Doctors and engineers may become worse because of AI

According to a survey released by enterprise solutions group Wolters Kluwer in early June, 70% of nurses and 77% of doctors are concerned that they will lose their skills because of over-reliance on AI systems.

This worry is well-founded. New evidence is showing that skill loss due to AI is starting to appear in medicine, computer science and other fields.

“Just being aware that this phenomenon exists, hopefully people will think about which skills they want to maintain and which skills they are willing to entrust to AI tools,” said Kevin Crowston, an information scientist at Syracuse University (USA).

A study conducted with endoscopists in Poland found that AI tools cause dependence and reduced skills. These doctors, all of whom had performed at least 2,000 colonoscopies in their careers, had access to an AI system capable of analyzing colonoscopy images in real time and flagging a type of precancerous lesion in the intestines, called adenomas. Doctors can only use the tool on some days and must examine themselves on other days.

In the 3 months before AI, doctors detected adenomas in 28.4% of colonoscopies. 3 months after the tool was put into use, the detection rate decreased to 22.4% without AI support, according to results published on The Lancet Gastroenterology and Hepatology.

Continuous exposure to AI tools can make clinicians “less motivated, less focused, and less accountable when making decisions without AI support,” according to the study authors.

The team notes that more research is needed to confirm this phenomenon, but those using AI tools should be aware that they risk losing part of their skills. Even highly skilled professionals can lose their skills when relying on AI tools.

Similarly in the field of computer science, researchers at Anthropic designed a randomized controlled trial in which 52 software engineers were asked to perform a basic programming task. All participants were able to search the web, but only half were suggested to use additional AI assistants.

The engineers were then asked to complete a quiz on what they learned from the task. The group that used AI assistants performed significantly worse than the group that did not use them, scoring an average of 50% compared to 67%.

AI-assisted people performed significantly worse on questions that required diagnosing errors in code, suggesting they did not grasp the concepts behind the code they had just created. The research is posted on the preprint database arXiv.

These findings are worrying, especially for students and new professionals. “People can perform at a fairly high level of performance because they are borrowing skills from AI, but not developing the skills themselves,” Crowston said.

Other technologies of the past have made certain skills obsolete, notes Tapani Rinta-Kahila, an information systems researcher at the Hanken School of Economics in Helsinki. For example, research shows that GPS navigation systems have impaired people’s navigation skills. However, generative AI tools are the first technology to automate many different cognitive capacities related to thinking and interpretation.

Rinta-Kahila once published a study of an accounting group that continuously used an automated, non-AI accounting system for more than a decade. When the tool was taken away, accountants forgot how to perform some routine tasks, suggesting a decline in skill when relying on the tool.

Not only in specialized tasks, AI also has an impact on common skills. A study by the Massachusetts Institute of Technology (MIT), published at the April conference of the Association for Computing Machinery, found that reliance on chatbots can impair skills in distinguishing misinformation.

In the study, 67 participants were asked to identify pairs of headlines and images as fake news or not. As a result, AI helped participants differentiate better and their ability to make correct decisions increased by 21%. However, the accuracy of self-assessment without AI support decreased by 15.3%.

“AI may help immediately, but it may ultimately degrade the ability to detect misinformation in the long term,” the study notes.

To prevent skill erosion caused by AI, people need to clearly determine how much work they are entrusting to AI tools, according to Rinta-Kahila. People also need to understand how generative AI models work, what their limitations are, and should avoid trusting AI output without asking questions.