Researchers at the Massachusetts Institute of Technology (MIT) have released a preprint study that is being described as the first brain scan investigation of ChatGPT users. The paper, titled “Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task”, suggests that prolonged use of artificial intelligence tools may reduce, rather than enhance, mental engagement.
Although the research is not yet peer reviewed, it provides a significant early glimpse into how reliance on artificial intelligence may shape brain activity, memory, and learning.
How the Study Was Designed
The MIT team recruited 54 participants and divided them into three groups. The first group used ChatGPT to assist or generate essays. The second group used a search engine such as Google. The third group wrote essays without any digital assistance.
Across four sessions, participants wrote under these conditions. During the fourth session, participants were switched: those in the ChatGPT group wrote without any tool, while those in the unaided group began to use ChatGPT. Brain activity was monitored using electroencephalogram (EEG) scans covering 32 regions. Essays were also evaluated using natural language processing and human graders.
A detailed overview of the project can be found on the MIT Media Lab website.
Memory and Recall Problems
According to the MIT Media Lab summary, participants in the ChatGPT group had difficulty recalling what they had just written. Participants who wrote without artificial intelligence did not show this problem.
The researchers wrote in their arXiv preprint that “LLM users also struggled to accurately quote from the essays they wrote just minutes prior”. This suggests that outsourcing writing to artificial intelligence may interfere with the brain processes involved in encoding and retaining information.
Weaker Neural Connectivity and Engagement
EEG connectivity scores were highest for the group writing without any tools and lowest for the ChatGPT group. The study reported that neural connectivity “scaled down in relation to external tool use” (arXiv).
Even after switching to unaided writing, the ChatGPT users continued to show under-engagement in their brain scans. By contrast, participants who first wrote without artificial intelligence and then adopted ChatGPT retained stronger brain activity and better recall.
Further details on the neural findings are explained in MIT’s project overview.
Writing Style and Perception
The study also assessed the essays themselves. Evaluators found that the essays produced with ChatGPT were technically sound but described them as “homogenous” and lacking originality. Participants who wrote without artificial intelligence showed stronger self-reported ownership of their work. These findings are outlined in the MIT Media Lab publication.
Interpreting the Results
Media coverage, including reports from Time and the New York Post, has amplified the potential consequences of these findings. However, the preprint itself does not provide some of the extreme numerical claims circulating online, such as a specific “47 per cent drop” or “83.3 per cent memory loss”.
Nevertheless, the evidence clearly shows reduced neural engagement and weaker recall for those relying heavily on artificial intelligence. The authors caution that their research is preliminary and limited to essay writing tasks.
The Productivity Paradox
One key implication of the MIT study is that speed and output may not equal learning. Artificial intelligence tools like ChatGPT can make writing faster, but the mental effort required for deeper processing and retention may decline. This aligns with concerns about “cognitive debt”, where overreliance on external systems leads to hidden losses in mental capacity.
An accessible overview of this concept is available on the MIT Media Lab project page.
Using Artificial Intelligence Intentionally
The findings suggest that people should use artificial intelligence tools to support, rather than replace, their own thinking. Strategies that may help include alternating between unaided and assisted work, building personal drafts before using ChatGPT, and actively revising or questioning AI outputs.
By approaching AI as an assistant rather than a replacement, individuals may preserve cognitive engagement and avoid long-term dependence.
The MIT brain scan study provides a critical early warning about the cognitive impact of overreliance on artificial intelligence. Delegating too much intellectual work to AI may save time but could erode memory, originality, and neural connectivity.
Further research, replication, and peer review are required, but the study signals that the way people use artificial intelligence matters as much as the tools themselves.