Author: Isabell Baumann | Editor: Laure Pauly, Anja Leist
These days, many of us casually hand over little bits of our thinking to AI. We ask it to tidy up our messages, simplify something complicated, or find the right words when we’re stuck. It’s quick, it’s easy, and before we know it, it becomes part of our everyday routine. What does this do to our brains? Do we become smarter or just more comfortable?
The use of software and chatbots that generate text, numbers, or graphs based on large language models (LLMs) or multimodal models is often called AI, and for simplicity, we use the term AI when speaking about how LLMs shape our learning and thinking.
We are starting to outsource parts of our thinking to machines. We ask AI to rephrase messages, explain difficult texts, or suggest words when we cannot find the right ones. This saves time, lowers barriers, and quickly starts to feel normal. But as this becomes routine, it is worth asking what this shift does to how we learn and think. What happens in our brains when we regularly use AI to learn, write, or understand things? Does this support make us smarter or mainly more comfortable?
Research on these questions is still at an early stage. Even so, there are growing signs that AI not only supports our thinking, but also changes it. In a country like Luxembourg, shaped by multilingualism and social inequalities in education, this development is especially relevant and deserves a careful and balanced discussion.
AI as a thinking tool: relief or at a cost?
From a cognitive science perspective, the underlying principle is not new: people have long offloaded thinking tasks to notebooks, calculators, navigation systems, or search engines. Generative AI, however, goes a step further. Systems like GPT not only help us find information, but also produce ready-made answers, texts, and arguments. This can be helpful, for example when learners spend less energy dealing with language barriers and can focus more on understanding content, although research on these effects is still limited (1).
Concerns arise when AI no longer supports thinking but begins to replace it. Studies suggest that more intensive AI use is linked to stronger offloading of cognitive processes. One study, for example, found that frequent use of AI tools was associated with lower levels of critical thinking, a relationship explained by cognitive offloading, the systematic transfer of thinking effort to technology (2).
Learning with AI: support yes, but no replacement for processes of comprehension and practice
Imagine a student preparing for a test with AI support. Answers come quickly and homework looks polished. But when the same student later sits an unexpected test without AI, things may feel harder.
Research suggests that this difference matters. In one study, students who used ChatGPT while studying performed worse in a later surprise test than those who studied without AI (3). Students who learned coding with the use of LLMs performed worse in later code writing than students who weren’t using LLMs during learning. However, use of LLMs for written code (e.g. for debugging = finding mistakes), did not make a difference in the later performance evaluations. When systems seem reliable, people also tend to check results less carefully, a pattern known as automation bias (4).
Learning that lasts usually requires effort and reflection. If AI removes too many of these steps, learning may feel easier at first but become less stable over time.
The developing brain: why children and young people matter
Children and teenagers are growing up with AI as a normal tool. Their brains are still developing, especially the skills that help them concentrate, regulate themselves, and think critically. This does not mean that AI is harmful for young people. But it does mean that how they use it matters.
What seems to help most is, first, information, and, second, dosed use of AI. Young people need to understand how AI works and where its limits lie. They can use it as support when they are stuck or want to explore ideas further. At the same time, they also need regular moments without it; to read, write, calculate, and think things through on their own. These independent phases are important because they build the mental effort and resilience that learning depends on.
Early research suggests that the order matters. Students who first try to work through a task themselves and only then turn to AI show stronger mental engagement than those who rely on AI from the start (5). While this research is still preliminary, it points to a simple principle for education and everyday life: think first, then use AI.
Multilingualism and AI: support and risk
Many students learn, think, and work in several languages at the same time.
AI can provide real support here. It can translate texts, simplify language, or explain content in a more familiar language. For learners whose home language differs from the school language, this can make it easier to access subject content (1). In this sense, AI may help reduce inequalities and make learning more accessible.
At the same time, there are risks. If AI constantly smooths out texts, takes over formulations, or simplifies meanings too quickly, active language production may decline. Language then becomes something to consume rather than to practise. In a multilingual context, however, this active effort is essential: searching for words, comparing meanings, and moving consciously between languages. If texts are only read as summaries, learners may miss the friction with ambiguity and counterarguments that helps develop judgement and understanding.
AI can support learning and living in a multilingual context, but it cannot replace active language production and comprehension. It should act as a bridge, not as a shortcut.
Same AI, unequal opportunities?
AI is often seen as a tool that could reduce educational inequalities. It is available at any time, explains patiently, and does not judge. For students who receive little support at home or feel insecure about language, this can be a real help, especially when questions arise outside school hours.
