AI makes doctors 20 % less efficient, reveals a disturbing study

AI assistance tools have become a daily resource that has increased productivity and speed in many industries. But despite their advantages, medical experts are concerned that an excessive dependence on artificial intelligence ends up causing more wrong than good.

IA dependence could reduce detection rates during colonoscopies

A recent study conducted with 1,443 patients revealed that endoscopists using AI agents during colonoscopies obtained a lower detection rate when they were then working without these tools.

The results, published this month in Lancet Gastroenterology & Hepatology, show that doctors reached a detection rate of 28.4 % of potential polyps With the support of AI. Without this aid, the rate fell to 22.4 %, a decrease of 20 %.

Dr. Marcin Romańczyk, gastroenterologist at the HT medical center in Tychy, Poland, and the main author of the study, said he was surprised by these results. He pointed out excessive dependence at AI as one of the major factors explaining the drop in performance.

We have learned medicine from books and from our mentors. We observed them, they guided us. And now there is an artificial tool that tells us what we should do, where to look … and in reality, we do not know how to behave in these situations

Dr. Marcin Romańczyk

This study does not only highlight the potential negative effects of too much confidence in AI: it also illustrates the evolution of medical practice, which goes from an analog tradition to a digital era.

Productivity gains at a cost price

Beyond the medical sector, AI automation is now omnipresent in working environments, where it is used to improve efficiency. Goldman Sachs even provided for in 2023 that AI could Climb productivity up to 25 %.

But this massive adoption has a reverse. Research by Microsoft and Carnegie Mellon university has shown that if theIA effectively increased efficiency From a group of knowledge workers, it also weakened their analytical skills and their “judgment”, which tended to atrophy.

The precedent of aviation: flight lessons Air France 447

The aviation sector has long illustrated the risks of excessive confidence in automation. In 2009, the Air France 447 flight connecting Rio de Janeiro to Paris crashed into the Atlantic Ocean, causing the death of 228 people.

The investigation revealed that the automated device system had transmitted erroneous information Due to dysfunction. The pilots, which are not very trained in manual piloting, have relied on automated indications rather than correcting the trajectory.

Find the balance between AI and human expertise

For Lynn Wu, a professor associated with the Wharton School of the University of Pennsylvania, these examples must serve as a warning. It insists on the need to maintain human expertise in the sectors with strong security issues.

What is important is that we learned lessons from this history of aviation and the previous generation of automation: AI can really improve performance. But at the same time, we must preserve these critical skills, so that when AI does not work, we are able to take over.

Lynn Wu

She added that if people lose their own skills, artificial intelligence will also work less well. For AI to progress, individuals must also continue to progress.

Romańczyk also accepts the use of AI in medicine, noting that “'The will be, or is, part of our life, whether we like it or not. He nevertheless highlighted the need to understand how artificial intelligence affects human thought and urged professionals to determine the most effective means of using it.