A collaborative study conducted by the Massachusetts Institute of Technology (MIT), Massachusetts College of Art and Design (MassArt), and Wellesley College has revealed that relying exclusively on large language models (LLMs) like ChatGPT during writing tasks leads to decreased brain engagement. The research indicates that an overreliance on these tools may contribute to cognitive atrophy, as users exhibit a reduced ability to recall information and a tendency to depend on automated outputs rather than active mental processing.
Brain Activity and AI Reliance
To determine the cognitive effects of LLMs, researchers monitored 54 participants between the ages of 18 and 39. The subjects were divided into three distinct groups to complete essay writing sessions under different constraints:
- LLM-only group: Restricted to using OpenAI’s ChatGPT.
- Search-only group: Restricted to using Google or other search engines without AI-enhanced queries.
- Brain-only group: Restricted to using personal knowledge.
Using electroencephalography (EEG) to record brain activity, the scientists found that the group using ChatGPT showed the lowest levels of brain engagement. By the third session, these participants relied heavily on copy-pasting content and struggled significantly to recall passages from their own essays. Findings published in arXiv show that when the groups were switched—forcing the LLM-users to rely solely on their own cognitive abilities—they demonstrated the worst performance and only marginal engagement compared to the initial baseline of the brain-only group.
Cognitive Risks and Professional Limitations
Beyond brain activity, other research highlights specific risks associated with the integration of AI into professional and personal environments. In healthcare, a study published in NPJ Digital Medicine found that AI can make unfounded assumptions and overlook critical clinical details, posing risks in medical settings.
The psychological impact of AI use also extends to self-perception and creativity. Research in Computers in Human Behavior suggests that frequent ChatGPT users tend to overestimate their own cognitive abilities. Furthermore, a study in the Journal of Creative Behaviour determined that AI creativity is limited to an average human level and fails to meet the standards required for professional or expert-level work.
Educational Integration and Knowledge Access
Despite these risks, some data suggests that LLMs can provide significant advantages when used as complementary tools. Between 2022 and 2025, students from more than 20 countries shifted from traditional search engines to ChatGPT to find information. According to research in Physical Review Physics Education Research, this shift has improved access to knowledge for non-English speakers by providing quality information in their native languages.
The effectiveness of AI in learning often depends on how it is paired with traditional methods. A study in Computers & Education found that while LLMs are useful for clarifying and contextualizing learning materials, they are less effective than traditional note-taking for retention. The highest levels of comprehension were achieved when students used note-taking either alone or in combination with AI tools.
Academic quality may also be evolving. Researchers at the University of Warwick analyzed nearly 5,000 student reports over a decade starting in 2016. Their findings, published in Computers and Education Artificial Intelligence, indicate that since 2022, students have utilized more sophisticated and formal wording in their papers. The evidence suggests that while the style of writing has shifted, core academic skills have not been overshadowed by the emergence of generative AI.
