Studies warn: daily AI use may erode professionals' core skills

Research conducted with doctors and software developers suggests that relying on AI tools in daily work may gradually reduce professionals' ability to solve problems independently. Studies from Poland, the US, and San Francisco highlight a growing gap between AI-assisted performance and actual learning, raising concerns particularly for younger specialists.

New research involving physicians and software engineers is raising uncomfortable questions about the long-term effects of AI on professional competence. Studies published in recent years, and reported by Nature News, indicate that the convenience of AI assistance may come at a hidden cost: the slow erosion of skills that specialists need to function independently.

Doctors losing their edge

The most striking findings come from a study of Polish gastroenterologists. Researchers tracked what happened to experienced physicians, each with at least 2,000 colonoscopies under their belt, when AI assistance was taken away. During the study, doctors were given access to a real-time AI system that analyzes colonoscopy images to detect adenomas, precancerous bowel changes. The key twist: the tool was only available on certain days.

Before the AI system was introduced, the specialists identified at least one adenoma in roughly one-third of procedures. After they had grown accustomed to using the technology, their detection rate on days without AI dropped to approximately 22%. The pattern strongly suggested that reliance on the tool had dulled the physicians' independent vigilance.

Robert Wachter, a physician at the University of California, said the findings were telling. According to Wachter, they clearly show that even highly qualified experts can see their performance gradually deteriorate when leaning on AI. The study's authors suggest that constant exposure to comfortable technological aids may make doctors less motivated and attentive.

Yuichi Mori, a researcher at the University of Oslo and co-author of the study, acknowledged that broader follow-up research is needed before the hypothesis can be scientifically confirmed. Nevertheless, he argued that users of new AI systems should already be aware of the potential risks.

Software developers hit similarly

A parallel pattern is emerging in the tech industry. Researchers at Anthropic, the AI company based in San Francisco, conducted a randomised controlled experiment with 52 software developers. All participants were asked to complete a programming task and had access to the internet and relevant documentation. Half the group was additionally allowed to use an AI-powered virtual assistant.

After completing the task, all participants sat a test measuring what they had actually learned during the process. The gap was stark: those who used AI scored an average of 50%, while the control group, who worked without AI, averaged 67%. Developers who relied on the AI assistant struggled most with questions that required identifying errors in code, indicating they had not fully grasped the underlying principles of the solutions they had produced.

A concern for the next generation

Researchers say the implications are especially sharp for junior professionals and students. A troubling disconnect is forming between performance and learning: with AI assistance, results may look good on the surface, but the person may not actually acquire the knowledge or skills that produced them, because those came from the machine.

This is not the first time technology has rendered certain skills obsolete. Tapani Rinta-Kahila, an informatician at the University of Queensland, points to GPS devices as a classic example of how technology can gradually hollow out a natural human ability, in that case, spatial orientation. But with generative AI, he notes, something new is happening: the technology is no longer replacing just routine manual tasks, but increasingly taking over skills that require abstract thinking and the interpretation of complex data.

Rinta-Kahila's own research, published in 2018, examined accountants who had used automated accounting software for more than a decade. When the software was removed, many of the financial professionals were unable to carry out several basic tasks on their own.

What can be done

When AI tools handle even beginner-level work, researchers argue, a crucial part of the learning pipeline disappears, the practice through which new specialists develop foundational competence. This creates a risk of producing generations of programmers, lawyers, or accountants who do not adequately understand the basics of their own fields.

The solution, researchers suggest, is not to abandon AI but to use it more consciously. Rinta-Kahila stresses that people need to understand how algorithms work rather than blindly trusting the answers machines provide. The goal, he argues, should be to integrate AI in a way that preserves, rather than replaces, human critical thinking.