According to the authors, "AI-alone medical care is likely to be better than physician-only or physician-AI hybrid care" when it comes to cognitive medical functions. They also argued that generative AI either rivals or outperforms physicians on several medical tasks, such as eliciting medically relevant information, establishing a differential diagnosis, specifying diagnostic testing, prescribing guideline-concordant treatment, and managing chronic diseases.

The authors cited multiple studies that found AI performed significantly better than human physicians in several areas. For example, one study found that Google's Articulate Medical Intelligence Explorer was "significantly better" than physicians at eliciting patient complaints, reviewing symptoms, and taking medical history. In another study, researchers found that ChatGPT o3 correctly identified a final diagnosis first in 60% of 377 complex real-world cases while 20 internal medicine physicians only did so in around 16% of 302 cases.

The authors also argued against the idea of physicians working alongside AI tools, saying that humans will likely contribute more errors when they incorrectly overrule AI. In a 2024 study, researchers found that introducing real doctors to AI-driven case studies led to significantly worse results than when the AI performed alone.

As AI models and tools continue to rapidly improve, "physicians, policymakers, and others need to urgently devise approaches to workflow, liability, regulation, reimbursement and medical education," the authors wrote.

According to Whyte, while the op-ed brings up important discussion topics, "is simply performing a diagnostic test the practice of medicine? I don't think it is."

Overall, AI tools "augment what the practice of medicine is," Whyte said. "It's an art."

AI sees data. Clinicians see people.

By Jolene Ackerson, senior director of provider technology consulting, Optum Advisory

While working as a physician assistant in the neuro ICU, there was more than one occasion when every indication — the Glasgow Coma Scale, the imaging, and the data — said the patient was unchanged. On paper, there was nothing to act on.

But something didn't sit right. Each time, I trusted that instinct and acted on it. And each time, had I deferred to the data alone, the outcome would have been different.

I think about those moments every time I read the current debate over AI in medicine.

In the recent JAMA perspective, the authors argue that autonomous AI may soon outperform not just physicians, but physician-AI hybrids. They also provocatively suggest that keeping humans "in the loop" can actually degrade performance. Soon after, AMA released its framework on physicians' role in the era of digital health and AI and clearly emphasized its own position: physicians must remain central, and humans can't be designed out of the loop.

Two credible sources. Opposite conclusions. Here's where I land as a provider:

• AI can generate insights and suggestions, but the provider owns the decision. Integrating evidence, weighing competing risks, and reading a patient's unique circumstances requires judgment and accountability technology can't replicate.
• Intuition is a clinical signal, not noise. The JAMA piece measures medicine as discrete cognitive tasks. But those ICU moments I experienced weren't tasks, and the instinct that flagged them is exactly the variable those studies can't capture. AI saw stable data. I saw the patient.
• Accountability can't be automated. When the algorithm is confident and wrong, someone has to answer for the outcome. Even the JAMA authors concede liability and regulation remain unsolved.
• Trust is earned, not assumed, and providers belong at the design table. We validate and verify the sources behind AI output before it touches a decision, and we help design the workflows rather than receive them.

AI will keep getting better, and I want it in the toolkit. But the day we treat clinical intuition as a bug to engineer out is the day we stop practicing medicine.

The best medicine won't come from choosing between a provider and a machine. It will come from someone who knows when to trust the data and when to trust their gut. Technology alone doesn't transform care; informed, clinician-led adoption does.

*Advisory Board is a subsidiary of Optum. All Advisory Board research, expert perspectives, and recommendations remain independent.