The American Medical Association has published a framework arguing that physicians must remain central to patient care as artificial intelligence moves into clinical work, and it lands the same week a prominent medical journal published the opposite case.
The framework, released by the Digital Medicine Society and first provided to Axios, holds that "the practice of medicine must evolve alongside" the technologies entering it, while maintaining that a physician's core responsibilities do not change.
For patients, the practical question is not which side is right in the abstract. It is unclear who will be responsible when an AI system is involved in their care, and neither document answers that.
Inside the Framework and What It Binds
The AMA document is a framework and implementation roadmap defining the role physicians should play as digital tools and AI enter practice. It argues that individual tasks change while underlying professional obligations, including direct human connection with patients and responsibility for clinical judgment, remain stable.
John Whyte, the AMA's chief executive and executive vice president, put the argument in concrete terms, asking whether a patient would want to be "treated by an LLM if you have chest pain."
What the document is matters as much as what it says. The AMA is a professional membership association, not a regulator. Its frameworks shape professional norms, inform advocacy and influence how members approach practice questions. They do not carry the force of law, do not bind hospitals or health systems, and do not determine what the FDA authorizes or what payers will cover.
The AMA also has an institutional interest in the answer, which does not make its position wrong but is worth naming. An organization representing physicians arguing that physicians are irreplaceable is advancing a view that aligns with its members' interests, and readers can weigh it accordingly.
A Directly Opposing Argument Published Days Earlier
An opinion piece in JAMA published this month argues that AI is already approaching or exceeding physician performance on core cognitive tasks, and that keeping humans in the loop could, in some circumstances, produce worse outcomes rather than better ones.
The authors write that when AI alone consistently outperforms humans alone, it also surpasses human-AI hybrids, and that "hybrid care" intended to catch AI errors is likely to worsen performance. They conclude that superior autonomous AI will likely be ready for deployment in some, and perhaps many, clinical workflows by 2030, and that physicians, policymakers, and educators need to address workflow, liability, regulation, reimbursement, and medical education urgently. The piece also raises the prospect of deskilling, in which physicians' own abilities erode through reliance on the tools.
Who wrote it is part of the story, and the draft version of this debate often omits it. The authors are bioethicist Ezekiel Emanuel of the University of Pennsylvania, research fellow Abe Baker-Butler, Curai Health chief executive Neal Khosla, and venture capitalist Vinod Khosla, an investor in Curai Health. Two of the four have a direct commercial interest in the autonomous model the piece advocates, which readers are entitled to weigh alongside the argument.
Whyte's response draws a line around what medicine is. Performing a diagnostic test is not the same as practicing medicine, he argued. The American College of Physicians holds a similar position, describing AI's proper role as supportive rather than autonomous.
Both documents are arguments rather than evidence. Neither is a trial, and the 2030 projection is a forecast by identified authors, not a finding.
Regulation Has Not Settled Liability or Payment
The gap both sides point at is real and unresolved.
The FDA issued a discussion paper this week contemplating how generative AI-enabled medical devices might be assessed and opened it for public comment. MedicalDaily reported on that paper and the comment window, which runs through October 19 and is open to anyone.
But device authorization is a narrower question than the ones that determine how AI actually reaches patients. Who bears liability when an AI system contributes to a harmful error remains largely unaddressed. So does how AI-provided care gets reimbursed, which in practice governs whether health systems adopt it.
Some deployment is running ahead of that framework. Utah has moved toward permitting AI-driven prescribing arrangements through a state pilot, and patient-facing tools are proliferating faster than the rules governing them.
Cost sits underneath the debate and is rarely stated openly. Autonomous or near-autonomous AI is attractive to health systems partly because it promises to deliver clinical work without clinician labor costs, in a sector where staffing is the largest line item. That is also why the reimbursement question is decisive: a service that cannot be billed will not be adopted at scale regardless of how well it performs, and a service that can be billed will be adopted quickly regardless of whether liability has been sorted out.
The AMA maintains its own augmented intelligence resources, and the FDA publishes guidance on AI and machine learning in medical devices. Neither addresses liability or payment, because neither organization owns those questions.
Reasonable Questions for a Patient to Ask
Nothing about this changes anyone's care today, and nobody should decline treatment or a test over it.
AI is already present in ordinary encounters in ways patients often are not told about, including ambient documentation tools that record and summarize visits and algorithms that flag imaging findings. The AMA's own survey research indicates that more than 80% of physicians now use AI in their professional work. Asking whether AI is being used in your care and whether a clinician reviewed its output is a reasonable question rather than a confrontational one.
If a practice uses ambient recording, patients can ask what happens to the recording and whether they may decline. If an AI tool contributed to a diagnosis, asking what the clinician independently concluded is fair. Patients using consumer chatbots to interpret symptoms should bring their findings to a clinician rather than acting on them, since those tools are not regulated as medical devices.
Access cuts both ways here, which is the part most coverage omits. Rural and underserved communities with few clinicians are the places where autonomous AI is most often proposed as a solution, and also the places with the least capacity to catch an error or seek a second opinion. A framework insisting on physician involvement means something different in a city with three hospitals than in a county with one clinic.
What remains unknown is whether autonomous AI care will expand before liability, reimbursement, and regulatory questions are settled, and whether professional frameworks like this one will shape that or simply describe a position that events overtake. The FDA comment period is the next opportunity for the public to weigh in, and MedicalDaily will report the outcome.
Key Questions Answered
What did the AMA release? A framework and implementation roadmap, produced with the Digital Medicine Society, argues that physicians must remain central as AI enters clinical care.
Does the framework carry legal force? No. The AMA is a professional membership association. Its frameworks shape professional norms and advocacy but do not bind hospitals, regulators or payers.
What does the opposing argument say? A JAMA opinion piece argues that AI already rivals or outperforms physicians on core cognitive tasks and that superior autonomous AI may be deployable in some workflows by 2030.
Do the JAMA authors have a stake in the outcome? Two of the four do. One is the chief executive of an AI telemedicine company, and another is an investor in it, a disclosure worth weighing alongside the argument.
Is either position backed by a trial? No. Both are arguments and analyses rather than experimental evidence. The 2030 figure is a projection by named authors.
Who is liable if an AI system contributes to an error? Largely unresolved. Neither the AMA framework nor current FDA activity settles liability or reimbursement, which are the questions that govern real-world adoption.
Can the public weigh in? Yes. The FDA opened a comment period on its generative AI medical device discussion paper through October 19, and it is open to anyone.
