The early evidence on AI and human expertise is in, and it points one direction: when professionals lean on AI tools, their own hard-won skills start to fade. That’s the finding making the rounds on Hacker News, drawn from research across medicine and computer science. The headline case is striking. Experienced doctors got measurably worse at spotting pre-cancerous growths during colonoscopies after they grew used to an AI assistant doing it for them.
This is significant because it’s not a story about junior staff or casual users. It’s about specialists at the top of their field losing ground.
What the researchers did
The core study followed endoscopy specialists in Poland, every one of whom had performed at least 2,000 colonoscopies. Researchers gave them an AI system that scans colonoscopy images in real time and flags adenomas, a type of precancerous intestinal lesion. The catch: the tool was available on some days and switched off on others. That setup let researchers watch what happened to the doctors’ unassisted performance once they’d grown comfortable with the AI.
The numbers tell the story:
- Before AI introduced: doctors found at least one adenoma in 28.4% of colonoscopies.
- After AI, working without it: the detection rate dropped to 22.4%.
That’s a six-point fall in unassisted skill in roughly three months. The work was published in The Lancet Gastroenterology and Hepatology.
Why it matters
The study authors put it bluntly. Constant exposure to AI can make clinicians “less motivated, less focused, and less responsible when making cognitive decisions without AI assistance.” In plain terms, the tool does the noticing, so the human stops noticing. When the tool isn’t there, the muscle has weakened.
What stands out here is the speed. We’re not talking about a slow decade-long slide. The drop showed up in a single quarter, among some of the most practiced specialists you could find.
The worry isn’t confined to one study. A survey of US health-care workers published this month found:
- 70% of nurses are concerned about losing skills to over-reliance on AI.
- 77% of physicians feel the same.
Those fears now have data behind them.
Beyond medicine
The pattern isn’t limited to hospitals. Researchers at AI firm Anthropic ran a randomized controlled trial with 52 software engineers performing a basic coding task. Everyone could search the web and read instructions; half were also prompted to use an AI assistant. It’s part of a broader push to measure whether “deskilling” is creeping into software work the same way it’s showing up in medicine.
Kevin Crowston, an information scientist at Syracuse University, frames the practical takeaway well: “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.
What you can do with this
The research doesn’t say drop your AI tools. It says be deliberate about them. A few practical moves:
- Decide what to protect. Pick the core skills your job actually depends on and keep practicing them by hand, AI off.
- Schedule unassisted reps. Like the Poland study’s AI-free days, build in regular sessions where you work without the assistant to keep your judgment sharp.
- Watch for autopilot. If you’ve stopped double-checking the AI’s output, that’s the warning sign the study describes.
The limitations
The researchers are careful not to overclaim. Co-author Yuichi Mori of the University of Oslo says more studies are needed to confirm the effect, and that no proven fix exists yet. “There is no established solution against deskilling right now,” he said. “It should be a very hot research topic in the next decade.”
That’s the honest state of play. We have early signals, not final answers. But the signal is consistent enough that anyone building AI into their daily work should treat their own expertise as something to maintain, not assume. More detail is available at the original source.