Madison, Wis.-based UW Health is reviewing more than 100 AI features arriving through an upgrade to one of its core technology platforms, illustrating the growing governance demands health systems face as AI capabilities become embedded in enterprise software.
Frank Liao, PhD, senior director of digital health and emerging technologies at UW Health, discussed the upgrade Sept. 14 during the panel “Building a Foundation: Key Considerations for Scaling AI in Healthcare” at Becker’s 11th Annual Health IT + Digital Health + RCM Conference in Chicago.
Dr. Liao said the upgrade spans multiple software versions and includes more than 100 AI features. UW Health is using its existing governance structure to review the capabilities and sort them into low-, medium- and high-risk categories, with only a small number falling into the highest-risk group.
“Governance actually allowed us to go faster,” Dr. Liao said.
Rather than having individual departments independently evaluate AI capabilities as they are released, UW Health can use a centralized process to assess them across the organization, he said. Dr. Liao described the approach as creating a “paved road” for AI, with governance intended to establish trust and safety without unnecessarily slowing adoption.
The health system has also expanded its AI governance structure as generative AI use has grown. What began with a heavier clinical focus now includes representatives from areas such as human resources and other administrative functions. UW Health has also updated how it evaluates large language models and increased the frequency of its governance work.
That oversight continues after deployment. Dr. Liao said responsibility for an AI tool’s outcomes ultimately needs to reside with the part of the organization using it.
“Every solution … has to live in some part of the business that’s accountable for the outcomes,” he said.
That accountability is becoming more important as health systems move toward agentic AI, he said. Unlike conversational AI, agents can integrate with systems and perform actions. UW Health is therefore looking not only at adoption and business outcomes, but also at what agents are doing and whether those actions align with what operational leaders expect.
Other panelists described similar efforts to bring greater structure to AI deployment. Paul Lukac, MD, chief artificial intelligence officer at Los Angeles-based UCLA Health, said the health system began holding an AI review meeting about nine months ago. The meeting takes place every two weeks and brings IT leaders together to review AI implementations requiring IT leadership involvement, giving the organization greater visibility into its AI portfolio.
