At the opening keynote of Becker’s 11th Annual Health IT + Digital Health + RCM Conference in Chicago, two health system technology leaders set up the argument that would carry through the rest of the week: The hardest problems facing AI in healthcare right now are not technical.
Bobbie Byrne, MD, executive vice president and CIO of Charlotte, N.C.-based Advocate Health, and Mitchell Schnall, MD, PhD, senior vice president for data and technology solutions at Philadelphia-based Penn Medicine, fielded questions from moderator Giles Bruce, a Becker’s assistant editor and conference co-chair, on where AI is delivering results for their systems and where it is still running into resistance. Four takeaways stood out.
1. AI will raise costs before it lowers them.
Dr. Schnall pointed to a roughly 30% efficiency gain in how Penn Medicine now handles pharmacy prior authorizations, translating into more captured pharmacy revenue. But Dr. Byrne pushed back on the idea that AI’s overall effect on healthcare spending will be to lower it.
“I do think AI in general is going to raise the cost of healthcare, not for our organizations, but in total,” Dr. Byrne said.
She argued the industry needs a different yardstick than raw dollars saved, pointing to a quality-adjusted life year, or QALY, as a fairer way to weigh cost against outcome.
2. Nursing units are where AI adoption stalled first.
Dr. Byrne described an early attempt to bring ambient, voice-based documentation into inpatient nursing care, expecting nurses to narrate their work conversationally the way physicians do with ambient scribe tools. It did not go as planned.
“We were really surprised at how little care is done in the hospital room by voice,” Dr. Byrne said.
She said nurses largely document through structured flow sheets rather than narrative notes, and a bedside interaction does not naturally lend itself to being spoken aloud the way a physician exam does. Advocate paused the rollout and built what Dr. Byrne called a “care out loud” program, coaching nurses to narrate their care so ambient tools have something to capture. She said staff hesitation on AI projects generally starts the same way.
“I think it starts with fear,” Dr. Byrne said. “Everybody kind of comes at it a little fearful, because they’re also thinking, this is going to replace me, and I have a family to feed.
“We’re not replacing jobs right now. We have so much more need than we have people, so that’s not really an issue.”
Dr. Schnall said Penn Medicine is careful not to promise staff their roles will stay static, even as it commits to retraining them.
“Jobs changed. My job changed. Everybody’s job changed,” he said. “Those are going to change at a faster pace.”
He said the health system’s commitment to employees is training for whatever those new roles become, not a guarantee the roles will hold still.
3. Patient trust in AI is decreasing even as patient use of it rises.
Dr. Byrne said Advocate has tracked a steady, month-over-month increase in patients connecting their MyChart records to outside AI tools such as ChatGPT, even as national survey data, including from Pew Research Center, shows rising skepticism toward AI.
“It is true that even as we see that people are more distrustful of AI, they are using AI more,” Dr. Byrne said. “Sometimes it’s the people who are highest users of AI who are also the most anti-AI.”
Dr. Schnall said the same split shows up in how patients want to receive care.
“Not all patients embrace interacting with the chatbot or embrace AI actively in their care,” he said. “There’s no way you can care for patients today with no AI. It’s so embedded in so many pieces, you can’t.”
He said that means systems will need to keep a more traditional path available for patients who want one, even while building AI further into the default experience.
4. Cybersecurity, not clinical AI, will be at the forefront a year from now.
Asked to look to a year from now, both said without hesitation that cybersecurity would dominate the landscape. Dr. Byrne said the speed with which threat actors can now access sophisticated, AI-driven attack tools is compressing the timeline organizations have to respond.
“Zero day is not going to be zero day,” she said. “It’s going to be zero hour.”
