Why AI Can Make Us Dumber — and How to Keep From Unlearning How to Think

AI doesn't automatically make us dumb — but it can wean us off thinking

The claim "AI makes you dumb" is deliberately provocative. A more accurate version would be: excessive dependence, too little practice of our own and the uncritical acceptance of automated results can cause us to lose certain abilities. What matters is not whether we use AI, but whether it supports our thinking or replaces it.

Imagine two people learning a new programming language. Person A first tries to solve a problem on their own, reads the documentation, makes mistakes and tracks down their causes — only when they get stuck do they ask the AI for a hint. Person B simply describes the desired result, takes the finished code and pastes every error message back into the chat.

Both end up with a working application. But Person A has gained problem-solving experience, while Person B remains reliant on the AI delivering a suitable solution next time, too. The danger point is reached when we produce results whose creation we can no longer follow ourselves.

Why effort is part of learning

The feeling of having understood something is not the same as being able to apply it on your own later. Lasting learning comes from active mental processing: retrieving information from memory instead of rereading it, putting explanations into your own words, working on problems without a ready-made solution, analyzing mistakes and applying knowledge in new situations.

Learning psychology speaks of the testing effect (retrieval practice) and of "desirable difficulties": methods that feel more laborious in the short term lead to better retention in the long term. This is exactly the effort an AI can cut short when it delivers the finished answer before we have thought for ourselves.

A randomized study of 120 students (Barcaui, 2025) offers a concrete indication: in an unannounced knowledge test 45 days after studying, the group that had used ChatGPT as a learning aid scored an average of 57.5% correct answers — the comparison group without AI scored 68.5%. The study has limitations and can't be generalized to every subject and way of using AI. But it shows why the quality of the learning process can matter more than the speed at which a result is produced.

Results we no longer understand

One of the most troubling developments is the widening gap between what people produce and what they actually understand. With AI, impressive projects can now be built with little prior knowledge — websites, databases, APIs, entire applications.

A typical example: someone has an AI build the frontend, database, API, Docker containers and server deployment, and accepts every suggested fix without checking how things fit together. After a few weeks, the website is running and looks professional. Then, after an update, a container won't start anymore, a database migration fails or a firewall rule cuts off two services from each other. The person can ask the AI again — but doesn't know which components are affected, where the logs are or which change might break other features.

How wide this gap can be is shown by a study of 32 computer science beginners (Zi et al., 2025): when it came to understanding AI-generated code, their per-task success rate was just 32.5%. Generating code and understanding code are clearly two different skills.

When the tool is suddenly gone

AI is a tool that can be taken away from you at any time: the service is down, your quota is used up, the model has been changed, prices go up or the provider shuts the service down. Then the crucial question arises: can you still keep your project going, investigate errors systematically and add a new feature without having everything regenerated?

If the answer is mostly no, there is a dependence that goes far beyond using a tool. A working product is not yet proof of your own competence. Competence shows in the ability to diagnose errors, justify decisions and keep developing a system without constant assistance — especially for applications that involve money, personal data or IT security.

Nobody has to master every task without tools. But anyone who uses a calculator should be able to tell whether a result is plausible; anyone who deploys AI-generated code should understand how it essentially works. It gets risky when we lose the ability to check our tools.

The illusion of competence

AI delivers explanations in clear, convincing language. That is exactly what can lead us to overestimate their accuracy: a well-phrased answer can be wrong, an extensive block of code can contain security holes, a detailed analysis can overlook important data. While reading, the steps seem logical — as soon as we have to apply them without a template, we realize that we only recognized them, but never understood them.

A Microsoft Research survey of 319 knowledge workers (2025) found a link between higher trust in generative AI and lower self-reported critical-thinking effort. At the same time, it shows that critical thinking is changing: instead of solving tasks themselves, people increasingly have to check AI output, put it in context and take responsibility for it. The most important skill of the future is therefore not just being able to operate AI — but being able to critically evaluate its results.

Are we becoming "prompt junkies"?

The term isn't a diagnosis but a behavior pattern: for almost every question, a prompt gets typed reflexively before any thinking happens. The answer is accepted, and at the next difficulty the cycle starts all over again — until any mental effort feels like an unnecessary detour. These questions can help you examine your own behavior:

• Do I start an AI query before I've thought about the problem myself?
• Can I explain the answers I accept in my own words?
• Could I repeat the task afterward without AI?
• Do I check important claims against reliable sources?
• In an AI-generated project, do I know which component is responsible for what?
• Do I feel unsure about simple tasks when no AI is available?

A single yes is no proof of dependence. What matters is the overall picture: does AI expand your abilities — or is it becoming a prerequisite for being able to act at all?

AI meets an attention span that is already overloaded

The debate about AI doesn't take place in a vacuum. Even before it, our attention was already being claimed by smartphones, social networks, short videos, advertising and notifications — by services designed, with autoplay, endless scrolling and personalized feeds, to hold on to us for as long as possible.

A meta-analysis published in Psychological Bulletin in 2025 evaluated 71 studies with a total of 98,299 participants: more frequent use of short videos was associated with poorer cognitive outcomes, particularly in attention and impulse control. Most of the data shows correlations, not clear causation — still, it is a clear warning sign.

