MindMarginIs AI making us dumber? What the research actually says
The short answer: nobody has shown that AI lowers anyone's intelligence. But several careful studies find that when people hand their thinking to AI, they do less of it, and in one real classroom experiment, they learned less.
The four studies
These are the studies behind the research cards on our homepage. Each page gives the numbers, a short quote from the paper, what it can't tell us, and a link to the paper itself.
- MIT Media Lab: Your Brain on ChatGPTPeople who wrote essays with ChatGPT showed the weakest brain connectivity of three groups, and felt the least ownership of their work. Preprint, not yet peer-reviewed, with a small sample.→
- Microsoft Research + Carnegie Mellon: Generative AI and critical thinkingThe more people trusted the AI, the less critical thinking they reported doing. Self-reported survey. It shows a link, not a cause.→
- Wharton + Penn Engineering: Generative AI without guardrails can harm learningHigh school students who practiced math with an unguided chatbot did worse on the exam once it was taken away. A tutor version with guardrails avoided the drop.→
- SBS Swiss Business School: AI tools and cognitive offloadingHeavier AI use went hand in hand with weaker critical thinking, linked to handing more of the thinking to the tool. Correlational. It shows a link, not proof of cause.→
What they add up to
The four studies look at different people doing different things: students writing essays, office workers using AI at their jobs, high schoolers learning math, and adults describing their own habits. They point the same way. When AI does more of the work, people tend to do less of the thinking themselves, and at least in math class, that showed up as lower scores once the AI was gone.
They also have real limits. One is a preprint that has not been peer-reviewed. Two rely on what people say about themselves. Two are correlational, so they show a link and cannot prove cause. Only one, the Wharton and Penn field experiment, randomly assigned students and measured learning directly.
That experiment also carries the most useful finding. A version of the chatbot with guardrails, built to guide students instead of handing over answers, avoided the drop. The problem it found was AI doing the thinking for students, and the design of the tool made the difference.
What is cognitive offloading?
Cognitive offloading means using something outside your head to do mental work for you: a calendar so you don't have to remember appointments, a calculator for arithmetic, and now AI for writing, summarizing and problem solving. Offloading is useful and everyone does it. The question the research raises is which skills you stop practicing when you offload them, and whether those are skills you want to keep.
A note on the word "dumb"
Our homepage headline says "AI is making you dumb." That's our marketing line, and we chose it on purpose, because it's how a lot of people already talk about this. The studies themselves are more careful. The MIT authors ask people not to describe their findings with words like "dumb", "stupid" or "brain rot", and none of our research cards or study pages do. We'd rather show you exactly what each paper found, limits included, than stretch it.
What you can do about it
None of this says stop using AI. It suggests keeping some of the thinking for yourself: trying a problem before asking, recalling things from memory instead of rereading them, and noticing when you're accepting an answer you haven't checked.
That's what MindMargin is for. It's a free Chrome extension that uses the seconds ChatGPT, Claude and Gemini spend writing to give you a quick piece of practice from memory: your own flashcards, SAT and GRE words, or a short game. You keep using AI the way you do now, and some of the practice stays yours.