Christopher Gardner and Jeffrey A. Singer
AI and Health Care: A Policy Framework for Innovation, Liability, and Patient Autonomy—Part 5
The American medical licensing system has caused a significant artificial shortage of primary care providers. The practical implication is that people are using general-purpose AI for their health concerns. AI has demonstrated incredible promise for use in primary care. Whether that potential is realized depends on the FDA.
The problem is that the FDA doesn’t regulate artificial intelligence as such—it regulates intended use. The moment a personal AI tool crosses the line from offering general information to giving individualized medical guidance, it risks being classified as a medical device. That triggers a costly, time-consuming approval process designed for static products, not for continuously learning software. Developers quickly learn the lesson: the more clinically useful their tool becomes, the more likely it is to face regulatory hurdles.
The result is a perverse incentive. To avoid FDA oversight, developers often limit what their tools can say or do, keeping them in the “wellness” category even when they could safely provide more meaningful guidance. In other words, the regulatory framework nudges innovation toward less capable products. At a time when millions of Americans lack access to primary care, that tradeoff is not just inefficient—it is harmful. The line the FDA draws is not between safe and unsafe—it is between general and useful.
What is Primary Care?
Primary care practitioners serve as the first point of contact for most patients. These are the professionals we turn to for basic checkups, immunizations, preventive care, and treatment of common, self-limited conditions. Primary care usually involves basic symptom gathering, analysis, and care coordination, but it is also essential for identifying complex issues early.
This critical role is why primary care is the medical field with the highest impact on a population’s health. One additional primary care physician per 10,000 people is associated with life spans nearly two months longer. US adults who “regularly see a primary care physician have 33% lower health care costs.” Primary care physicians decrease the number of visits to the ER, lower the likelihood of recurrent strokes, and increase the willingness of men to pursue active treatment for depression.
Despite these well-documented benefits, 92 million Americans live in areas with a shortage of primary care providers, and even more do not have a PCP. These numbers will likely balloon over the coming years as the PCP shortage is expected to grow from 15,600 to 70,600 by 2038.
Yet the supply of primary care providers is not solely determined by patient needs—it is also influenced by policy. Medical licensing laws limit nurse practitioners, physician assistants, and other clinicians from practicing to the full extent of their training, even though many could safely deliver routine primary care. Meanwhile, the third-party payer system encourages physicians to pursue higher-paying specialties rather than primary care.
The result is a workforce bottleneck, not caused by a lack of capable providers but by regulatory constraints. Patients, however, don’t wait for licensing boards to catch up—they seek care wherever they can find it. Increasingly, that means turning to general-purpose AI, which is already capable of handling many routine primary care tasks. In effect, licensing laws are restricting human care while accelerating the demand for machine substitutes.
It should come as no surprise if incumbent licensed professionals respond by urging lawmakers to restrict patients’ ability to consult AI unless these tools are overseen by licensed clinicians. Framed as a matter of “patient safety,” such proposals would, in effect, maintain the same licensing system that caused the shortage initially and protect existing professionals from new competition.
This is a familiar pattern: when innovation threatens established interests, those interests often try to steer it through existing regulatory structures. Lawmakers should view these arguments with appropriate skepticism. Protecting patients does not require protecting a system that already leaves millions without access to basic care.
The Potential of Artificial Intelligence
One of the most promising solutions to the primary care shortage has been the development of high-quality, health-focused AI. Optimized to manage even the basics of primary care, AI could boost the productivity of current medical professionals and enable them to concentrate on more complex cases. It would also give Americans and their caregivers instant access to information about their ailments, injuries, and emergencies around the clock. This is true even in the most remote areas of the US, where open-source models could be downloaded and used without any internet or service connection.
AI can also help tie together the silos of the American healthcare system and give primary care providers back the time they need to help manage complex or unfamiliar cases. Today’s medical students will spend the next four years getting a generalist education in medical school before spending the following three to seven years as a resident focusing exclusively on their specialty. While these programs develop exceptional expertise for doctors in their specialty, they also limit their scope, reducing their ability to consider differential diagnoses for conditions that span multiple systems or organs. The natural consequence is that specialists exhibit significant bias towards “pulling” conditions into their specialty during diagnosis. This bias directly contributes to the 2.6 million diagnostic errors made by American doctors each year, costing an estimated $870 billion.
Mark Atwood’s story illustrates the shortcomings of a system that focuses almost exclusively on managing disease progression, not disease prevention, as well as a small part of the potential of AI. Having lived most of his life with an undiagnosed chronic nerve condition, Mr. Atwood suffered symptoms that sent him to neurologists, dermatologists, allergists, and endocrinologists. But in his words, “each specialist peered through a keyhole and caught one sliver of the whole.” This fragmented and frustrating approach to care is sadly not uncommon. Patients with chronic illness, pain conditions, or other complex problems and medical profiles suffer as our medical system commonly fails to support them. The result is that patients often drift from treatment to treatment or between providers, with none of them able to put all the pieces together.
This is where Mr. Atwood’s story is unique. After 50 years of doctors’ visits, scans, and tests—often with specialists focused narrowly on their own domains—it took a single conversation with an AI system to connect the dots and identify the underlying problem. His experience highlights a structural weakness in modern medicine: fragmentation. When care is divided across silos, no one is responsible for integrating the full picture.
Mr. Atwood’s case demonstrates the potential of AI not just to expand access, but to plug the gaps that patients routinely fall through. AI systems can synthesize information across specialties in a way that time-constrained, highly segmented human care often cannot.
Plugging those gaps isn’t just about giving primary care providers more time—it’s about restoring continuity, perspective, and ultimately autonomy to patients. Today’s AI tools can help patients see the bigger picture: understanding their options, identifying the right questions to ask, and deciding when a situation truly requires urgent care. In a system where information is fragmented, AI’s ability to integrate may be its most valuable feature.
AI has given patients the ability to manage and monitor their own health safely when doctors wouldn’t otherwise be available. America has a long history of promoting and celebrating this type of patient autonomy, but existing FDA regulations heavily limit the functionality of new AI models and disincentivize innovation.
These regulatory barriers don’t operate in a vacuum—they often align with the interests of incumbent providers, who benefit when new tools must fit within existing, license-centered models of care.
Conclusion:
Federal regulators should reconsider the barriers they place in front of this technology. AI is not a silver bullet for the primary care shortage, but it has the potential to expand access and help patients navigate a fragmented system that too often fails them. That fragmentation is not accidental—it is, in part, the product of licensing rules and regulatory constraints that limit who can provide care and how that care is delivered.
Every day, Americans suffer—and sometimes die—not because solutions don’t exist, but because they can’t access timely, affordable care. As policymakers restrict the supply of human providers, patients are already finding alternatives. The question is whether regulators will allow AI to help close that gap—or allow incumbent interests to shape rules that keep it from doing so. Loosening the FDA’s approach to software-based medical tools would not replace clinicians, but it could empower patients and extend the reach of care to millions who currently fall through the cracks. At a time when the system is strained, policymakers should be looking for ways to enable innovation, so fewer patients are left to navigate it alone.
To read other parts of this blog series, go here.