm learned in three months.

It is tempting to conclude that this kind of pattern-matching accuracy should replace the specialist. But the same logic that seemed to doom the radiologist over a decade ago has consistently underestimated what these clinicians actually do. Whenever we make grand predictions about the replacement of jobs, almost none of us actually know anyone else’s job.

A radiologist does not just interpret pixels; they design the scan protocol, integrate findings across imaging modalities over time, weigh a finding differently depending on whether the patient is immunocompromised or an elite athlete, and consult directly with the ordering clinician to shape the next diagnostic step. None of that lives in a training dataset. It lives in the accumulated judgment of clinicians who have done the job for years.

Medicine Begins Where Certainty Ends

Every new technology that has entered medicine over the past 20 years—CT scans, genetic testing, rapid PCR testing—has ushered in great advances but also generated new categories of diagnostic uncertainty, rather than eliminating it. Diagnostic dilemmas appeared that require the most experienced clinicians to sort out. If there were no uncertainty in a clinical encounter, you wouldn’t need a doctor.

But uncertainty is the permanent condition of medicine, and managing it is a core job description of the physician. It transcends the nominal task of “matching symptoms to diagnosis” that AI genuinely excels at.

When a young person asks Dr. Dhaliwal whether they should worry about machine learning before committing to medicine, his answer is simple: Don’t worry about machine learning. Worry about how you will become a learning machine. AI will change how the job gets done, but history suggests it will not eliminate the need for the judgment that makes the job possible in the first place.

Gurpreet Dhaliwal, MD, is a clinician-educator and Professor of Medicine at the University of California, San Francisco. He is the site director of the internal medicine clerkship at the San Francisco VA Medical Center, where he teaches medical students and residents in the emergency department, urgent care clinic, inpatient wards, outpatient clinic, and morning report. He studies, writes, and speaks about how doctors think—how they make diagnoses, how they develop diagnostic expertise, and how they interface with technology to augment their thinking.

The opinions expressed here do not necessarily reflect the views of The Doctors Company. We provide a platform for diverse perspectives and healthcare information, and the opinions expressed are solely those of the author.