Scientists Scanned ChatGPT Users' Brains In A Groundbreaking Study. The Results Are Horrifying

When it comes to new technologies, none of them have been adopted by the general public as fast as large language models (LLMs) like ChatGPT, Claude, Copilot, and Gemini. How quick the technology has expanded and been accepted has garnered many concerns. In 2015, Stephen Hawking had a grim warning about artificial intelligence (AI) potentially accelerating economic inequality and increasing the poverty gap. Meanwhile, some researchers worry that an overreliance on this tech is one of the worst ways AI has affected school, and a groundbreaking study observing participants' brain activity while using ChatGPT may have proved it.

Publishing their findings in arXiv, scientists at Massachusetts Institute of Technology (MIT), Massachusetts College of Art and Design (MassArt), and Wellesley College collaborated to determine the cognitive effects of using LLMs while writing essays. They divided 54 participants aged 18 to 39 into three groups based on what they were restricted to using to write the essays — OpenAI's ChatGPT, Google or other search engines without AI-enhanced queries, and nothing except their personal knowledge. All of them completed three essay writing sessions. For a fourth session, the researchers switched the groups of 18 participants; the LLM-only group was restricted to using their brains, while the brain-only group was restricted to ChatGPT.

The participants' brain activity was recorded with electroencephalography (EEG), which showed that the ChatGPT-only group had the least brain engagement. Compared to the other groups, the participants also mostly relied on copy-paste by session three and found it more difficult to recall passages from their own essays. When the groups were switched, the ChatGPT-to-brain group had the worst performance and just slightly better engagement compared to the first baseline session of the brain-only group. These findings are preliminary but suggest that ongoing AI use may "contribute to cognitive atrophy through excessive reliance."

Alongside this MIT Media Lab study, others have made disturbing discoveries about LLMs. One study published in NPJ Digital Medicine found that AI is capable of making assumptions and overlooking critical details, which isn't ideal in healthcare settings. While a study published in Computers in Human Behavior showed that people who regularly use ChatGPT tend to overestimate their cognitive abilities, another study published in the Journal of Creative Behaviour determined that the creative capacity of AI is limited to that of an average person and can't meet expert or professional standards. In the debate regarding whether artificial intelligence is good or bad, though, scientists have also seen some positive aspects.

For example, research published in Physical Review Physics Education Research showed that, between 2022 and 2025, students from over 20 countries increasingly migrated from using Google's search engine to using ChatGPT as a tool to find information. That shift has allowed non-English-speaking students to obtain quality details in their native languages, potentially improving access to knowledge. In Computers & Education, researchers published findings that note-taking is still critical for student comprehension and retention but that LLMs can be useful tools for clarifying, exploring, and putting learning materials into more understandable context. The scientists also determined that note-taking alone or in combination with LLMs improved understanding and retention better than LLM-use alone.

A third example of positive AI-study results was published in Computers and Education Artificial Intelligence. For a 10-year period that started in 2016, researchers at University of Warwick analyzed nearly 5,000 student reports. They found that, since generative AI emerged in 2022, the students have used more formal, positive, and sophisticated wording in their papers. Despite that, the evidence suggests that LLMs haven't overshadowed core academic skills.