
Millions of people now ask a chatbot before they ask themselves. They let it draft the email, outline the paper, debug the code, and summarize the book. It feels like a superpower. So it's fair to ask: is AI making us dumber?
The honest answer from the research is more useful than a yes or no. AI doesn't lower anyone's intelligence. But it can stop us from building and using our own thinking skills, especially while we're still learning them. Whether AI makes you sharper or duller depends on when you use it and what you hand over.
Here's what the evidence shows.
What Is Cognitive Offloading?
Psychologists use the term cognitive offloading for any time we use an external tool to reduce mental effort: writing a shopping list, setting a phone reminder, or tilting our head to read rotated text. In a widely cited review, Evan Risko and Sam Gilbert describe offloading as a normal, often smart, part of human cognition.
Offloading isn't new, and it isn't bad. The question is what happens when the tool can do the thinking itself, not just hold the information.
Earlier research gives a hint. In 2011, Betsy Sparrow and colleagues found that when people expected to have later access to information on a computer, they remembered the information itself less well and remembered where to find it better. Our brains adapt to the tools around us. Generative AI is the most powerful such tool ever built.
Four Recent Studies on AI and Thinking
1. Heavy AI users score lower on critical thinking. A 2025 study by Michael Gerlich surveyed and tested 666 participants in the UK. It found a significant negative correlation between frequent AI tool use and critical thinking ability, and cognitive offloading explained much of the link. Participants aged 17 to 25 showed the highest AI dependence and the lowest critical thinking scores. Higher educational attainment was associated with stronger critical thinking regardless of AI use. The study shows a correlation, not proof that AI causes the decline, but the pattern is hard to ignore.
2. Trusting AI more means thinking less. Researchers at Microsoft Research and Carnegie Mellon surveyed 319 knowledge workers about 936 real examples of using generative AI at work. The more confidence workers had in the AI, the less critical thinking they reported applying. The more confidence they had in themselves, the more critical thinking they applied. The researchers also observed a shift in the work itself, from solving problems to verifying and integrating AI output.
3. AI boosts practice scores and lowers test scores. In a randomized field experiment with nearly 1,000 high school math students in Turkey, a team led by Hamsa Bastani at the University of Pennsylvania gave some students a GPT-4 tutor during practice. Students using a standard ChatGPT-style interface solved 48% more practice problems correctly. But when the AI went away for the exam, they scored 17% lower than students who never had access. Many had used the tool to get answers rather than to learn. A second version, designed with teacher input to give hints instead of answers, largely eliminated the harm.
4. Writing with ChatGPT lowers brain engagement. An MIT Media Lab team led by Nataliya Kosmyna recorded EEG activity from 54 participants writing essays with ChatGPT, with a search engine, or with no tools. The ChatGPT group showed the weakest brain connectivity, and many couldn't accurately quote from essays they had just written. In a later session, participants who had first written on their own and then used ChatGPT showed stronger engagement than those who had started with it. The study is small and still a preprint, but it points the same direction as the others.
So, Is AI Making Us Dumber?
Taken together, the research supports a clear and practical conclusion:
- AI is safest in the hands of people who already know how to think. Experts who have built deep knowledge can use AI to move faster, because they can judge its output.
- AI is riskiest for learners. Students who outsource the effortful parts of learning, like recalling, reasoning, and writing, skip the very struggle that builds skill. Learning scientists Robert and Elizabeth Bjork call these productive struggles "desirable difficulties."
- Design matters. The Penn study shows that an AI built to give hints instead of answers removed most of the harm. How we use the tool matters as much as whether we use it.
This also fits a broader trend. After rising for most of the 20th century, IQ scores in Norway began to decline for people born after the mid-1970s, and a large study of Norwegian men by Bernt Bratsberg and Ole Rogeberg found the cause was environmental, not genetic. That study predates generative AI, so it says nothing about chatbots directly. But it shows that the way we live and learn shapes how well we think.
How to Use AI Without Losing Your Edge
The research suggests a few simple habits:
- Think first, then ask. Write your own outline, attempt the problem, or form your own opinion before you open a chatbot. Then use AI to check and extend your thinking.
- Ask for hints, not answers. Tell the AI to question you, point out gaps, or give the next step only.
- Explain it back. If you can't explain an AI-assisted answer in your own words, you haven't learned it yet.
- Keep your confidence calibrated. Treat AI output like a smart but unreliable colleague's draft: useful, and always worth checking.
- Protect real conversations. Discussing ideas with a teacher, mentor, or peer forces you to retrieve, defend, and revise your thinking in real time, something a chatbot can't do for you.
Why We Built a Post-AI University
At Elton, we take this research seriously. No AI teaches at Elton. Every student learns 1-on-1 with expert human professors and coaches, who ask questions, push back, and help students build the knowledge and judgment that let them use AI well. Read more about our approach in AI in Higher Education: The Case for a Post-AI University.
If you want an education that makes you sharper in the age of AI, explore our academics or book a free info session.
Sources
- Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., & Mariman, R. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. Proceedings of the National Academy of Sciences, 122(26).
- Bjork, E. L., & Bjork, R. A. (2011). Making things hard on yourself, but in a good way: Creating desirable difficulties to enhance learning. In Psychology and the Real World.
- Bratsberg, B., & Rogeberg, O. (2018). Flynn effect and its reversal are both environmentally caused. Proceedings of the National Academy of Sciences, 115(26), 6674–6678.
- Gerlich, M. (2025). AI tools in society: Impacts on cognitive offloading and the future of critical thinking. Societies, 15(1), 6.
- Kosmyna, N., et al. (2025). Your brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing task. MIT Media Lab, arXiv 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. CHI 2025.
- Risko, E. F., & Gilbert, S. J. (2016). Cognitive offloading. Trends in Cognitive Sciences, 20(9), 676–688.
- Sparrow, B., Liu, J., & Wegner, D. M. (2011). Google effects on memory: Cognitive consequences of having information at our fingertips. Science, 333(6043), 776–778.