Learning brief
Generated by AI from multiple sources. Always verify critical information.
TL;DR
MIT researchers found that using ChatGPT to write essays literally decreased brain activity in regions responsible for thinking. Students who used the AI showed lower neural engagement, couldn't remember what they wrote, and by their third essay were just copy-pasting. If Google search changed how we remember, AI chatbots might be changing how we think.
What changed
ChatGPT users showed measurably lower brain activity than students writing from scratch or using Google search.
Why it matters
Your brain stops working when AI does the thinking — and the effect compounds each time you use it.
What to watch
Will schools adopt AI tools before we understand the long-term impact on developing brains?
What Happened
MIT researchers strapped EEG sensors (think: swim cap covered in electrodes that measure brain electricity) to 54 college students and asked them to write SAT essays. One group used ChatGPT. One used Google search. One used nothing but their brains (Source 2, 3).
The ChatGPT group's brains were measurably quieter. The EEG showed less activity in neural networks responsible for critical thinking, memory processing, and creativity — the same networks that light up when you're actually solving a problem (Source 3). They also couldn't quote from their own essays afterward, suggesting they never deeply processed what they'd written (Source 2).
By the third essay, most ChatGPT users had stopped trying. They'd evolved from "help me outline this" to "write this for me and fix the grammar" — essentially copy-paste (Source 3). Two English teachers who graded the essays called them "soulless" and noted they all used identical phrasing and ideas (Source 3). The brain-only group, meanwhile, showed the highest neural connectivity in areas tied to creativity and semantic processing.
A separate study from Carnegie Mellon and Microsoft surveyed 319 office workers who use AI tools weekly. The finding: the more confident someone was that AI could handle a task, the less critical thinking they applied to checking its output (Source 2). Think of it like spell-check — you probably don't scrutinize every red underline anymore because you trust it. Now apply that to data analysis or legal research.
MIT researcher Nataliya Kosmyna released the findings before peer review specifically because she's worried a school district will launch "GPT kindergarten" before we understand what we're risking (Source 3). The concern isn't theoretical — a UK survey found 60% of students felt AI had already damaged their schoolwork skills (Source 2).
So What?
The real story here is cognitive atrophy, not productivity. Yes, ChatGPT users finished essays faster. But their brains worked less each time they used it — and the laziness compounded (Source 3). This isn't like using a calculator for arithmetic while learning advanced math. It's like having someone else do your physical therapy exercises. Your brain is a muscle, and these tools are doing the reps for you.
The uncomfortable truth is we already have a preview of this effect. Remember when GPS apps became ubiquitous? Studies showed people's spatial memory declined because they stopped mentally mapping routes (Source 4). Google search created the "Google Effect" — we stopped remembering facts we knew we could look up (Source 4). Both tools made life easier. Both also rewired how our brains work. AI chatbots are the same pattern, but for higher-order thinking: analysis, synthesis, argumentation. The things that separate critical thinking from information retrieval.
For developing brains, the stakes are higher. Adults have already built neural pathways for problem-solving. Kids haven't. If a 10-year-old outsources essay writing to ChatGPT, they're not just skipping homework — they're skipping the cognitive development that homework was designed to trigger (Source 3). And unlike calculators (which handle tedious arithmetic so students can focus on concepts), AI chatbots handle the concepts. There's nothing left to learn.
Sources