A new study from MIT’s Media Lab is raising red flags about the impact of generative AI tools like ChatGPT on human cognition—particularly among students. The study suggests that using ChatGPT for academic work may reduce brain activity, diminish creativity, and impair memory formation.
The experiment involved 54 participants aged 18 to 39, who were divided into three groups: one using ChatGPT, another using Google Search, and a control group using neither. Each group was asked to write multiple SAT-style essays while wearing EEG devices to measure brain activity across 32 regions.
Results showed ChatGPT users exhibited the lowest neural engagement, underperforming across behavioral, linguistic, and cognitive measures. Their essays were also deemed formulaic and lacking originality by English teachers. Alarmingly, as the study progressed over several months, many in the ChatGPT group abandoned active writing altogether, opting instead to copy-paste AI-generated responses with minimal editing.
Lead author Nataliya Kosmyna explained her urgency to publish the findings ahead of peer review, saying, “I’m afraid in 6-8 months some policymaker will propose ‘GPT for kindergarten.’ That would be absolutely detrimental to developing brains.”
In contrast, the group that relied solely on their own brainpower showed stronger neural connectivity in alpha, theta, and delta bands—regions linked with creativity, memory, and semantic processing. These participants felt more ownership over their work and reported higher satisfaction. The Google Search group also demonstrated high engagement and satisfaction, suggesting traditional web research supports more active learning than LLM use.
In a follow-up test, participants had to rewrite a previous essay—this time without their original tool. ChatGPT users struggled, barely recalling their previous responses, and showed weaker brain wave activity. In contrast, the brain-only group, now using ChatGPT for the first time, exhibited increased cognitive activity, suggesting that AI can support learning—but only when foundational thinking is already in place.
Kosmyna warns that heavy AI use during critical learning phases could impair long-term brain development, particularly in children. Psychiatrist Dr. Zishan Khan echoed this concern: “Overreliance on LLMs may erode essential neural pathways related to memory, resilience, and deep thinking.”
Ironically, the paper itself became a case study in AI misuse. Some users summarized it using ChatGPT, prompting hallucinated facts—like falsely stating the version of ChatGPT used was GPT-4o. Kosmyna had anticipated this and included “AI traps” in the document to test such behavior.
MIT researchers are now expanding their work into programming and software engineering, and early results are even more troubling—suggesting broader implications for industries seeking to automate entry-level tasks.
While previous studies have highlighted AI’s potential to boost productivity, this research underscores the urgent need for responsible AI use in education, backed by policies that balance efficiency with brain development.
OpenAI did not respond to a request for comment. Meanwhile, the debate on the role of AI in learning continues—with growing calls for regulation, transparency, and digital literacy.
