ChatGPT Use May Reduce Cognitive Engagement During Learning Study Finds

ChatGPT has quickly become part of everyday work and study, but new research is raising questions about what happens when people rely on it too heavily for thinking and writing tasks.

A study from the MIT Media Lab examined how different levels of external assistance affected people while writing essays. The researchers compared participants using ChatGPT, a search engine and no external tools. The findings suggested that the ChatGPT group showed the weakest overall brain connectivity during the task.

However, the findings need some context. The MIT team itself describes the research as preliminary and says it has not been peer reviewed. The study also focused specifically on essay writing, so the results cannot automatically be applied to every kind of ChatGPT use.

What the MIT study found

The research involved 54 participants across the first three sessions, with participants divided into three groups. One group used an LLM, specifically ChatGPT, another used a search engine and the third completed the work without external tools.

Each participant completed a series of essay writing sessions over four months. Researchers used EEG measurements to examine brain activity and also assessed the written work using linguistic analysis and other evaluation methods.

The researchers reported that the brain connectivity of the three groups differed. The brain only group showed the strongest and most widely distributed connectivity, while the search engine group showed an intermediate level. The ChatGPT group recorded the weakest overall coupling.

Memory and sense of ownership also stood out

The study did not only look at brain activity. Researchers also examined how well participants remembered and connected with the essays they had produced.

According to the findings, people in the ChatGPT group had more difficulty accurately recalling quotations from their own essays. They also reported a lower sense of ownership over their work compared with participants who wrote without external assistance.

The researchers described this pattern using the term cognitive debt. The idea is that relying heavily on an external tool may reduce the amount of mental effort invested in a task, potentially affecting engagement and learning in that particular setting.

An interesting part of the research came in the fourth session. Some participants who had previously used ChatGPT were asked to complete the task without it, while some participants from the brain only group were moved to the LLM condition. The researchers observed weaker neural connectivity in the participants who moved from LLM use to working without the tool.

Does ChatGPT actually make people less intelligent

That conclusion would be too strong based on this study alone.

MIT has specifically cautioned against describing the findings as evidence that ChatGPT makes people dumb or causes brain damage. The researchers say the work is an early investigation into the cognitive costs of AI assistance, not proof that AI permanently reduces intelligence.

There are also clear limitations. The participant pool was relatively small and came from a specific geographic and academic setting. The researchers noted that the experiment focused on essay writing and that the results may not apply to other tasks or to every AI model.

What it could mean for students

The study does offer a useful warning for students. Using ChatGPT to understand a difficult topic, generate ideas or check completed work is different from handing an entire thinking task over to an AI system.

For learning, doing some of the difficult work independently can still matter. Writing an answer, recalling information and working through a problem without immediate assistance can require more active mental engagement.

So the bigger issue may not be whether students use ChatGPT, but how they use it. Treating AI as a shortcut for every assignment could reduce opportunities to practise independent thinking, while using it as a supporting tool may leave more room for the student to do the actual reasoning.

The MIT research therefore adds to an ongoing discussion about AI and education. More studies involving larger and more diverse groups will be needed before researchers can determine the long term effects of frequent AI assistance on learning, memory and reasoning.