MIT Media Lab: “Your Brain on ChatGPT”

People who wrote essays with ChatGPT showed the weakest brain connectivity of three groups, and felt the least ownership of their work.

Authors
Kosmyna et al., MIT Media Lab
Published
2025, arXiv preprint
Peer review
Not yet peer-reviewed
Participants
54
Method
Essay writing with ChatGPT, with a search engine, or with no tools; brain activity recorded with EEG
Length
Four months

What they did

Researchers at the MIT Media Lab asked 54 people to write essays over four months. One group used ChatGPT, one used a search engine, and one used no tools at all. While they wrote, the researchers recorded their brain activity with EEG, then looked at the essays and asked people about them.

What they found

The ChatGPT group showed the weakest brain connectivity of the three groups while writing, and felt the least ownership of what they wrote. The no-tools group showed the strongest connectivity and the most ownership.

“Over four months, LLM users consistently underperformed at neural, linguistic, and behavioral levels.”

Kosmyna et al., 2025, abstract

“Self-reported ownership of essays was the lowest in the LLM group and the highest in the Brain-only group.”

Kosmyna et al., 2025, abstract

What it doesn't show

  • It is a preprint. It has not been through peer review yet.
  • The sample is small: 54 people, all doing one kind of task, essay writing.
  • EEG connectivity measures how engaged people's brains were during the task. It is not a measure of intelligence.

A request from the authors

The authors' own FAQ asks people not to describe this paper with words like "stupid", "dumb", "brain rot", "harm", "damage", "brain damage", "passivity", "trimming" or "collapse". We follow that. Our homepage headline makes a broader marketing claim, and this page sticks to what the paper found. Read the authors' project page and FAQ.

Read the paper

Read the paper on arXiv“Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task.”

How MindMargin relates

The ChatGPT group did less of the work themselves, and felt less connected to the result. MindMargin does not change how you use AI for the task in front of you. It adds a few seconds of your own effort (recalling a word, answering a flashcard) during the wait, so some practice stays yours.

All four studies, and what they add up to