Doctors are still skeptical of AI—health systems are trying to change that

Katherine Fung

By Katherine Fung

Senior Writer

See more of our trusted coverage when you search.Prefer Newsweek on Googleto see more of our trusted coverage when you search.

Only about a third of Parkview Health's clinicians are enthusiastic about artificial intelligence. Another third are indifferent. The rest have tuned out entirely.

That assessment, offered by Dr. Mark Mabus, the chief medical informatics officer and senior vice president of electronic health records at Parkview Health in Indiana, during Newsweek's March 5 webinar, "Outdoing Dr. Google: How to Answer Patients' Questions Before the Internet Does," summed up the challenge ahead for health systems.

Mabus and his fellow panelists—Dr. Bridget Duffy, co-chair of the PatientAI Collaborative, Dr. Atiq Bhatti, the faculty administrator and clinical research operations lead in the department of medicine at Eastern Virginia Medical School and Virginia Health Sciences and Dr. Tanvi Jayaraman, the clinical lead of Health AI at Oura—agreed that 2026 is less a year for deploying new AI tools and more a year for bringing the medical profession up to speed on the tools already here. Thursday's discussion was moderated by Newsweek Health Care Editor Alexis Kayser.

Read More on Health Care

"This is the year we need to get everybody on board and educate," Mabus said. "Maybe we move that dial to half-and-half, or 60/40, but our clinicians have a bunch of AI thrown at them. They need to know how to best utilize this."

Patients, on the other hand, are already using AI, whether their doctors are ready or not. A 2025 study from the University of Pennsylvania's Annenberg Public Policy Center found that roughly 8-in-10 American adults go online to answer questions about their health conditions or symptoms, and nearly two-thirds find the AI-generated responses to be somewhat or very reliable.

Bhatti's own research has found that, on average, the gap between when a patient begins searching for information and when a patient can get in to see a provider is 24 days. For Bhatti, that's not a patient behavior problem, but a health system design failure. He called the source that patients turn to first as the "digital front door."

"There's a huge gap, not only on the side of what we are building in terms of the tools and the technology we have, [but also] that technology doesn't reflect the patient's needs and the data and the demand signal that patients are putting in front of us," Bhatti said. "That's where we have to intervene."

Part of that intervention means building AI tools that clinicians can actually trust and stand behind. Jayaraman said this means picking and choosing what training data to feed the AI and opting for information that can be held to a clinical standard rather than giving it "just everything on the internet." She used Oura's new women's health AI model as an example.

"We have curated that set of resources that the model has been trained on," Jayaraman said. "We have board-certified OB-GYNs on staff who have made sure these are resources and textbooks and guidelines that they rely on day-to-day in their practice and they would feel they could stand by when a patient is reading up on their health information."

Dr. Duffy's Patient AI Collaborative was built to do just that. Duffy has spent the past several years building "a checklist of always events" that health systems can use to evaluate whether a patient-facing AI tool is safe, culturally informed and clinically sound.

Citing her own experience having watched loved ones learn devastating diagnoses from an app and turning to ChatGPT in a panic, Duffy said, "We took an oath as doctors to first do no harm. And I thought to myself, could we not create that standard for how AI interacts with patient-facing communication?"

Mabus said that Parkview is already moving in that direction, giving AI tools to all of its graduate medical education residents so that the next generation of physicians doesn't have to unlearn a resistance to technology that their predecessors developed. The harder lift, Thursday's panelists said, is not by dismissing what patients are finding on their own, but engaging with it.