The combination is where it gets problematic: short videos train rapid switching between stimuli, notifications interrupt periods of focus, and as soon as a task gets difficult, AI delivers an instant answer. Thinking independently seems slow and strenuous by comparison. That leaves fewer and fewer opportunities to practice exactly the stamina needed for demanding work and for learning new things.

AI as a tutor instead of an answer machine

The same AI that can replace thinking can also train it. The difference lies in the kind of question you ask. Instead of asking for the finished solution, you can deliberately use it as a personal tutor:

• "Ask me questions that help me figure out what I don't understand yet."
• "Give me a hint, but not the full solution."
• "Let me solve the problem myself first, then evaluate my approach."
• "Create three practice exercises of increasing difficulty."
• "Check my explanation and show me which connections are missing."

That way, you stay active — and AI becomes a training partner instead of a crutch.

Using AI wisely in your personal and professional life

AI is a powerful tool that makes knowledge more accessible and saves unnecessary effort. It is especially helpful for research and getting oriented, routine work, learning support, programming, brainstorming, translations, accessibility and organization. To make sure the benefits don't come at the expense of your own competence, a few clear rules help:

• Define the goal yourself. If you don't know which problem needs solving and what a good result looks like, you can't judge whether the AI has achieved it.
• Understand the solution before you sign off on it. Have the key steps explained to you, check assumptions and test the result — with software, especially security-relevant changes.
• Automate repetition, not your expertise. AI can take over formatting and standard reports; for new, complex or risky tasks, human oversight stays intensive.
• Document decisions. Record why a solution was chosen — that way, the knowledge is preserved even when an AI service isn't available.
• Keep practicing without AI. Deliberately schedule tasks that you solve entirely without assistance.
• Protect your data. Private documents, login credentials, customer data or confidential business information don't belong in an AI chat without careful consideration.
• Keep the responsibility. In matters of finance, law, health or security, AI provides arguments — the decision and its consequences remain yours.

The method: think first, then AI, then check

A simple routine combines the advantages of AI with active thinking of your own — for programming, languages, research, project planning and much more:

• Step 1 — Try it yourself. Describe the problem in your own words and write down what you know, what you suspect and which questions remain open.
• Step 2 — Ask targeted questions. Ask the AI for hints, explanations or counterarguments, not automatically for the complete solution.
• Step 3 — Compare critically. Look for false assumptions, missing information and alternatives.
• Step 4 — Implement it yourself. Implement the solution yourself or solve a similar task without help.
• Step 5 — Check the result. Verify that it works and that you could repeat the process later.

How to protect your focus and your own thinking

If you want to guard against digital dumbing-down, you shouldn't look only at your AI use, but at your entire digital environment:

• Cut down on notifications. Turn off everything unnecessary and set fixed time slots for messages and social media.
• Consciously limit short videos. If you often scroll longer than planned, a fixed usage limit helps.
• Create periods of focus. Work on one task for a manageable stretch of time without jumping between browser, messenger and AI chat — and gradually increase the duration.
• Read instead of just consuming summaries. For important topics, longer texts, documentation and original sources are worth the time.
• Allow yourself to be bored. Not every wait has to be filled with your smartphone — undirected thinking is often where your own ideas begin.
• Test your dependence. Regularly solve a familiar task completely without AI and check whether you still know the essential steps.

PRO6SELLER: awareness for a self-determined digital life

Digital developments are changing how we take in information, make decisions, work and handle our privacy. That is why PRO6SELLER sees itself not only as a provider of digital tools, but as a place where people learn to use these tools consciously, critically and on their own terms.

The magazine translates complex topics — from AI and automation to digital traces and surveillance to encryption and data minimization — into accessible articles with practical recommendations. Every service is paired with clear information: what does a tool do, which data is processed and where are its limits? The goal is not to tie people as tightly as possible to digital services, but to give them the knowledge they need to make independent decisions.

This principle also applies to our own services: with the Security Setup, all security-relevant settings are documented in a traceable way, and Bot & Script development can be used not only to build an automation but also to create clear technical documentation for an existing project — especially valuable when nobody fully understands an AI-built system anymore. For projects beyond the standard range, you can request an individual offer via Mission.

The best technology isn't necessarily the one that makes every decision for us. Often, it's the one that empowers us to act in a more informed and independent way.

Conclusion: let AI work for you — but don't unlearn how to think for yourself

AI can be an excellent teacher, assistant, source of ideas and technical tool. But it can't automatically give us the understanding that comes from learning on our own, hands-on experience and critical engagement. If you merely accept answers instead of understanding them, you feed the illusion of competence. If you have entire projects built without knowing how they work, you become dependent on the very tools that were supposed to help.

That's why we shouldn't only ask what AI can do for us — but also what we ourselves still want to be able to do. Automate the routine. Use AI as support. Check its results. Learn the fundamentals. And keep the ability to carry on working on your own.

The future doesn't automatically belong to those who write the most prompts, but to those who understand the results, ask the right questions and can take responsibility. Stay curious. Stay critical. And don't hand over your own thinking.

🔗 Related reading: Using technology mindfully starts with your own data — which traces we leave while browsing and how to reduce them is explained in the article Hide Your IP Address: VPN, Proxy and Tor Explained. How to regain control over your own device instead of relying on factory defaults is shown in our article Root Smartphones: More Control, More Freedom, More Privacy.