AI May Not Be the Best Tool for Doctors All the Time

Imagine if relying on Google Maps too much slowly eroded your sense of direction. One day, you glance at a paper map and feel like you’re deciphering hieroglyphics. Now imagine that—but inside your colon.

That’s the uncomfortable implication of a new study that’s sending ripples through the medical community. No, it’s not about robots taking over. It’s about something quieter, subtler, and more human: the slow deskilling of doctors. Not because AI is replacing them—but because they’ve started to forget what it feels like to work without it.

The Colonoscopy Study That Should Worry You

At first glance, the news was encouraging. AI-assisted colonoscopies have been shown to improve detection rates of adenomas—those potentially cancerous polyps that hide in the bends of your bowel. Who wouldn’t want a digital eagle-eye backing up their doctor?

But the real story was buried in a follow-up study out of Poland, part of the ACCEPT trial. Researchers looked at how doctors performed colonoscopies without AI—before and after they’d grown accustomed to using it. The results were startling.

In 1,443 non-AI procedures across four endoscopy centers, doctors who had not yet been exposed to AI had an adenoma detection rate (ADR) of 28.4%. After AI had been introduced into their regular routines, but then removed for certain procedures, that number dropped to 22.4%.

That’s not a minor dip. That’s a 20% relative reduction in detection performance—just from the psychological effect of having gotten used to AI.

Put bluntly: using AI regularly made doctors worse at their jobs when they weren’t using it.

The ‘Google Maps’ Effect, In a Lab Coat

Dr. Marcin Romańczyk, the study’s lead author, likened the phenomenon to the Google Maps effect. You follow the blue dot long enough, you stop noticing landmarks. AI gives clinicians a sense of security, a second pair of eyes—until it doesn’t. When that assistance disappears, so does a chunk of their diagnostic sharpness.

It’s a subtle erosion, and that’s what makes it dangerous. No alarm bells. No flashing warning signs. Just a creeping complacency, wrapped in good intentions and clinical efficiency.

Even more unsettling? The AI-assisted procedures themselves only nudged ADR up to 25.3%—better than the AI-free post-exposure group, but still below the pre-AI baseline. So while AI “helped,” the long-term effect may be undermining the very skill it was meant to complement.

Why This Isn’t Just About Your Colon

Let’s zoom out. The implications go way beyond endoscopy.

Dr. Omer Ahmad of University College London pointed out that even a 1% drop in ADR can translate to a 3% increase in colorectal cancer risk for a population. A 6% drop? That’s not academic—that’s lives.

And if it’s happening here, where else might it be happening? Radiology. Pathology. Surgery. Anywhere doctors rely on visual cues and pattern recognition. If the human part of the process atrophies, you don’t just risk errors—you risk losing the very thing that makes clinicians effective in the first place: judgment.

And no, this isn’t about rejecting technology. It’s about the deeply human tendency to offload effort when given a shortcut. You don’t have to be a Luddite to recognize that convenience can make us careless.

Automation Isn’t the Enemy. Over-Reliance Is.

We’ve seen this movie before. Autopilot in aviation made flying safer, but it also introduced new risks—pilots who weren’t as ready to manually land a plane in an emergency. The tech got better. The humans got worse.

AI in healthcare risks following the same script. It starts as a co-pilot. But over time, doctors lean back. They trust. They tune out. And when they’re asked to fly solo again, they fumble.

The irony? AI was supposed to raise the floor of performance. Instead, it may be quietly lowering the ceiling.

What Smart Systems Need: Smarter Humans

So what do we do? We don’t ditch the AI. We ditch the delusion that AI can replace human vigilance.

Hospitals and clinics should introduce what Ahmad calls “AI-off” sessions—regular, intentional practice without tech support. Track individual detection rates with and without AI. If you’re slipping in solo mode, that’s your wake-up call.

Medical training also needs to evolve. Future doctors should learn how to work with AI, but also how to work without it—on purpose. Not because the machine will fail (though sometimes it will), but because skills are like muscles. Stop using them, and they weaken.

This isn’t just about keeping the scalpel sharp. It’s about preserving the very instinct to look closer, think harder, question more. That instinct isn’t coded into AI. It lives—and dies—in us.

AI May Hold the Microscope, But We Still Need to See

AI is not the villain here. It’s a tool. A very good one. But tools can dull our senses when they become crutches.

In the end, the scariest part of the Poland study wasn’t that AI made things worse. It’s that we didn’t notice we were slipping.

Until someone looked.

And that, more than any fancy algorithm, is still the heart of medicine: the ability to notice what others don’t. To catch the thing that isn’t blinking on a screen. To see.

Let’s not forget how to do that.