AI Is Forcing Doctors to Ask What Remains Uniquely Human in Medicine

Artificial intelligence is now advanced enough that leading physicians and researchers are openly debating whether it can outperform human doctors at core medical tasks within the next few years. A new article in the Journal of the American Medical Association argues that autonomous AI may surpass doctors and doctor-AI combinations by 2030 in diagnosis, treatment recommendations, and chronic disease management—raising a bigger question about what the medical profession becomes next.

The debate matters because it is no longer about whether AI will assist clinicians. It is about whether the most reliable medical judgment could soon come from systems that work without humans in the loop, a claim that has startled doctors, ethicists, and health-care leaders alike.

The paper, co-authored by high-profile physician-ethicist Ezekiel Emanuel and venture capitalist Vinod Khosla, says the evidence from recent AI-in-medicine studies points toward a near-term transition in which autonomous systems may outperform physicians on a handful of essential tasks. Critics say that conclusion goes too far, too fast, and rests heavily on studies that do not always mirror real-world care. Still, even skeptics concede that medicine is moving into unfamiliar territory.

What the JAMA paper is arguing

The authors’ central claim is straightforward: AI is approaching a point where it can deliver better medical decisions than people. Their analysis, which reviews published research on AI in medicine since the start of 2024, suggests that autonomous systems could soon exceed both physicians and hybrid human-AI teams in five foundational activities: taking patient histories, reaching diagnoses, deciding on tests, recommending treatment, and managing chronic illness.

That is an unusually sweeping claim for medicine, where incremental improvement is the norm and patient safety is paramount. But the article does not present the outcome as speculative hype. Instead, it frames the shift as increasingly likely, even if uncomfortable.

The paper’s authors argue that medicine is nearing a threshold at which AI alone could provide the best care for several essential tasks, and they suggest that involving humans may sometimes reduce performance rather than improve it.

That last idea is especially provocative. The authors contend that human intervention can interfere with AI’s effectiveness, a position that runs against the common instinct that human oversight should always make automated systems safer or better.

How did a longtime skeptic become convinced?

Ezekiel Emanuel says he did not begin this project as an AI evangelist. For years, he dismissed predictions that machines would replace core physician functions. Vinod Khosla, who has promoted this idea for more than a decade, had been telling audiences for years that AI would eventually take on the bulk of what doctors do. Emanuel says he once rejected that view outright, believing the work of doctors was too complex to automate.

His thinking changed after he read a manuscript by Robert Wachter, the chair of medicine at UCSF, who described a future in which premium, high-touch care would still involve physicians, while lower-cost care might increasingly rely on AI systems. Emanuel says that book prompted him to reconsider a question he had long assumed was rhetorical: if AI can handle much of diagnosis and treatment, what role is left for doctors?

Once that question seemed plausible, Emanuel helped assemble a research team with Khosla and others to examine the available evidence. The group’s paper acknowledges conflicts of interest, including Khosla’s investments and the affiliations of the co-authors, but argues that the data warrant a serious new conversation.

Why are critics pushing back?

Critics say the paper draws too much from studies that are not equivalent to actual bedside medicine. John Whyte, chief executive of the American Medical Association, argues that much of the literature reviewed by the authors includes simulations or structured tests rather than direct, blind comparisons in the real world.

Whyte also points to recent evidence suggesting that patients often struggle to use large language models effectively in clinical settings. In his view, that limits the practical value of AI as a standalone medical provider.

The AMA’s position is that AI can be valuable, but it should be used within a physician-led care plan rather than as a replacement for trained clinicians.

That concern reflects a broader skepticism across medicine: performance in a controlled study is not the same as performance when the patient is anxious, symptoms are incomplete, histories are messy, and decisions carry real consequences.

What do the studies actually show?

The evidence base is growing, but it is uneven. AI systems have made notable progress in tasks that require pattern recognition, information retrieval, and differential diagnosis. At the same time, many medical studies still evaluate systems in simplified environments, using carefully prepared inputs rather than the fragmented, contradictory information physicians often receive in practice.

That distinction matters. A model that appears outstanding at predicting a diagnosis from a clean case summary may perform less well when the information is incomplete, the patient describes symptoms poorly, or multiple conditions overlap.

The JAMA paper’s authors say the trend line is clear enough to justify their forecast. Still, even they acknowledge that predicting medicine four years from now is hard, and that their conclusion is unsettling.

Why does the debate matter to patients?

Because the outcome could change how medical care is delivered, how quickly patients are seen, and which parts of the health system remain human-led. If AI can reliably triage symptoms, identify likely conditions, suggest tests, and recommend treatment, it could lower costs and expand access—especially in places with physician shortages.

But the stakes are higher than efficiency. Medical decisions involve fear, ambiguity, communication, and trust. Patients are not simply bundles of symptoms. They need explanations, reassurance, and someone responsible when things go wrong.

That is why the question is not only whether AI can match doctors on narrow benchmarks. It is also whether people will accept a system where the best medical answer may come from software rather than a person.

What role would remain for human doctors?

Even the most aggressive AI advocates do not argue that every doctor will disappear overnight. Vinod Khosla says physicians will still be needed in areas such as surgery and emergency care, at least for now. But he sees expertise and judgment as increasingly replicable by software, with doctors perhaps serving mainly to test and refine the systems.

