'An unstable situation': The AI trust problem

The AI trust problem

More than 1,000 artificial intelligence (AI) tools have received U.S. Food and Drug Administration (FDA) validation — but that number barely scratches the surface of what's actually being used in practice.

Medical Economics asked Richard Anderson, M.D., FACP, chairman and CEO of The Doctors Company and TDC Group, where clinician trust in AI most often breaks down. His answer centered on a paradox that has no clean resolution yet.

"Physicians have essentially no ability to thoughtfully purchase or select among the thousands of AI applications — which ones are better, which ones are dangerous and which are not," Anderson said. Large health systems can build dedicated evaluation teams, but even then, assessing how a tool fits into a broader care ecosystem is difficult. For independent and smaller practices, the challenge is steeper.

The deeper problem, Anderson said, is the collision between AI recommendations and the legal standard of care. When AI suggests something different from what a physician would traditionally do and the outcome is good, there's no issue. But when the outcome is adverse, the physician who followed that recommendation has, by legal definition, deviated from the standard of care. "Clinical adoption of AI is actually going to be slowed by this paradox," he said.

• Related content: ”The new malpractice frontier: Who’s liable when AI gets it wrong?”

Anderson expects the paradox to resolve itself eventually, as AI becomes woven into the standard of care itself — but he's clear that the path there will be uneven.

"How that happens will be piecemeal," he said. "It will be different venue to venue, and system to system."