The World Health Organization (WHO) reports that 4.5 billion people do not have access to primary care. In the US alone, the American Medical Association (AMA) claims 83 million Americans do not have access, and the National Association of Community Health Centers (NACHC) puts the number even higher at one-third of the US population or over 100 million, including many children (NACHC).

Since losing my brother, a cardiologist, to a misdiagnosis some ten years ago, I have been studying the medical field of primary care, including three years at Stanford as a fellow, researcher, and grant recipient, and two at Harvard as a visiting scientist. This includes delving into a plethora of published studies, white papers, and other clinical references that underscore the multifaceted issues plaguing our current primary care landscape. From access disparities and escalating costs to diagnostic errors and provider shortages, the need for innovative solutions has never been more apparent.

One of the foremost challenges facing primary care today is the shortage of qualified providers. According to the Association of American Medical Colleges (AAMC), the United States could see a shortage of up to 139,000 physicians by 2033, with primary care bearing the brunt of this deficit. This shortage will exacerbate access issues, particularly in underserved rural and urban areas.

Moreover, the traditional model of primary care delivery is burdened by numerous systemic flaws. Diagnostic errors and misdiagnoses, estimated to affect 12 million adults annually in the US alone, highlight the fallibility of human decision-making in complex medical scenarios. According to a Johns Hopkins study published last year, nearly 375,000 people die annually from these errors, with another 400,000 suffering permanent disabilities (JH). The migration from physicians to lesser-trained nurse practitioners (NPs) and physician assistants (PAs), and increased consideration of pharmacists as prescribers, while necessary to alleviate physician shortages, elevates concerns about the quality of care and diagnostic accuracy. Burnout among NPs has also been reported to be rising. Desperate to alleviate the physician shortage, some US states have already passed laws for foreign medical school graduates to practice without residency training.

Furthermore, time-constrained primary care providers grapple with daily decision fatigue, especially in the afternoon, which can compromise the quality of care delivered. A study of 19,000 women showed that those requiring cancer screening referrals were less likely to be referred in the afternoon. Additionally, prevailing financial incentives often prioritize volume over value, perpetuating a fee-for-service model that incentivizes unnecessary procedures and tests. This is motivating an increasing number of primary care physicians to adopt a concierge model, charging additional out-of-pocket fees, sometimes cutting off as many as 80 percent of their patients from their services.

In full transparency, I admit to joining such a practice as a patient after exhausting a search for a quality doctor who accepts regular insurance in my home area.

Cognitive biases, such as confirmation bias and availability bias, further impede accurate diagnosis and treatment planning. Communication errors between patients and providers, exacerbated by time constraints and administrative burdens, contribute to suboptimal outcomes and patient dissatisfaction.

Physician burnout, a pervasive issue in modern health care, is driven by factors including excessive workload, administrative tasks, and a lack of work-life balance. This burnout not only harms individual providers but also exacerbates workforce shortages as disillusioned physicians exit the field.

On top of all of these concerns is the ever-expanding universe of medical knowledge, which already far exceeds a single human’s intellectual capacity to draw upon.

In light of these challenges, the concept of semi-autonomous AI-powered primary care providers emerges as a promising solution. By harnessing the capabilities of artificial intelligence, these providers can offer 24/7 accessible, cost-effective, and evidence-based care to patients across diverse settings. AI systems are not susceptible to human factors such as fatigue, cognitive biases, or burnout, offering consistent performance and decision-making.

You may ask, how do we circumvent the well-known flaws in AI diagnostic models? Studies have reported that the best large language models powering generative AI both hallucinate and demonstrate bias. Granted, but so do humans. The most recent statistics demonstrate that 47 percent of physicians report burnout, 27 percent depression, and 10 percent suicidal ideation. That is not meant to excuse inaccuracy by an AI system; it merely points out that AI would be expected to be more scrutinized than humans to identify and eliminate errors, if not held to a higher standard. AI engineering to correct errors could be implemented faster than modifying ingrained human behavior. Also, as chess computers did in the 1990s, AI is improving with each iteration.

Moreover, AI-enabled autonomous systems will not be without human oversight. While a human will remain in the loop, they will not have the classical training of today’s physicians; rather, they will be dually trained to understand both AI and medicine. One medical school program has already adopted this curriculum. Further oversight will be discussed shortly.

AI doctors can already augment the capabilities of primary care providers, assisting in diagnosis, treatment planning, and patient education. By leveraging vast amounts of clinical data now part of emerging large language models and machine learning algorithms, these systems can offer personalized care tailored to individual patient needs, reducing diagnostic errors and improving outcomes.

Due to regulatory and liability issues, some may argue that AI is not even ready for semi-autonomy. The biggest regulatory hurdle may well be state medical boards, which may feel threatened by or disagree with the quality of semi-autonomous AI-powered systems to practice medicine. In the United States, medicine is a licensed profession regulated by individual states. One of the most important functions of the nation’s state medical boards is issuing licenses to physicians. Boards are also responsible for evaluating whether a physician’s professional conduct or ability to practice medicine warrants modification, suspension, or revocation of the license to practice. I propose these same boards be given the same rights to license, govern, and oversee AI systems, permitting usage where there are known shortages within their states to stretch existing capabilities.

For liability concerns, let us consider the medical standards for malpractice. It consists of four components: duty to care, breach of duty to care, causation, and damages. Can an AI system have a duty of care? I say yes, because the developer could be held to that standard. The Food and Drug Administration (FDA) would review the product’s capabilities and determine its clinical validity and utility before allowing release to market, control claims, and hold the company accountable, as the FDA does with pharmaceuticals and medical devices. Therefore, between the state medical boards and the FDA, AI systems will be carefully regulated.

In the case of AI, a breach of duty would be determined by a product’s owner failing to maintain the product as approved or making improvements as determined by evolving medical standards of care. Users should expect no less. Causation would be determined by experts, as in any product lawsuit, and damages have already well-established precedents. Liability is not an insurmountable hurdle to giving tens of millions of people access to primary care where there is none today.

Implementing semi-autonomous AI-powered primary care providers requires careful consideration of ethical, regulatory, and privacy concerns. However, the potential benefits of improved access, reduced costs, and enhanced quality of care make it a compelling avenue for further exploration and development.

The challenges facing primary care demand innovative solutions that transcend the limitations of traditional care delivery models. Semi-autonomous AI-powered primary care providers promise to revolutionize health care delivery, ensuring all individuals have equal access to high-quality, patient-centered care regardless of time, geography, or socioeconomic status. They can become great health care equalizers.

Steven Charlap is a physician executive.