Doctors using AI may “forget” some skills: Shot

Artificial intelligence is beginning to help doctors screen patients with some routine diseases. However, new research raises concerns about whether doctors are too dependent on AI.

A study looking at Polish gastroenterologists found that after becoming accustomed to using AI-assisted systems, it appears that it is about 20% worse to spot polyps and other abnormalities during colonoscopy.

Research results published in the journal Lancet Gastroenterology and Hepatology; We suggest that even after a short period of time with AI, experts can become overly dependent on AI to do certain aspects of the job.

“We were very surprised,” says Marcin Romańczyk of MD-PH.D. He was a gastroenterologist at the HT Medical Centre in Tykie, Poland, who led the research.

However, not everyone is convinced that this paper proves that doctors have lost critical skills due to AI.

“Three months seem like a very short period of time to lose skills that took 26 years to accumulate,” says Johann Haleman, a researcher at the University of Manchester, who relies on humans for artificial intelligence.

Hulleman believes statistical variation in patient data may be part of an explanation for why numbers are falling. Factors such as the mean age of patients used in different sections of the study may explain the variation, he says.

Medical AI

Artificial intelligence is becoming more and more common in many everyday medical scans. The next time you get a scan for breast cancer or colon disease, your AI may be analyzing the images.

“AI is everywhere,” says Romachik. At the same time, many doctors are catching up. Because learning how to use technology was not part of the training.

“We've been taught by books and teachers,” he says. “No one could tell me how to use AI.”

A few years ago, four Poland clinics attempted AI systems to detect polyps and other abnormalities during colonoscopy. AI works in real time and analyzes video from the camera in the colon.

If you find something, emphasize the area your clinician sees.

“This particular one has a green box that shows where the polyps are,” he says.

In the box

The clinic was collecting data on whether the AI system worked. That was true, but when Lomakujik and his colleagues recovered the data, they found something else. After the system was introduced, doctors got significantly worse when AI was turned off.

According to their analysis, Once the new AI system was turned off after the doctor acquired AI, the rate of possible polyp detection fell from 28.4% to 22.4%. In other words, doctors seemed to immediately rely on AI systems that caught polyps. Romasik says he's not quite sure why it's happening, but he has some theories.

“We unconsciously wait for the green box to come out and show us the area where the polyps are.

There are other examples that support that idea. Similar studies have shown that if you know that nonexpert can get an AI system that helps you press a button, you will do a worse job than scanning mammograms.

Johann Halman, who led the mammogram research, described it as “safety effect.” He says these latest results may be interesting, but he is skeptical. Colonoscopy studies were conducted over three months, with participating physicians having decades of experience. He believes there are statistical variations as many factors, such as the age of the patient, can be behind the obvious decline.

He further says, “I don't know the basic truth because I don't know how many polyps there are actually.” This means that of the polyps that doctors may have missed are unknown in fact medically significant.

The study author, Romańczyk, believes the drop is authentic, but admits that studying AI in such a clinical setting is difficult. There are many variables that the researchers could not control.

He is not against the use of AI. He actually thinks that a small green box will help him do a better colonoscopy. But he believes there should be more research like these examining how AI is changing the way doctors work in the real world.

“Look at what's going on,” he says. “There are AI systems available, but there is no data.”