- MIT researchers found LLM users showed the weakest brain connectivity after writing with ChatGPT across three sessions.
- The study tracked 54 participants via EEG—Brain-only writers built the most distributed neural networks; AI-assisted writers’ cognition scaled down with each session.
- LLM users couldn’t accurately quote their own essays, and former AI users who switched back to brain-only writing still showed reduced engagement weeks later.
Ten minutes. That’s roughly how long it takes for measurable cognitive atrophy to set in when you outsource your thinking to ChatGPT—at least according to an MIT Media Lab study that now has the brain scans to back it up.
Researchers at MIT, led by Nataliya Kosmyna, fitted 54 participants with EEG caps and had them write essays across four sessions. One group used ChatGPT, another used search engines, and a third relied on nothing but their own brains. The results, published in arXiv under the title “Your Brain on ChatGPT,” showed that LLM-assisted writers developed the weakest, most fragmented brain connectivity of all three groups.
The gap wasn’t subtle. Brain-only participants built the strongest, most distributed neural networks—engaging memory, executive function, and creative processing simultaneously. Search engine users landed in the middle. LLM users? Their brains barely lit up. Cognitive activity, the researchers found, “scaled down in relation to external tool use.” Translation: the more the AI did, the less the brain bothered.
Why Your Brain Stops Trying After ChatGPT Takes Over
It gets worse. In a fourth session, researchers swapped groups—former ChatGPT users were told to write on their own, and brain-only writers got access to the AI. The results were lopsided. Former LLM users who went back to writing by hand showed “reduced alpha and beta connectivity,” a clinical way of saying their brains had gotten lazy and couldn’t snap back immediately. Brain-only users who switched to ChatGPT, meanwhile, showed higher memory recall and activation in the prefrontal cortex—regions tied to decision-making and working memory.
LLM users also reported the lowest sense of ownership over their own essays. When asked to quote their own work, many couldn’t. They had produced text, sure—but their brains hadn’t processed it as something they’d created. It was output, not thought. The Harvard Gazette spoke with multiple faculty experts who flagged a similar pattern: students using AI “without a good understanding of how it works in a computational/Bayesian sense” tend to put “too much confidence in its output.”
The study has its skeptics—and rightfully so. A comment paper published on arXiv criticized the sample size (54 participants, only 18 in session 4), flagged reproducibility concerns with the EEG analysis, and noted “inconsistencies in the reporting of results.” The authors of the comment were careful to praise the research direction but argued some findings “could be interpreted more conservatively.”
That caution hasn’t stopped the alarm bells from ringing. A Stanford study from March found similar risks in AI chatbot interactions, noting that users develop dangerous dependency patterns when chatbots reinforce their existing beliefs. The broader pattern—what outlets have called the “boiling frog” effect on cognition—suggests this isn’t a ChatGPT problem. It’s a human-AI interaction problem that scales with every new assistant, copilot, and chatbot that promises to think for you. Frontierbeat has previously reported on how agentic AI systems amplify these dependency risks when they move from passive tools to active decision-makers.
The practical concern isn’t that people will stop thinking entirely. It’s that casual AI use—asking it to draft emails, summarize articles, brainstorm ideas—creates what the researchers call “cognitive debt.” Small, compounding. Invisible until you need your brain to do the heavy lifting and realize the muscle has atrophied. As one Harvard professor put it: “If AI is doing your thinking for you, that is undercutting your critical thinking and your creativity.”
The study, “Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task,” is available on arXiv.
