Are you surrendering your brain to an AI chatbot? A study from the MIT Media Lab offers an early picture. Researchers had people write short essays while measuring their brain activity and testing their memory. Writers who used a chatbot showed a drop of up to 55% in brain network activity, the flow of information between brain regions, compared with writers who worked alone.

The study, titled Your Brain on ChatGPT, followed 54 participants aged 18 to 39 from five Boston-area universities, most of them students. The team assigned each person at random to one of three setups, with 18 people per setup, balanced for age and gender. Every essay had a 20-minute limit, and the prompts came from SAT tests.

The three setups gave writers different amounts of outside help. Chatbot writers received a ChatGPT account and were barred from other websites and other chatbots. The Google group searched the web with AI answers filtered out, which gave the team a second comparison: outside help without a chatbot. Writers in the third group had no chatbot, no websites, and no tools, only their own minds. People used to write essays this way. Each person kept the same setup for three sessions, and the study ran for four months.

Every writer wore a headset with sensors on the scalp, and the headset recorded brain activity in each session. The authors used the readings to assess how engaged each writer was and how much mental effort the task took. Writers working alone had the strongest and widest-ranging brain network activity and served as the baseline. The Google group showed 34% to 48% less. The chatbot group sat lowest, at up to 55% below the baseline. The authors describe the pattern as activity scaling down with the amount of external support: brain regions passed less information to each other as outside help grew.

The chatbot group also fell behind on recall. After each session, researchers asked each writer to quote a sentence from the essay without looking at the text. The question tested memory, and the essays were minutes old. In the first session, the question was new to every writer. Of the 18 chatbot users, 15 struggled to quote anything from their essays, and none gave a correct quote. Correct quotes came from 15 of 18 writers in the Google group and 16 of 18 writers working alone. The difference shrank once writers knew the question was coming. In the third session, 6 of 18 chatbot users still failed to give a correct quote, against 2 of 18 in each of the other two groups.

A fourth session took the chatbot away. Participants were not told about the change in advance, and each person chose a topic from the ones they had already written about. Only 18 of the 54 came back for the session, and nine of those had written with the chatbot for three sessions. These nine wrote with no tools at all. Their brain network activity was weaker, and the authors describe the result as under-engagement, a lower level of involvement in the task.

The first two results describe writers with a chatbot at hand. The third describes writers after the chatbot was gone. All three come from one task, essay writing. So what happens if you use ChatGPT, Gemini, or Claude for a long list of tasks? The study does not say. AI companies are not going to tell you to stop using their products, so the decision falls to you.

The evidence behind this decision remains early. The researchers posted the paper on a public research server before outside experts reviewed the work. Lead researcher Nataliya Kosmyna said the conclusions should be treated with caution and as preliminary. She called the work one of the first studies of AI’s effect on the human brain. The authors say chatbots offer immediate convenience and point to potential cognitive costs. Chatbots now do many tasks. Should they do all of them? Where do you stop?

Sources

arXiv, Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task — arxiv.org
The Register, Brain activity much lower when using AI chatbots, MIT boffins find — theregister.com