Imagine you’re a great chef who decides to marry and live with a person who happens to be an even better chef. Naturally, they start doing most of the cooking.
A few years later, you’re very happy with your diet, but your cooking skills have atrophied. You’ve forgotten your favorite recipes and lost your instinct for when a steak is perfectly seared, or a soup is over-salted.
Another example that might be more relevant to people reading this: If you’re one of the millions of people who rely on GPS to get around, how confident are you that you can still navigate using a map?
This is the principle of AI deskilling. And worryingly, research is starting to show that it’s happening more quickly than we’d thought.
Deskilling itself isn’t new; skills have always come with a “use-it-or-lose-it” clause. With AI, though, the risks are magnified simply because it can potentially replace so many skills.
Day to day, millions of us use it to write, create images and generate code. These are skills that can take years to perfect, but are we risking throwing them away if we rely on machines?
How real is the threat, and more importantly, short of a total self-ban on using AI, what can you do to prevent it from happening to you?
Is The Threat Real?
From the evidence, it seems that fears are well-placed. A 2026 survey of US healthcare workers found that 74 percent of clinicians are worried about losing skills due to over-reliance on AI.
More direct evidence comes from a study published by The Lancet last year, which found endoscopists who reverted to using non-AI methods became seven percent less able to detect adenomas after just three months of AI use.
That study notes that frequent use of AI assistants appears to make clinicians “less motivated, less focused and less responsible when making decisions.”
Outside of medicine, Anthropic, developers of the Claude AI platform, found that software engineers using an AI tool completed their work slightly faster but scored lower on follow-up comprehension tests than those who didn’t.
And researchers at MIT coined the term “cognitive debt” to describe a long-term decline in neural network connectivity and memory recall observed among writers using AI during a trial.
Researchers have often taken care to point out that these are early results and we don’t know how the advantages offered by AI, such as increased speed and access to information, offset these downsides.
But it certainly suggests that this is a danger we should be aware of, and we should be doing our best to avoid falling victim to it. So what can we do?
So What Can We Do?
Digging a little deeper into some of the research, it becomes apparent that the biggest risks are around how we use AI, rather than whether we use it.
The Anthropic researchers found the negative impact on skill was lessened among participants who acted in a way that meant they stayed cognitively engaged while learning, solving problems and making decisions with AI.
Going back to our home chefs, it’s the difference between the competent chef watching and learning as the excellent chef makes dinner, and putting their feet up and waiting to be served.
This gives us a good place to start thinking about solving this problem. So here are some pieces of practical advice on implementing it in your own AI use.
First, understand what’s at risk. What are you relying on AI for that you used to do just fine (if more slowly) than before?
If this is something central to what you are or do (writing, communicating, coding, design), then you won’t want to lose it. The human qualities you bring to these tasks are what set your work apart from the work of a machine. Without them, your value proposition drops dramatically.
Next, always remember AI is the assistant and humans are responsible for doing the work. That means taking the lead, forming your opinion and being certain of what you want to do before turning to AI. Going in with vague ideas about what you want and expecting AI to give it purpose and direction is a recipe for low-quality, generic results, often called “AI slop”. You’ll be less engaged with it because it isn’t your own work.
It’s also a good idea to sometimes remove the safety net of AI. Force yourself to finish jobs or carry out tasks without using AI, for the valuable purpose of knowing you still can. If you’ve been using AI heavily for a while, gathering data for a report without ChatGPT’s ability to summarize dozens of sources into a couple of paragraphs can seem daunting. But make sure you do it from time to time anyway to keep the gears working.
Finally (for now, as I’m sure this is a subject I’ll come back to), human oversight of AI output is as critical in this instance as it is in many others. We know it’s essential for mitigating the ever-present danger of AI hallucinations. But it’s also necessary to ensure we understand what AI is doing for us and that whatever it’s helping us to create is in line with how we’d build, design or write it ourselves, when steered by our human insight and values.
That’s just a few examples of ways that maintaining cognitive engagement both protects us from deskilling and keeps our work authentic. It involves being honest with ourselves about our AI use and creates a sound footing for building defenses against decline.
So What’s Next?
None of this means abandoning AI. The research is new, and there’s no long-term data. There’s also evidence that shows using AI is helping some people work faster and to a consistently higher standard.
But while the benefits to productivity and cost-saving potential of AI can’t be ignored, neither can the risks of losing something of what makes us human.
I believe this shows us that the leaders and professionals who thrive in an AI-driven world won’t be those who use AI most, but those who use it best.
Rather than simply how to use it, focusing on when it can be trusted, when we should challenge it or push back, and identifying what it might have missed can help to keep us sharp. This should be considered an essential aspect of AI literacy that everyone should understand.