Neal Khosla, who leads the AI-enabled health company Curai Health, says the policy environment will likely evolve to allow AI systems to prescribe medication in the future. His company already blends automation with physician oversight, a model that sits between today’s medical norms and the autonomous future envisioned by the JAMA paper.

For skeptics, that future is far from inevitable. For enthusiasts, it is already underway.

What about empathy and judgment?

Those may become the most defensible human advantages. Robert Wachter argues that even if AI proves superior at many technical tasks, clinicians will still matter for communication, comfort, and delivering difficult news.

He suggests that the most valuable doctor of the future may be less a diagnostician and more a guide, interpreter, and trusted human presence in moments of uncertainty.

Wachter says AI may win on information processing, but physicians could remain essential for the parts of care that involve trust, emotion, and nuanced decision-making.

That view does not deny the advance of AI. It argues that medicine contains layers beyond inference, and that some of them may remain stubbornly human.

What is the “doorman fallacy” and why does it matter?

The doorman fallacy is the idea that automation will eliminate a job because it appears to cover the core function, while missing all the other duties that make the role valuable. Wachter uses the example of apartment doormen, whose jobs survived automatic doors because they also accept deliveries, help residents, and provide a human presence.

Applied to medicine, the argument suggests that even if AI can handle diagnosis and treatment recommendations, physicians may still offer services that are harder to automate: context, advocacy, bedside judgment, and reassurance.

But the analogy has limits. Opening a door is not the same as managing complex disease. In health care, the most important tasks are also the most consequential. A mistaken diagnosis can have life-or-death consequences, which is why some observers think the doctor’s role is more vulnerable than the doorman’s ever was.

Could AI deskill the next generation of doctors?

Yes, and that may be one of the biggest hidden consequences of widespread AI adoption in medicine. If trainees rely on software for instant answers, they may never develop the deep pattern recognition and judgment that come from making difficult decisions unaided.

That concern is already surfacing in medical education. Some educators worry that permitting students and residents to lean too heavily on AI could weaken their ability to take histories, perform physical exams, and reason through uncertainty without digital assistance.

The problem is circular: the better AI gets, the more tempting it becomes to rely on it; the more people rely on it, the less practice they get; and the less practice they get, the more dependent they become.

1. Students use AI for fast answers.
2. They practice fewer independent assessments.
3. Clinical judgment develops more slowly.
4. Future doctors become more dependent on AI.

Supporters of AI in training argue the opposite: that refusing to use the best available tools could itself be irresponsible. If an AI system can improve care today, they ask, why deprive learners of it?

How would regulation and liability change?

Regulation is likely to be the gatekeeper between prediction and reality. Even if AI systems become strong enough to prescribe or direct treatment, health authorities and medical boards would still need to decide when and how such systems can be used independently.

That raises thorny questions: Who is liable if an autonomous system gives the wrong recommendation? How do insurers reimburse AI-led care? What standards will determine whether a model is safe enough for unsupervised use? These are not academic issues; they are the mechanisms that determine whether a technological breakthrough becomes routine practice.

Liability may prove especially important. Hospitals and physicians have long operated within a framework where accountability is traceable to a human professional. An autonomous AI model blurs that chain in ways current law is not fully prepared to handle.

What does this mean for the broader economy?

The debate over AI doctors reaches beyond health care. If machines can outperform highly trained professionals in one of the most trusted and skill-intensive human occupations, then similar pressure may spread across other fields built on judgment, expertise, and certification.

That broader anxiety is part of why the discussion has attracted so much attention. Health care is not just another industry. It is one of the places where society has long assumed that human expertise must remain central. If that assumption breaks, the implications for labor, education, and social stability could be profound.

Some observers have started to imagine a world in which certain roles are preserved primarily because society wants a human face on decisions that are increasingly machine-made. Others think that market forces will simply reward the cheaper, faster, and often more accurate option: AI.

Could some jobs become “human-reserved”?

Possibly, but only if policymakers or institutions decide that preserving human labor has social value beyond productivity. One argument circulating in the AI debate is that certain tasks may need to remain human-led to maintain trust, dignity, and employment—even if machines can do them more efficiently.

In medicine, that could mean reserving some roles for doctors as counselors, interpreters, and decision partners even when AI is technically capable of doing more.

Where does this leave doctors now?

For the moment, doctors remain essential. AI is spreading quickly, but medicine still depends on human institutions, clinical judgment, and legal responsibility. Most patients will continue to see doctors who use AI as a tool rather than a replacement.

Still, the direction of travel is difficult to miss. The most striking part of the current debate is not that AI might someday help doctors. It is that respected voices are now asking whether the best care could eventually come from AI without doctors at all.

That does not mean the profession is disappearing next year. It does mean physicians are facing an identity crisis that extends beyond technology. The question is no longer whether AI will change medicine. It is which parts of medicine remain meaningfully human once the software becomes better than the average clinician at the work doctors have traditionally called their own.

The answer may depend on whether patients, regulators, and doctors themselves decide that the future of care should be optimized solely for accuracy—or whether medicine must also preserve the human relationship that has always defined it.