This week a headline made the rounds: students are losing the ability to read. The usual reflexes followed at once – «of course, AI is destroying education» versus «overblown moral panic». Both skip the only thing that matters: what does the research say once you actually read the studies? And above all – this is not a youth problem. The people studied were adults, knowledge workers, professionals. It concerns us all.
What the Studies Show
The brain measurably powers down – and does not recover at once. At the MIT Media Lab, 54 people (aged 18–39) wrote essays while an EEG recorded their brain activity across 32 regions. The ChatGPT group showed the weakest neural connectivity, the «brain-only» group the strongest and most widely distributed. 83 % of the AI users were afterwards unable to quote a single line from their own text. In a fourth session without AI, their brain activity did not return to the normal level – the authors call this «cognitive debt» (Kosmyna et al. 2025). Important: a preprint, small sample, not yet peer-reviewed, methodologically criticised (Stanković et al. 2026). The direction, though, matches everything else.
Those who trust the AI think less critically. A peer-reviewed study by Microsoft Research and Carnegie Mellon University surveyed 319 knowledge workers and analysed 936 real AI tasks. The pattern is unambiguous: the greater the trust in the AI, the less critical thinking is applied. Those who trusted their own competence checked the results carefully (Lee et al. 2025).
Frequent AI use correlates with weaker critical thinking – most strongly among the young. Michael Gerlich of the SBS Swiss Business School found, among 666 people, a marked negative relationship between frequent AI use and critical thinking ability, mediated by «cognitive offloading» – the outsourcing of thinking to digital tools. Younger participants were the most dependent and performed worst (Gerlich 2025). A Swiss study, incidentally.
With children the risk is fundamentally different in kind. Adults lose an ability they already possess. A child whose brain has not yet formed independent pathways of thought may never build them at all. The AI then does not compete with existing capacity – it replaces its development.
Those affected see it themselves. A RAND survey (December 2025) of more than 1,200 school and university students shows: 70 % of university students fear AI is harming their critical thinking. Among high-school students this share rose within a single year from 55 % to 65 % – while usage increased at the same time (Schwartz & Diliberti 2026). When those affected sound the alarm themselves, that is a data point.
The Decisive Counterpoint
The tool is not the problem – the process is. Studies consistently show: those who engage critically with AI – questioning, editing suggestions, pushing back instead of blindly copying – learn better and tire less. AI as a sparring partner rather than a ghostwriter demonstrably improves argumentation and critical thinking (de la Puente et al. 2024; Wang & Fan 2025). Whether AI builds thinking up or replaces it is decided by design, use and intent.
Conclusion
The answer is neither a ban nor unrestricted access. Today we use a single all-purpose tool for everything – for postdoctoral research as much as for a twelve-year-old's first algebra problem. That is the design flaw. Instead of retrofitting all-purpose chatbots with filters, research and investment energy should flow into specialised, purpose-built applications: a learning AI following the Socratic method, one that does not hand over the answer but forces the learner to think. First attempts exist – they are simply far too rare.
With social media, it took a decade of documented harm before regulation followed (the United Kingdom is now banning it for under-16s). With AI we are earlier on the same curve – and this time we see the pattern coming. Social media hijacked our attention. AI hijacks thinking itself. The right demand is therefore not «ban it», but: urgently better research, longitudinal, across age groups – and tools that make us think harder, not less.
Sources
- Kosmyna, N., Hauptmann, E., Yuan, Y. T., et al. (2025). Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task. arXiv:2506.08872 (preprint, n=54).
- Stanković, M., et al. (2026). Comment on: Your Brain on ChatGPT — methodological critique of Kosmyna et al. arXiv:2601.00856 (comment, preprint).
- Lee, H.-P., et al. (2025). The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers. Proceedings of CHI 2025. Microsoft Research & Carnegie Mellon University (n=319, 936 tasks).
- Gerlich, M. (2025). AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking. Societies, 15(1), 6. SBS Swiss Business School (n=666).
- Schwartz, H. L., & Diliberti, M. K. (2026). More Students Use AI for Homework, and More Believe It Harms Critical Thinking. RAND Corporation, RR-A4742-1 (n≈1,214).
- de la Puente, M., et al. (2024); Wang, J., & Fan, W. (2025). Meta-analysis on the effect of ChatGPT on learning performance and higher-order thinking. Humanities and Social Sciences Communications, 12.