Panelists explored the present and possible future uses of artificial intelligence in healthcare during the 2026 In Business Madison Healthcare Summit on Oct. 30, and the need for guardrails and collaborative governance.
The Healthcare Summit, held at the Park Hotel on Madison’s Capitol Square, featured panels on AI in healthcare and the local health tech startup ecosystem, and also recognized this year’s Healthcare Heroes, exemplary healthcare professionals nominated by their peers and colleagues.
The panelists on AI in Healthcare included Annette Fox, chief information officer for Group Health Cooperative-South Central Wisconsin; Becky Kohler, chief nursing informatics officer for UW Health; and Sean McGunigal, software development lead for AI research and development for Epic.
Before addressing questions to the AI panel, moderator Dr. Frank Byrne, president emeritus of SSM Health-St. Mary’s Hospital, said all sorts of issues have arisen around AI, but that’s not unusual for technology in general.
“What new technology in the history of humankind hasn’t had the potential to do both great good and great harm?” he asked.
Ambient AI
Most of the AI panel discussion centered on the healthcare benefits, and so far most of the benefits stem from the use of ambient AI, a technology used to record clinical notes and reduce time-consuming documentation for healthcare nurses and physicians.
Kohler said nurses are very protective of patients, so one of the very first things a nurse will ask about new technology, including ambient AI, is: How is this going to work for a patient?
“If it doesn’t work for a patient or if they get the least sense that’s not going to be in the best interest of a patient, (nurses) will reject it,” Kohler said.
Kohler said discussions about new technology must include its safety and reliability.
“The education around artificial intelligence has been really important because the nurses, again, want to know and understand and feel that that tool is going to be safe and reliable — not only for their use, but for the patient as well,” she said.
Fox, whose GHC-SCW team is responsible for everything related to the Epic electronic health record product, was asked whether a healthcare system that sits on both sides of healthcare — caregiver and insurer — changes the way it evaluates AI tools.
Fox said GHC-SCW takes a 360 degree view of AI governance because it’s wary of anything that could replace human judgment, and any new technology must be evaluated for potential cybersecurity risks and inherent bias.
With new systems, there are also cost factors to consider, including their effect on insurance premiums and whether the technology can actually lower costs and improve the quality of care.
Once adopted, Fox said technology must be monitored to make sure it does what it promises.
“I think we’re at an advantage being both a provider and a payer at the same time,” Fox said. “About a year ago, we created an AI governance committee that is a central place where all this evaluation is done. It is not an IT-only committee by any means.”
McGunigal, who has been with Epic for 10 years, started working on Chart with Art, the company’s ambient documentation product, two years ago. He was asked how Epic decides what to build.
“For us, it always comes down to the need,” McGunigal said. “We don’t sell a solution just because it’s fun to develop or it’s the latest and greatest tool. We jump to a tool because it solves a problem that we otherwise couldn’t solve.
“Fortunately, there are a lot of problems that AI can solve that were not solvable with the code that we would develop and release before that,” he said.
McGunigal agreed that cost is an important consideration for adoption because with AI, customers are paying for higher intelligence.
“So, if you’re in a position where you’re not going to recoup that benefit, it’s probably not the right solution,” he said. “It probably means you need to look to a cheaper model or not to AI tools.
“For us, it’s all about what’s the problem that we’re solving,” he said. “How valuable is that to an organization either in time saved or money recouped? And does that value justify the cost of using the intelligence to help assist in that way?”
Risk mitigation
Byrne, noting a recent tragedy in which a woman with epilepsy died after ChatGPT advised her to lower her medication, asked panelists how to mitigate AI risks.
McGunigal, quoting a prominent AI researcher, said AI models have “jagged intelligence,” meaning they are exceptionally smart at some things and exceptionally dumb at other things.
“The problem is that as a patient, you can be lured by the exceptionally smart things that the model’s telling you and ignore the dumb things that come out of it,” he said. “And so, at Epic it’s really our job to build the guardrails and safety rails around it so that you’re only dealing with the intelligence that’s capable and that it knows, and you’re not dealing with the intelligence that it doesn’t know.
“One of the big limitations with AI is that … it’s coming from the origin of a chatbot and the chatbot just wants to make you happy,” he said.
