Leading physician groups reject AI-over-doctor claims in joint statement

Six national physician organizations said Sept. 30 that statements casting artificial intelligence (AI) as inherently better informed than physicians, or suggesting physicians cannot make clinical decisions without first consulting it, "diminish physician expertise and risk undermining patients' trust."

The statement came from the American Academy of Family Physicians, American Academy of Pediatrics, American College of Obstetricians and Gynecologists, American College of Physicians, American College of Surgeons and the American Medical Association (AMA). It named no one.

It followed remarks from Health and Human Services Secretary Robert F. Kennedy Jr. at the Make America Healthy Again Summit in Washington one day earlier, where Kennedy said AI can give patients a second opinion "much better informed than any doctor in the country."

Kennedy also said OpenAI CEO Sam Altman had told him it would be malpractice for a doctor to diagnose or prescribe without checking AI. He pitched AI as a way for patients to digest medical records that can run a thousand pages.

AI plays some role in patient visits at 83% of medical groups, according to an Aug. 4 Medical Group Management Association (MGMA) Stat poll with 189 responses. In a January MGMA Stat poll of 328 respondents, 20% of medical group leaders said their organization has AI governance or a formal AI-use policy, 22% said they were developing one and 56% said they had neither.

What did the physician groups say?

"Physicians evaluate patients in context, drawing on years of training and experience," the statement reads. "They ask questions, understand each patient's unique needs and circumstances, exercise professional judgment and take responsibility for the care they provide."

"Physicians have long embraced technologies that improve care, and AI has tremendous potential to provide new tools and insights," the groups wrote. "But patient safety, physician expertise and the humanity of clinical practice must guide how these tools are developed and used."

The statement closes with a commitment "to ensuring physicians help lead the responsible development and use of AI in medicine." In an AMA survey of nearly 1,700 physicians released this year, 85% said they want a say in AI adoption in their practice, and respondents ranked clear liability frameworks among their top regulatory priorities.

What should an AI policy cover?

"Practices need to have some degree of governance in their practice, and it's important that there are policies in place," Anders Gilberg, MGMA senior vice president of government affairs, told Physicians Practice in September. "You don't want some physicians using maybe ChatGPT to query about some HIPAA protected information about a patient only to then get an incorrect or a hallucination and be liable for that hallucination."

Gilberg separates AI into categories a policy should address: AI embedded in existing systems, administrative AI, and what he called "the wild west of the large language models" that clinicians may access through personal accounts on their phones. "There's no AI compliance per se," he said. "It's the existing laws that you can go sideways with pretty quickly."

"Banning the use of AI is not a sustainable strategy for anyone," Asha Palmer, J.D., senior vice president of compliance solutions at Skillsoft and a former assistant U.S. attorney, said in an April interview. "If people are using it in the shadows, you're losing control of your data, you're losing control of your people, and you'll eventually lose control of your practice."

When a practice leader discovers unapproved tools, Palmer recommends asking why before imposing discipline. "What are we not providing you that you need or you feel is so necessary to deliver your standard of care, that you need this tool?" she said. She also recommends pressing vendors before anything goes live. "Ask them tough questions about where your data is going, how the models are being changed or trained," she said.

How should staff and clinicians handle patients who arrive with AI answers?

"Patients have always brought in their own information into these appointments because they are trying to be resourceful and trying to show up with good intentions to educate themselves," Amber Maraccini, Ph.D., vice president and head of health care and life sciences at experience management company Medallia, said in a February interview. "The most productive path would not be to dismiss the effort. Instead it's looking at the information and say, let's look at this together."

"There's no way for ChatGPT to have the entire context," Maraccini said. "It only has the context of the information you are feeding it, but the clinician is going to have a broader range of information to help paint that additional piece of the picture."

Nancy Cibotti, M.D., a primary care physician and the U.S. chief medical information officer at AI scribe company Heidi, described her practice's approach during an April Medical Economics and Physicians Practice webinar that Heidi sponsored. "We do get consent with every patient to use an AI scribe," Cibotti said. Some patients were initially very uncomfortable, she said, but that has eased as patients started using tools such as ChatGPT themselves.

Where does the liability land?

"If AI produces a recommendation that's different than the standard of care, and the doctor follows it, and the outcome is actually adverse, well, then by definition, the doctor has violated the standard of care," Richard E. Anderson, M.D., FACP, chairman and CEO of The Doctors Company and TDC Group, said in February on the Off the Chart podcast.

Large systems have "much more ability" to create dedicated teams to evaluate AI, Anderson said. "Every hospital, every doctor, has to make almost individual decisions about how AI is going to be used, and they have very little basis for doing it."

Marc Succi, M.D., executive director of the Mass General Brigham MESH Incubator, said in an April interview that he is comfortable with AI for low-risk, high-feasibility tasks such as documentation, summarization, patient-friendly explanations and billing.

For clinical decision support, patient message replies, lab orders and medication renewals, "that's where you got to stop and look at the level of performance very critically in multiple different ways, not just trust what the vendors say."

Gilberg said there is "probably a government role" in helping people understand how AI will affect health care and the human element.

"Retaining the human element is very important as well," he said. "Making sure that it doesn't override or interfere with the physician-patient relationship."