In Luxembourg, educational success is still closely linked to social background and language (6,7). AI could help reduce some of these gaps, but only if all students learn to use it well. However, current evidence suggests that not everyone benefits equally. Those who already have strong learning strategies and digital skills often use AI more reflectively and therefore more effectively.
What does this mean for schools, parents, and society? Any recommendations?
The key question is not whether we should use AI or ban it, but how we shape its use so that it supports learning rather than shortens it.
Research highlights the pros and cons of AI. It can reduce cognitive effort, but it may also contribute to a gradual loss of reflection and self-regulation if too much thinking is outsourced to technology (8).
For schools and teachers, this means that AI should not be used in isolation or without reflection. It should be integrated into learning processes in ways that encourage explanation, reasoning, and independent thinking (2).
- For parents, the message is simple: accompany rather than control. Children are already using AI at home for homework, writing, and understanding texts. The key question is not whether they use AI, but how. Conversations about when AI is helpful, when it replaces instead of supporting thinking, and why not every answer is correct are more effective than strict bans. Parents do not need to be AI experts, but they can show interest, ask questions, and make use visible.
- For society as a whole, the question is which skills we want to strengthen. If critical thinking, independence, and language skills remain central educational goals, then dealing with AI must become part of these goals, not their replacement.
- For everyone, adopt a critical mindset when interacting with LLMs. Ask the LLM not to validate your ideas, but to critically reflect on them.
All in all …
AI is entering classrooms, homes, and everyday life faster than research can fully keep up and that is important to remember. Although early findings offer valuable insights, we are still only beginning to understand how AI shapes learning, thinking, and the developing brain. What we do know so far points toward a balanced path: AI can offer genuine support, but it works best when it complements and not replaces the mental effort that builds knowledge and thinking skills.
As studies continue to evolve, the most helpful approach for schools, parents, and society is to stay curious, stay reflective, and stay engaged. Rather than asking whether AI is “good” or “bad,” the more meaningful question becomes: How do we use it in ways that strengthen our ability to think, learn, and stay independent? In this sense, the story of AI in education is still being written and we all play a part in shaping it.
________________________________
Take part in AI events organized by the University of Luxembourg: https://www.uni.lu/research-en/research-areas/artificial-intelligence/ai-activities/.
Want to learn how to use AI? We recommend the Elements of AI Luxembourg programme supported by the Luxembourg government as part of the national AI strategy, coordinated by the Department of Media, Connectivity and Digital Policy, implemented with the Digital Learning Hub.
________________________________
For better readability of the text, the assistance of Microsoft Copilot, an AI language model based on the GPT-4 architecture, secured with UL enterprise data protection, has been used.
- Bai, L., Liu, X., & Su, J. (2023). ChatGPT: The cognitive effects on learning and memory. Brain-X, 1(3), Article brx2.30. https://doi.org/10.1002/brx2.30
- Gerlich, M. (2025). AI tools in society: Impacts on cognitive offloading and the future of critical thinking. Societies, 15(1), 6. https://doi.org/10.3390/soc15010006
- Barcaui, A. (2025). ChatGPT as a cognitive crutch: Evidence from a randomized controlled trial on knowledge retention. Social Sciences & Humanities Open, 12, 2025, 102287.
- Parasuraman, R., & Manzey, D. (2010). Complacency and bias in human use of automation. Human Factors, 52(3), 381–410.
- Kosmyna, N., Hauptmann, E., Yuan, Y. T., Situ, J., Liao, X.-H., Beresnitzky, A. V., Braunstein, I., & Maes, P. (2025). Your Brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing task (arXiv:2506.08872) [Preprint]. arXiv. https://arxiv.org/abs/2506.08872
- LUCET, SCRIPT, & MENJE. (2021). Nationaler Bildungsbericht Luxemburg 2021: Bildungsmonitoring für Luxemburg. Luxembourg: University of Luxembourg & Ministère de l’Éducation nationale, de l’Enfance et de la Jeunesse.
- LUCET, SCRIPT, & MENJE. (2024). Nationaler Bildungsbericht Luxemburg 2024: Bildungsmonitoring für Luxemburg. Luxembourg: University of Luxembourg & Ministère de l’Éducation nationale, de l’Enfance et de la Jeunesse.
- Chirayath, G., Premamalini, K., & Joseph, J. (2025). Cognitive offloading or cognitive overload? How AI alters the mental architecture of coping. Frontiers in Psychology, 16, 1699320. https://doi.org/10.3389/fpsyg.2025.1699320
