When researchers at four medical centers in Poland introduced AI to help endoscopists detect precancerous polyps, the tool worked. Detection rates improved. But the same endoscopists, when tested without the AI, performed measurably worse than they had before the technology arrived.
That finding — from a study published in Lancet Gastroenterology & Hepatology — is the kind of data that Benjamin Schwartz, MD, president of academic delivery at Phoenix-based Banner Health, is watching closely as health systems accelerate AI adoption across clinical workflows.
“Before AI was introduced, the same endoscopists were detecting adenomas at a rate of roughly 28%,” Dr. Schwartz said in a recent “Becker’s Healthcare Podcast” interview. “And after AI was introduced, which was intended to help them detect suspicious lesions, when they did it without AI, their detection rate dropped to 22%. So what’s implied by that is that there’s a risk that, relying on the machine, you lose the practice of the skills that you’ve had and that you might lose them.”
The study involved roughly 1,400 procedures. The AI was working as designed: total detection rates held because the tool was present. What changed was something harder to measure — the clinician’s independent ability when the tool was not.
The pattern is playing out in Banner’s own AI decisions, including a debate the system had over ambient listening technology, tools that capture caregiver-patient interactions and automatically populate the clinical record. The productivity benefits of those tools are clear. The training implications are more complicated.
“We debated quite a bit whether or not the trainee should have that for fear that they would never learn the skill of writing those notes and without that skill, they would become completely dependent on the machine,” Dr. Schwartz said.
But the deskilling fear shouldn’t stop progress. Technology continues to evolve and the skills behind it change as well, and health systems are balancing the forward progress with ensuring their teams have foundational skills.
“The outcome of that discussion was that they should have the technology,” he said. “But we go through every AI intervention through that scrutiny to make sure that we’re not creating a next generation without any of the basic skills that would be needed if the machine weren’t there.”
At the Becker's 11th Annual IT + Revenue Cycle Conference: The Future of AI & Digital Health, taking place September 14–17 in Chicago, healthcare executives and digital leaders from across the country will come together to explore how AI, interoperability, cybersecurity, and revenue cycle innovation are transforming care delivery, strengthening financial performance, and driving the next era of digital health. Apply for complimentary registration now.
