Search author on:PubMedGoogle Scholar
Electroencephalography can be used to study brain activity while people are completing tasks with and without the help of chatbots such as ChatGPT.Credit: Leah Nash/Washington Post/Getty
The brains of people writing an essay with ChatGPT are less engaged than are those of people blocked from using any online tools for the task, a study finds1. The investigation is part ofa broader movement to assess whether artificial intelligence (AI) is making us cognitively lazy.
How close is AI to human-level intelligence?
How close is AI to human-level intelligence?
Computer scientist Nataliya Kosmyna at the MIT Media Lab in Cambridge, Massachusetts, and her colleagues measured brain-wave activity in university students as they wrote essays either using a chatbot or an Internet search tool, or without any Internet at all. Although the main result is unsurprising, some of the study’s findings are more intriguing: for instance, the team saw hints that relying on a chatbot for initial tasks might lead to relatively low levels of brain engagement even when the tool is later taken away.
Echoingsome posts about the study on online platforms, Kosmyna is careful to say that the results shouldn’t be overinterpreted. This study cannot and did not show “dumbness in the brain, no stupidity, no brain on vacation”, Kosmyna laughs. It involved only a few dozen participants over a short time and cannot address whether habitual chatbot use reshapes our thinking in the long-term, or how the brain might respond during other AI-assisted tasks. “We don’t have any of these answers in this paper,” Kosmyna says. The work was posted ahead of peer review on the preprint server arXiv on 10 June1.
Kosmyna’s team recruited 60 students, aged 18 to 39, from five universities around the city of Boston, Massachusetts. The researchers asked them to spend 20 minutes crafting a short essay answering questions, such as “should we always think before we speak?”, that appear on Scholastic Assessment Tests, or SATs.
Are the Internet and AI affecting our memory? What the science says
Are the Internet and AI affecting our memory? What the science says
The participants were divided into three groups: one used ChatGPT, powered by OpenAI’s large language model GPT-4o, as the sole source of information for their essays; another used Google to search for material (without any AI-assisted answers); and the third was forbidden to go online at all. In the end, 54 participants wrote essays answering three questions while in their assigned group, and then 18 were re-assigned to a new group to write a fourth essay, on one of the topics that they had tackled previously.
Each student wore a commercial electrode-covered cap, which collected electroencephalography (EEG) readings as they wrote. These headsets measure tiny voltage changes from brain activity and can show which broad regions of the brain are ‘talking’ to each other.
The students who wrote essays using only their brains showed the strongest, widest-ranging connectivity among brain regions, and had more activity going from the back of their brains to the front, decision-making area. They were also, unsurprisingly, better able to quote from their own essays when questioned by the researchers afterwards.
The Google group, by comparison, had stronger activation in areas known to be involved with visual processing and memory. And the chatbot group displayed the least brain connectivity during the task.
More brain connectivity isn’t necessarily good or bad, Kosmyna says. In general, more brain activity might be a sign that someone is engaging more deeply with a task, or it might be a sign of inefficiency in thinking, or an indication that the person is overwhelmed by ‘cognitive overload’.
Gain free access to this article, as well as selected content from this journal and more onnature.com
doi: https://doi.org/10.1038/d41586-025-02005-y
Kosmyna, N.et al.Preprint at arXivhttps://doi.org/10.48550/arXiv.2506.08872(2025).
Makransky, G., Shiwalia, B. M., Herlau, T. & Blurton, S.Educ. Psychol. Rev.37, 60 (2025).
Are the Internet and AI affecting our memory? What the science says
How close is AI to human-level intelligence?
Do AI models produce more original ideas than researchers?
How does ChatGPT ‘think’? Psychology and neuroscience crack open AI large language models
Eureka! The brain science behind lightbulb moments
Obesity drugs show promise for treating a new ailment: migraine
First ever skull from ‘Denisovan’ reveals what ancient people looked like
‘I didn’t know someone could pursue astronomy as a career’
Boys surpass girls in maths in the first year of school
How many PhDs does the world need? Doctoral graduates vastly outnumber jobs in academia
Don’t sleepwalk from computer-vision research into surveillance
Computer-vision research is hiding its role in creating ‘Big Brother’ technologies
Controlling diverse robots by inferring Jacobian fields with deep networks
Founded in 2018, Westlake University is a new type of non-profit research-oriented university in Hangzhou, China, supported by public a…
The University of Toronto now recruiting for the Eric and Wendy Schmidt AI in Science Postdoctoral Fellowship. Valued at $85,000 CDN per year.
The Integrated Science Lab (IceLab) offers three postdoctoral scholarships that will be affiliated with one of six possible multidisciplinary projects
Aim to a global leader in materials science research society, and create a diverse, innovative, and collaborative research environment
Department of Frontier Materials, Suzhou Laboratory
The Max Planck Institute of Molecular Cell Biology and Genetics (MPI-CBG) in Dresden, Germany, is seeking an outstanding individual to fill the po…
Max Planck Institute of Molecular Cell Biology and Genetics
