As professionals across industries race to adopt artificial intelligence tools, a troubling question is moving from speculation into the realm of measured evidence: are we losing the skills these tools were meant to assist?
A report published in Nature finds that early results are in — and they are not encouraging. Reliance on AI-driven systems is beginning to erode the abilities of physicians, software engineers, and other skilled workers, raising urgent questions about how to preserve human expertise in an age of automation. For more context on this story, see our ongoing AI trends.
Doctors Perform Worse Without AI
One of the clearest examples comes from medicine. A study of Polish physicians specializing in endoscopy examined what happens when experienced doctors grow accustomed to an AI assistant and then lose access to it.
The physicians, all of whom had performed at least 2,000 colonoscopies during their careers, were given an AI system that analyzes colonoscopy images in real time and flags a type of precancerous intestinal lesion called an adenoma. The tool was available on some days but not on others, creating a natural experiment.
The results were stark. During the three-month period before the AI tool was introduced, the specialists found at least one adenoma during 28.4 percent of colonoscopies. After the tool was introduced, the adenoma detection rate for colonoscopies performed without AI assistance dropped to 22.4 percent.
In other words, once the doctors began relying on the AI system, their own diagnostic performance declined when it was unavailable. The findings were published last October in The Lancet Gastroenterology and Hepatology.
The study's authors concluded that continuous exposure to such tools can cause clinicians to become "less motivated, less focused, and less responsible when making cognitive decisions without AI assistance."
A Survey Shows the Worry Is Widespread
The fear of skill loss is not confined to a few studies. A survey of US health-care workers published earlier in June found that 70 percent of nurses and 77 percent of physicians are worried about losing their skills because of over-reliance on AI systems.
That anxiety appears to be justified by the emerging data. As more professionals integrate AI into their daily workflows, the gap between what they can do with the technology and what they can do without it may be widening.
Kevin Crowston, an information scientist at Syracuse University, framed the challenge plainly: "Just being aware that this phenomenon exists hopefully provokes some self-reflection about which skills people want to maintain and which they're willing to outsource" to AI tools.
Software Engineers Show Similar Patterns
The deskilling effect is not limited to medicine. To investigate whether skills are being lost in computer science, researchers at the AI firm Anthropic designed a randomized controlled trial involving 52 software engineers.
All participants were asked to perform a basic coding task. Every engineer could search the web and access instructions on how to complete it. Half were also prompted to use an AI assistant. The study, released as a preprint, was designed to isolate the effect of AI assistance on fundamental coding ability.
The findings add to a growing body of evidence that offloading routine cognitive work to AI can weaken the underlying competencies that professionals once developed through repeated practice. When the AI handles the heavy lifting, the human practitioner gets fewer opportunities to build and maintain mastery.
The Mechanics of Deskilling
Researchers describe deskilling as a gradual process. When a tool reliably handles a task, a worker's engagement with the underlying skill declines. Over time, the neural pathways and judgment honed through years of practice begin to fade.
In medicine, this can be especially dangerous. A physician who once relied on keen visual judgment to spot suspicious tissue may, after months of AI assistance, become dependent on the system's prompts. If the system fails or is unavailable, the doctor's unaided performance may no longer meet the standard it once did.
Yuichi Mori, a physician-researcher at the University of Oslo and a co-author of the colonoscopy study, warns that people who use AI tools should be aware they risk losing some of their skills. "There is no established solution against deskilling right now," he said. "It should be a very hot research topic in the next decade."
Why Standard Testing Misses the Problem
Part of the difficulty in addressing deskilling is that it is invisible until the AI is removed. Performance metrics look strong when the tool is in use, masking the gradual erosion of human capability underneath.
This creates a feedback loop. Workers perform better with AI, so they use it more. The more they use it, the less they practice the unassisted skill. By the time the decline becomes apparent — when the tool is down, retired, or replaced — the skill may have already atrophied.
The phenomenon also raises questions about how organizations should structure training and assessment. If professional competence is increasingly measured while AI assistance is active, evaluations may overstate what workers can truly do on their own.
Preserving Human Expertise
Researchers and workplace experts are now debating how to preserve important human skills in the AI era. The solutions are not yet clear, but several principles are emerging.
First, awareness matters. Simply recognizing that deskilling can occur may prompt individuals and organizations to deliberately maintain unassisted practice. Second, professionals may need to make conscious choices about which skills to keep sharp and which to delegate. Not every skill is worth preserving at full strength, but some — particularly those tied to safety, diagnosis, and critical judgment — clearly are.
Third, organizations may need to redesign workflows so that AI augments rather than replaces core cognitive work. Instead of letting the tool do the thinking, AI could be positioned as a check on human judgment, requiring the professional to arrive at an answer first before consulting the system.
A Defining Challenge for the AI Era
The Nature report makes clear that deskilling is no longer a hypothetical concern. The evidence, while still early, is consistent across fields: when humans hand over cognitive work to AI, the underlying skills can weaken.
For the millions of professionals now integrating AI into their daily routines, the stakes are personal as well as societal. A surgeon who loses the touch to spot a tumor, an engineer who forgets how to debug without an assistant, a nurse whose clinical intuition dims — these are not abstract risks.
As Mori noted, there is no established solution yet. But the first step, researchers agree, is taking the threat seriously. The early results are in, and they are a warning: the tools designed to make us better may, if used carelessly, leave us less capable than before.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
