Healthcare AI Must Evolve Beyond Automation to Transform Care

Healthcare organizations invest two to three times more in artificial intelligence than other industries, yet most providers concentrate AI efforts on administrative tasks rather than clinical transformation. Senior health system executives are now calling for a fundamental shift in how the industry deploys AI technology.

Dr. Michael Pfeffer, senior vice president and chief information and digital officer at Stanford Healthcare, and Dr. Eric Alper, senior vice president and chief quality officer at UMass Memorial Health, outlined this evolution during the HIMSS AI in Healthcare Forum in Boston. Their message: the industry has barely scratched the surface of AI's clinical potential.

Why it matters

The healthcare industry risks creating a two-tier system where well-funded hospitals can afford advanced AI diagnostic tools while smaller community providers and safety-net hospitals fall behind. This digital divide could mean patients receive different quality of care based solely on which facility an ambulance delivers them to—a fundamental equity issue that threatens to stratify healthcare access by institutional resources rather than clinical need.

From queries to diagnostics

Stanford Healthcare has implemented large language model access directly within electronic health records, enabling real-time patient chart queries. One cardiology fellow told Pfeffer the tool "changed my life" during a particularly demanding service rotation.

But both executives emphasized that advanced diagnostic support represents AI's true future in healthcare. Alper envisions AI agents that review charts, close care gaps, and ensure patients receive appropriate screening—pushing healthcare toward Six Sigma levels of diagnostic accuracy.

"It's no longer about 'Should we use AI?' It's about 'Where should we use AI and how exactly should we use it?'" Alper said.

The governance challenge

UMass Memorial has established governance processes to evaluate AI risk upfront, but Alper acknowledged the organization lacks resources to monitor models in real time after deployment. This limitation highlights a critical industry challenge: clinicians cannot realistically audit every AI-generated output.

Pfeffer noted that users initially verify AI summaries for accuracy, but quickly begin trusting them without verification. "You see this summary once—it looks good. You see it twice—it looks good, and then you trust it forever," he explained. "We cannot put AI out there that clinicians have to check."

The solution, both leaders suggested, involves secondary AI agents that automatically monitor outputs and flag inaccuracies for human review, rather than requiring constant human oversight.

Sustainability through smaller models

The computational costs of large language models pose sustainability concerns. Pfeffer predicted the industry will need to shift from "throwing everything at these large language models to much more thoughtful, customizable, smaller and at-the-edge use cases."

Edge computing with smaller models could reduce both financial costs and environmental impact while maintaining clinical utility. This approach becomes especially important as AI moves toward precision health applications using genomic data for population-wide screening.

Pfeffer illustrated the utility challenge: identifying every patient needing colon cancer screening would overwhelm available gastroenterologists. Healthcare systems must balance model capabilities with actual delivery capacity and measurable outcomes.

The digital divide warning

Alper expressed concern about growing disparities between resource-rich systems that can build, experiment with, and monitor AI versus smaller community providers and independent clinics unable to access these transformative tools. He noted that UMass Memorial, despite its achievements, remains "underbedded" with full emergency rooms and access challenges.

Both executives acknowledged reliance on EHR vendors to help deliver AI tools more broadly, suggesting vendor partnerships may be essential to bridging the resource gap.

These details were first reported by Healthcare IT News, based on the executives' keynote presentation at the HIMSS AI in Healthcare Forum.