Do I unlearn things by using AI?

There are many fears around artificial intelligence, and most of them come from the outside: someone might misuse my data, automate my job away, watch me. This one is different. It arises from your own doing. And perhaps that is exactly why it sits so deep.

That is what it comes down to: if I let the AI write, calculate and think, then one day I will not be able to do it myself. I make myself dependent on a tool I do not control. And if it should one day be unavailable, or stop producing usable results, I will be standing there unable to do it myself.

How widespread this thought is can even be put in numbers. In a Bitkom survey in early 2026, 34 percent said they feared relying too much on AI and unlearning skills as a result. Whether it turns out to be exactly a third, a little more or a little less, honestly changes little for you. The only thing that counts for you day to day is something else: you are not alone with this worry.

First the uncomfortable part: yes, there is something to your worry

It would be dishonest to talk this worry away. What you stop practising gets weaker. That holds for mental arithmetic, for spelling, for writing an email and for writing code. Someone who has been handed every route by a navigation app for ten years will not necessarily find their own way through the city.

This site exists to meet fears with facts, not to pat them away. So: the effect is real. The only question is whether losing is the right word for it.

The pocket calculator has been here before

We had this discussion once already, almost word for word. When calculators arrived in schools, the worry was the same: the children will unlearn arithmetic. And many do recognise it in themselves – after years with a calculator, mental arithmetic comes less easily than it used to.

But look more closely at what actually happened. What declined was the doing: running the column of numbers in your head. What stayed is the judging: when a result appears on the display that is off by a factor of a thousand, you notice. You still know what order of magnitude to expect.

And after a learning phase, something arrived that had not been possible before: calculations you simply would not have had time for. The ability did not disappear. It shifted – from calculating to estimating – and what you can calculate has grown in both volume and range.

What this feels like for me

I have worked in quality assurance for around twenty years. My craft is taking software apart to see whether it holds. When I let an AI write code today, that craft has not become redundant. Quite the opposite: it is exactly what decides whether usable results come out at the end.

The AI quickly delivers something that looks plausible. Whether it really does the job, whether the failure cases are covered, whether it still works under load – you cannot see that at first glance. That is still my expertise. My tester's eye still goes in fully; it just applies at a different point than before.

Depending on how intensively you use AI, the feeling of having made something yourself can start to blur. Probably you have to come to terms with the idea that giving the right instructions is also an achievement you can claim for yourself.

Abilities rarely vanish, they shift

In programming the pattern shows up especially clearly. If an AI types code faster and more cleanly than I do, I will probably not get better at pure typing. Possibly even slower.

What grows instead is the layer above: how do I structure a system most sensibly? Where do I still need a human decision? What happens if parts of the system fail? That is architecture, and here a door opens: behind it the bigger tasks are waiting – because the bottleneck is no longer the typing.

On balance I build more today than I used to. Not because I can do more, but because what my ability is applied to has moved.

Where it does go wrong

Now the catch, and it matters because it carries the whole argument: this shift only works if you understand enough to be able to judge.

Someone who never learned mental arithmetic will not notice a misplaced decimal point on the calculator. Someone who never understood what a program is supposed to do cannot check whether the AI implemented it correctly. At that point it is no longer a shift but a handover. And then the worry is correct.

The difference is not whether you use AI, but what you keep while doing so. Hand over the doing if you like. Keep the judgement.

How to notice the balance tipping

Three questions to ask yourself about anything you regularly hand to an AI:

• Could you explain why the result is correct? Not produce it yourself – just justify it. If nothing comes to mind, you have handed over the judgement as well.
• Would you notice if it were wrong? That is the real question. An AI is rarely obviously wrong; it is plausibly wrong. Whoever can no longer see the error no longer has control, only trust.
• Could you manage without it if you had to? Slower, clumsier, with some looking up – but at all. If the answer is no and the matter is important, it is worth catching up on the basics.

Conclusion

The worry is justified, but it aims slightly off target. It is less about loss than about relocation. What you hand over gets weaker; what you need in order to judge what you handed over becomes more important. And usually something arrives that was not possible at all before.

Abilities are rarely lost; far more often they shift. What matters is keeping the ones that let you judge the result.

So engaging with AI does not simply make you unlearn something. You learn something else – necessarily, because handling the tool is itself an ability.

If the topic stays with you: in Fear of AI we look at the fear as a whole, and under the biggest AI fear drivers you will find the other worries broken down one by one.