By Alexis Kayser
Healthcare Editor
0
Share
Newsweek is a Trust Project member
See more of our trusted coverage when you search.Prefer Newsweek on Googleto see more of our trusted coverage when you search.
Artificial intelligence has quickly become part of physicians’ everyday work. For a profession built on trust, evidence and accountability, deciding how much confidence to place in the technology is a more complicated question.
Eighty-one percent of physicians now use AI professionally, more than double the share reported when the American Medical Association first surveyed doctors on the technology in 2023. More than three-quarters believe AI improves their ability to care for patients, according to the AMA’s 2026 survey of nearly 1,700 physicians.
But greater use has not erased clinicians’ concerns about what happens when AI gets something wrong. Eighty-eight percent of physicians said robust safety and efficacy validation is important for broader adoption, while 86 percent said the same of data privacy. Physicians also ranked clear liability frameworks among the most important regulatory steps for increasing their trust in AI.
Read More on Health Care
The more AI becomes part of clinical practice, the more important it becomes for clinicians to understand when, and why, they can rely on its recommendations.
Dr. Chris DeRienzo, chief physician executive at the American Hospital Association, framed trust and transparency as closely linked while moderating a panel at Newsweek's AI Health Summit on September 23 in New York City.
"It seems to me that you've got to find the right point at the intersection of trust and transparency in order to enable [clinical] judgment," DeRienzo posed. The health care leaders who joined him on-stage differed on what ultimately earns clinicians’ confidence.
“Transparency drives trust,” Dr. Matt Sakumoto, chief clinical product officer at the ambient AI provider Nabla, told Newsweek's gathered audience during the panel.
Sakumoto said clinicians need evidence behind the tools they use, including information about the data used to train models and how they have performed previously. The same questions matter for patients as AI becomes more deeply embedded in clinical care, he said.
Yet transparency does not necessarily mean that clinicians need to understand everything happening inside an algorithm.
Theo Zanos, head of the Division of Health AI at Northwell Health, argued that sustained performance may ultimately matter more for gaining physicians' trust. A tool can offer extensive information about how it was built, he said, but clinicians will quickly stop using it if it does not reliably work.
“Performance also drives a lot of that," Zanos said, both, "gaining trust and maintaining trust."
Research also suggests that distinction matters. A systematic review of studies examining explainable AI found mixed evidence that explanations themselves increase clinicians’ trust. Five of 10 included studies found explainable AI increased trust; two found no effect on trust, two showed mixed results and one was inconclusive. The authors concluded that simply providing an explanation does not automatically make clinicians more trusting of an AI system.
Patients appear to weigh performance heavily, too. A study published this year in JAMA Network Open found that AI performance was the factor most strongly associated with whether U.S. adults preferred an AI-assisted medical encounter. Having a clinician involved, representative training data and governance mechanisms were also associated with greater trust.
That puts pressure on health systems to make AI evaluation an ongoing process rather than a hurdle cleared before deployment.
Dr. Sophie Bakri, medical director of the Center for Digital Health at Mayo Clinic, said that AI is evaluated much like other clinical innovations at the world-renowned health system based in Rochester, Minnesota. Leaders consider why a tool was developed, its intended use and limitations, then validate its reliability and performance. But those assessments cannot end when a product goes live.
“We have to be prepared to stop. We have to be prepared to reduce usage, to pivot,” Bakri said, if concerns arise about reliability, privacy or safety.
At the health insurer Elevance Health, transparency increasingly means explainability, according its Chief Digital Information Officer Ratnakar Lavu. The company centralizes AI development through a platform that gives it visibility into which models are being used and how, then subjects clinical applications to a responsible-AI framework that includes testing for bias and hallucinations before and after deployment.
“We talk a lot about transparency, but the transparency comes through explainability,” Lavu said: “How did that AI actually come back with that response?”
Clinicians themselves remain central to that process. Eighty-five percent of physicians surveyed by the AMA said they want to be consulted or directly involved in AI adoption decisions.
Bakri said clinicians should similarly remain involved after deployment, continuously evaluating whether tools remain reliable as data and models change.
“The accountability rests with the clinicians as well as the organization,” she said. “As clinicians, we decide ultimately what we use.”
View 0 Comments
Submit Correction
View Editorial & AI Guidelines