What if the smartest tool you’ve ever used is quietly making you worse at your own job? We’re glad you’re here, friend. Settle in, and stay with us to the end — the numbers in this story might change how you reach for that “Ask AI” button tomorrow morning.

AI Deskilling: When Smart Tools Make Us Forget How to Think

Picture a surgeon who has performed thousands of procedures. Hands steady, eye sharp, instincts honed over a decade. Now hand that surgeon a brilliant assistant that whispers the right answer at every turn. Helpful, right? For a while, yes. Then the assistant steps out of the room — and the steady hands hesitate.

That uneasy scene isn’t science fiction. It’s the early reading from fresh research, reported in Nature on 18 June 2026, and the signal is hard to ignore. As more professionals lean on artificial-intelligence tools, their hard-won abilities may start to fade. The word researchers use for it is deskilling.

We’ve written before about how machines reshape the way we reason, in our look at whether AI is reshaping how you think. This time the worry comes with measurements. Let’s walk through them together.

What does “AI deskilling” actually mean?

Deskilling is simple to picture. You practise a skill, you get good at it, you keep it sharp through repetition. Take the repetition away, and the skill softens. Athletes know this. Musicians know this. Now doctors and programmers are learning it too.

A recent survey of US health-care workers, published by Wolters Kluwer in 2026, put real figures on the fear. Seventy per cent of nurses and 77% of physicians said they worry about losing their abilities through over-reliance on AI systems. That’s not a fringe anxiety. That’s most of a profession raising its hand.

Kevin Crowston, an information scientist at Syracuse University in New York, frames the choice plainly. Knowing the phenomenon exists, he says, should push each of us to ask which skills we want to keep — and which we’re willing to hand over to a machine. Fair question. Let’s see what the data says before we answer it.

How fast can a doctor lose a life-saving skill?

Here’s the study that made people sit up. Researchers followed endoscopy specialists in Poland — physicians who use flexible probes to look inside the human body. Every one of them had already performed at least 2,000 colonoscopies. These were experts, not trainees.

They were given an AI system that scans colonoscopy images in real time and flags a precancerous growth called an adenoma. The catch: the tool was switched on some days, switched off on others. A natural experiment, hiding in plain sight.

The result was stark. Before the AI arrived, these specialists spotted at least one adenoma in 28.4% of colonoscopies. After they’d grown used to the tool, their detection rate on the AI-free days slid to 22.4%. Same doctors. Same skill. Lower numbers.

Six percentage points may sound small on paper. In a screening room, those points are missed growths that should have been caught. The study, published last October in The Lancet Gastroenterology and Hepatology, suggests even top professionals can get worse at the very tasks their jobs demand, once they grow dependent on a tool.

Robert Wachter, a physician at the University of California, San Francisco, and the author of a book on how AI is reshaping health care, reads the finding as a warning. The study’s own authors put it sharply: steady exposure to such tools can leave clinicians “less motivated, less focused, and less responsible” when they have to decide without AI at their side.

Mori, a co-author of the colonoscopy study, is careful. More research is needed to confirm the pattern, he says. Yet the message to anyone using these tools is direct: you risk losing some of your skill, so go in with your eyes open. We explored a cousin of this problem in our piece on algorithmic neo-paternalism in medicine, where the machine doesn’t just help the doctor — it starts deciding for the patient.

What happened when coders let AI do the work?

Doctors aren’t alone. Researchers at the AI firm Anthropic, based in San Francisco, ran a tidy little experiment on software engineers. They wanted to know a sneaky thing: when AI helps you finish a task, do you actually learn anything?

Fifty-two engineers were asked to do a basic coding task. Everyone could search the web and read instructions. Half were also nudged to use an AI assistant. Afterwards, all of them sat a quiz on what they’d just done.

The split was telling. The AI-assisted group scored an average of 50%. The group that worked without AI scored 67%. The AI users stumbled hardest on questions that asked them to find bugs in code — a sign they never really grasped the ideas behind the work they’d produced.

The coding study was posted on the preprint server arXiv, ahead of peer review, so we hold it gently. Still, Crowston finds it troubling, especially for students and early-career coders. He describes “a very odd disconnect between performance and learning.” People perform at a high level, he explains, by borrowing skills from the AI — without ever building those skills themselves.

Read that line twice. You can look productive while learning nothing. That’s the trap.

Why is generative AI unlike every tool before it?

You might shrug and say technology has always retired old skills. True enough. Tapani Rinta-Kahila, an information-systems researcher at the Hanken School of Economics in Helsinki, points to GPS navigation, which has quietly worn down our sense of direction. Who reads a paper map anymore?

His own earlier work tells a sharper story. Back in 2018, he studied accountants who’d relied on an automated, non-AI system for over a decade. When the tool was taken away, they’d forgotten how to do several routine tasks. The muscle had wasted.

So why treat generative AI as something new? Rinta-Kahila’s answer cuts to the heart of it. Earlier machines automated muscle and memory. Generative AI is the first technology to automate the cognitive work of thinking and interpretation — the faculties we long believed were ours alone.

That’s the difference. A calculator does your arithmetic. A chatbot can do your reasoning. And reasoning, once outsourced, is a harder thing to win back. He worries the next generation of programmers may never grasp the foundations of code, for lack of hands-on practice — and that the same risk shadows law, accounting, and other knowledge-heavy work. We touched a related nerve in our essay on whether AI is quietly stealing your choices.

Can we use AI without losing ourselves?

Now for the hopeful part, the part we care about most. None of these researchers is telling you to throw your tools in the sea. AI catches cancers that tired eyes miss. It ships code faster. The point isn’t fear — it’s awareness.

Rinta-Kahila offers a short, sane playbook. To guard against skill erosion, he says, do three things:

Know what you’re handing over

Notice how much thinking you offload. If the machine does every hard step, your skill never gets its workout. Keep a few tasks for yourself, on purpose, the way a pilot still hand-flies to stay sharp.

Understand the tool’s limits

Learn roughly how these models work and where they fail. A tool you understand is a tool you can question. One you don’t is one you obey. We unpacked that gap in our piece asking whether AI truly understands you, or is just a clever parrot.

Question the output, every time

Don’t trust an answer simply for arriving fast and sounding confident. Rinta-Kahila’s phrase is worth taping to your monitor: stay “mindfully vigilant.” Rely on the tool, and watch the tool at the same time.

That balance — leaning in while staying awake — is exactly the tightrope our President Gerd Dani walks in his reflection on robots, humans, and why AI is changing his mind. Use the machine. Don’t become it.

A quiet trade we should make on purpose

Let’s gather the thread. Skilled doctors lost ground on a cancer-screening task once an AI did the spotting for them. Engineers who leaned on AI performed well yet learned little. Accountants forgot the basics after a decade of automation. Across medicine, code, and the ledger, the same shape keeps appearing.

The deeper question isn’t whether AI is good or bad. It’s about agency. Every skill you outsource is a small piece of yourself you choose to set down. Sometimes that’s wise. Sometimes it’s a loss you won’t notice until the tool blinks off — and your steady hands hesitate.

So pick on purpose. Keep the skills that make you you. Hand over the rest with open eyes. That’s the trade worth making.

Frequently asked questions