At UCI Health, the promise of AI is not measured by hype or speed, but by a simple question: “What problem are we trying to solve?”
Deepti Pandita, MD, UCI Health’s chief medical information officer, said that question guides nearly every AI discussion inside the Irvine, Calif.-based academic medical center, particularly those surrounding AI governance.
“We don’t look at innovation for the sake of innovation,” Dr. Pandita said. “If there is a burning problem we need to solve, then we look at AI more thoughtfully. But everything has to follow the same governance process.”
Any AI proposal — whether floated by a researcher on campus or a frontline provider — first passes through a rigorous procurement review. Vendors must complete an extensive questionnaire covering data privacy, algorithm transparency and potential bias. The responses are then vetted not just by IT staff but also by data scientists and AI researchers.
If the proposal clears those hurdles, it advances to the AI Governance Committee, which Dr. Pandita leads. The committee’s range of skill sets and job functions is deliberately broad: The chief medical officer, chief nursing executive, compliance officers, legal counsel and operational leaders all weigh in.
“Everyone who will be responsible for using, implementing, maintaining or governing the AI tool has a seat at the table,” she said.
For Dr. Pandita, AI is not a switch to be flipped but an iterative process. Tools must be monitored constantly for regression, bias and reliability. Even when using algorithms embedded in the electronic medical record system, the organization runs an internal “second pass” to verify safety and transparency.
“You have to be good stewards of your AI product,” she said. “It’s not something you just let loose in your system.”
That vigilance stems from both optimism and caution. Within the next three to five years, she believes AI will take on many of the administrative burdens that plague physicians — scheduling, documentation or inbox management.
“The scary part is if AI gets automated to the extent that it’s functioning independently and we don’t have good guardrails,” she said.
Already, the health system is seeing measurable benefits. Ambient scribe technology has been “very, very well received” by physicians, allowing them to finish charting on time, reduce burnout and reclaim hours once lost to late-night “pajama time.”
Another AI-enabled tool helps triage patient messages in the electronic inbox, with algorithms drafting responses that physicians then edit. Nearly 40% of messages are now generated by the system, saving providers countless clicks and keystrokes.
Perhaps the most ambitious project, though, is still in development. To avoid patient data leaking into open-source models, Dr. Pandita and her team are building their own large language model housed entirely on internal servers.
“For the length-of-stay problem, this will be a game changer,” she said.
The model is currently being tested against guardrails for accuracy and safety, with hopes for implementation in the coming months.
As AI reshapes workflows, Dr. Pandita sees less of a role change for clinical informatics leaders than a need to upskill.
“Informatics is at the intersection of people, process and technology. We are changing processes and technology with AI, so of course the ‘people side’ has to change,” she said. “Leaders must now be fluent not only in translating tech speak into clinical language, but also in navigating the complexities of machine learning.”