Emmie, which is Epic’s patient-facing AI product, tries not to give guidance or diagnose, which is the physician’s job, he said.
“Don’t try to go beyond what the provider has already done,” he said. “If there is any doubt, if there is any question, refer back to the human being. That’s one of the ways we’re able to put the guardrails around it.”
Fox said one of the things she goes back and forth on is whether the government should get involved in creating the guardrails.
“We have to find a sweet spot,” she said. “There have to be some kind of guardrails because if you just allow any AI out there to do whatever it wants to do, that puts the burden on us to try to figure out how safe it is.”
Fox said staff training and patient communication are big parts of AI implementation.
“All of us have Googled in the past about these symptoms,” she said. “What do we have? AI makes that even easier, but you have to use it the right way to not diagnose yourself, but to ask better questions.”
Kohler said UW Health has an AI governance council that has evolved as it gets more AI tools from Epic and other systems as well.
She said UW Health’s protective clinicians are engaged in UW Health’s AI governance.
“One of the things we think about when we think about safeguards is what makes me feel confident, and part of that is our … clinicians’ engagement in our AI governance and the fact that they are very protective (of patients),” she said.
Difference makers
Panelists were asked where else AI is making a difference in local clinics and hospitals.
Fox said Epic’s Insight, an AI tool that provides patient chart summaries, is an improvement over the old process of examining paper charts for relevant information.
“Using AI, it can pull all that together for you,” she said. “That’s been really a big time saver for our clinical staff. … Even if it only saves one minute per patient per visit, you see 20 patients a day. Times that by how many days in a week, how many days in a year, and that really adds up.”
One of the benefits of Chart with Art is that rather than facing a computer screen while typing in notes, doctors and nurses have more face-to-face interaction with patients, Fox said.
So, in addition to reducing the cognitive load with ambient notetaking, they can look at the patient for the entire visit.
“And on the flip side, we’re hearing lots of feedback from our patients that they noticed that too,” Fox said. “They noticed the more personal interaction with the provider during the visit, so that one has been an absolute gamechanger for us.”
Kohler also mentioned the easing of the cognitive burden.
“For inpatient nurses, the cognitive burden is one that has really stood out for those nurses. … The thing that we’re hearing from the nurses in particular is, ‘I don’t have to remember the thing that the patient told me an hour ago or when I saw them three patients ago.’ It’s that cognitive burden release that they’re really appreciative of.”
In radiology, UW Health has found AI value in incidental findings, where clinicians look for one thing and something else appears.
“AI can help to see that and bring it forward,” Kohler said.
McGunigal said one technology that is not in widespread use but that he’s really excited about is Epic’s Cosmos, which he described as “a big de-identified data set.”
“Lots of Epic organizations around the world have generously donated their de-identified data to help with medical research,” he said. “It’s basically leveraging AI to help improve patient outcomes and actually improve the field of medicine.
“It’s not going to save time today,” he said. “It’s not going to save lives even in the short term, but it has pretty broad implications. I’m very excited for it. I don’t see it really coming into the mainstream too soon, but it’s one AI technology that’s less talked about that I think is pretty impactful.”
What the future holds
Panelists were also asked what else people can expect from AI over the next three to five years.
Kohler said Cosmos holds promise for precision health. With the integration of biometric data and consumer data into the electronic health record, she said there is a lot to unpack.
“But the possibilities there are incredible with that amount of information,” she said. “There’s already so much information in the electronic health records, and the reason we’re seeing such possibilities around outcomes and so on is now with artificial intelligence and computing, now we can actually consume that information.
“So now with biometric information … there’s precision health and the ability to actually have more individualized care,” she said.
Fox wondered what patients will create with AI tools outside professional healthcare settings. She said families can create AI agents to plan a trip. She said she has a feeling that we’ll start to see that type of use in healthcare on the patient side for planning things such as their kids’ sports physicals, which she often got a late start on.
“And it’s kind of exciting in some areas, but it’s also scary because a lot of patients aren’t going to understand that once they take the information out of Epic and they put it into a different system, those (privacy) protections are gone,” she said. “So, I think there’s going to be a lot of education needed for people in this.”
Coming later this week: The health of the healthcare startup ecosystem in Madison.
