Major medical professional organizations see promise, too.

The head of the American Academy of Family Physicians, which boasts around 124,000 members, said AI could be seen as a capacity enhancer, giving doctors more space to see more patients on a per-hour-per-day basis by, in part, embedding an AI agent into each patient’s electronic health record. The ability for medical systems to use AI agents to schedule and remind patients about health screenings is another area AAFP CEO Shawn Martin envisions as transformative, as it could check or reduce the impact of high-cost health events such as cancer.

There is broad alignment on the upsides that AI holds for the medical industry. There is also collective concern that it risks eroding the human connection vital in the field, along with fears about mistakes it could make, including through so-called hallucinatory episodes.

The head of the American College of Physicians, which boasts around 163,000 members, sees AI as a “potential opportunity” to help professionals focus on “doing the things that really bring the joy of medicine” in an industry she said has a burnout prevalence rate of 10% among internal medicine physicians. But ACP president Jan Carney warned AI must not be viewed as a substitute for physicians. 

“That patient-physician relationship, it’s relational. It’s not simply between an individual and a computer,” Carney said.

Martin echoed the thought. 

“I still think there’s a humanism all about family medicine and primary care, that it’s going to be a one-to-one relationship for a long time, hopefully forever,” the AAFP CEO said. 

AI agents in anonymized health records

Harris, a practicing board-certified anesthesiologist with a specialty in obstetric anesthesiology, said he believes by the end of 10 years, AI could be doing “most of” medical diagnoses. AI is now involved with radiology, reading X-rays, CAT scans, and MRIs, the Maryland doctor noted. It is already being used to help speed up prior authorizations, though he warned that human judgment must remain in play.

A PA is a complex process that requires healthcare providers to obtain approval from insurance payors before delivering certain medical services, medications, or procedures. The 2024 American Medical Association of 1,000 practicing physicians reveals that 93% report care delays due to the PA process. A recent nationwide survey of health insurers found that most are already using AI systems for PA requests, including up to 12% that use AI to process automated denials.

“A physician or health care provider feels a patient needs a certain test or surgery or expense, even some medication — the insurance company will probably be running it through artificial intelligence to see if that’s an appropriate use,” Harris said. “I can see how again, you know, insurance companies can see that cuts down on what they … consider waste. On the other hand, I think it can empower patients to challenge the companies when they do deny care. Because you can go and present the facts, and you can present your insurance policy to a AI platform, and they can actually comment on whether or not, you know, you should have been denied that care.”

Harris argued the “real promise” of AI lies in giving it access to tens of millions of anonymized health records from Americans. The government has access to many patients’ health records, primarily through Medicare and Medicaid/CHIP. Harris revealed he is “having discussions with the government agencies that hold the data” on allowing AI agents like Claude and Gemini to access the anonymized health records.

“I would urge them that, if it’s truly anonymized data, to allow that access sooner rather than later, so that Americans get the benefit of this large database. I mean, it’s their data in the database; they should derive the benefit,” Harris said. “That information is very useful in terms of long-term public health, what we call longitudinal health, and that information is really only available by looking at large databases.”

Martin said a sweeping database already exists to a certain extent, pointing to the “massive repositories of data today” held by systems such as Medicare claims, UnitedHealthcare, and Elevance, which could use AI agents to sift through information. Bera said he thinks companies like Epic, which is the largest medical electronic medical record company in the country, will “largely become an AI company” analyzing such records.

“We will have to make sure they’re HIPAA-compliant, and a non-identifiable way for patients so you’re protecting patient privacy — but you’re learning — you’re now looking at a million patients learning about disease evolution and what happens over a lifetime,” Bera said. “Yeah, I think the more data you give to AI, the more it learns, and the better it becomes.”

Harris argued that the ramifications of the government allowing access to all the anonymized health information it holds would be “very, very powerful.” 

“You go in, and your blood pressure is slightly elevated, and your physician says, Well, I think this is the medication that is going to work for you,” he added. “And your physician has seen perhaps thousands of patients. But when you enter the information into the AI, the AI says, okay, Jane Doe, this is how old she is. This is the other medication she’s on. If you take that individual and you look into that 100-million-person database, you might find 10,000 people who are pretty similar, and that maybe a thousand are taking the same medication, and then you can see over time, is that medication effective at reducing blood pressure in that person? You can answer that relatively instantaneously from a database that is just huge.”

AI in drug discovery and FDA approval

Bera had a 20-year medical career and served on the House task force on AI. He noted that Microsoft has rolled out a diagnostic tool with 85% efficiency, with OpenAI recently announcing a similar tool. The California doctor said the system will be transformative, “if used properly,” as it could dramatically reduce the timeline for diagnosis and eliminate rounds of testing for patients. 

“If AI can go through all the literature, everything that’s out there, go algorithmically suggest for me as a doctor, it’s like, ‘Hey, doctor, you might want to order this test, because this is a problem.’ … That is a good thing, because it gets me to the diagnosis faster,” he said. “So it’ll improve efficiency outcome, reduce medical spending by reducing inappropriate testing.”

Bera emphasized AI’s impact on drug discovery, especially in oncology, by modeling cancer at the genetic level over time rather than just at a single moment. He envisions a genetic-level cancer registry that tracks how tumors evolve, allowing researchers both to forecast disease progression and reconstruct where it began.

“You could do genetic testing, theoretically down the road, that says, ‘Okay, you’re predisposed to this type of cancer based on our genetic modeling,’” Bera said. “I think that day is coming too, fairly quickly.” 

“It could improve efficiency, it could speed up the rapidity of diagnosis. It could reduce the administrative burden. It can save lives,” he said of AI. 

Harris said AI could have a significant impact on the FDA’s drug approval process, starting by speeding up the process for rare disease treatments to be authorized for public use, and then tracking their success. FDA applications are “literally 100,000 pages long,” and the agency is “already using AI to review those applications, to summarize the applications and to pull targeted answers when a reviewer has a specific question, he said.

THE AGE OF AI IS HERE: HOW SHOULD IT BE REGULATED?

After the treatment is approved, AI‑driven analysis of anonymized health records can accelerate and enhance the FDA’s post-marketing surveillance of drugs approved for public use, Harris said. He believes such a form of post-marketing surveillance could substitute some of the required pre-marketing testing, such as traditional randomized controlled trials, which are often “very expensive to do, and especially in the case of rare diseases, very difficult to do.” He said patients with rare conditions could hypothetically “get a certain treatment” sooner and then be tracked by AI agents embedded in their electronic records “to know whether those rare disease treatments were, in fact, effective.”

“I think it’ll allow patients access to those treatments earlier with the FDA, hopefully coming to the decision that we’re just going to follow more people after they get the therapy in real-world situations, which is actually the most important question to answer anyway,” Harris said.