Researchers Reveal Students Who Use AI Models To Write Essays Face Cognitive Challenges

A recent MIT study, focused on the cognitive cost of using AI models to write essays, revealed that students who rely more on large language models (LLMs) may face harmful consequences and cognitive challenges.

The study, titled Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task, found that the use of an AI models can significantly affect students and their learning processes, including what the researchers call a “cognitive cost.”

The research involved 54 participants and revealed that the group using ChatGPT to write essays showed weaker neural connectivity and had difficulty remembering and quoting their own essay just minutes after finishing the task.

While the research team acknowledged the limitations of their small sample size, they hope the findings will serve as “a preliminary guide to understanding the cognitive and practical impacts of AI on learning environments.”

For the study, the researchers divided the participants into three groups: one that could use LLMs such as ChatGPT, another that could access traditional search engines like Google, and the third group that could only use their knowledge—called the Brain-only group.

The participants completed four essay writing and analysis sessions—three with the original group setup, and a final session in which access to tools was changed, requiring the LLM group to write using only their brains.

As measurement instruments, the scientists used an electroencephalography (EEG) to register brain activity considering engagement, and load—Scientists have also recently developed an e-tattoo to detect mental fatigue. The study also included NLP analysis, participant interviews, and essay scoring by both human teachers and an AI tool.

The experts revealed a strong correlation between brain connectivity and the use of external tools. The Brain-only group had the highest levels of neural connectivity, while those who used AI showed the weakest.

Memory retention was also negatively affected. The group that used AI models had more difficulty quoting their own essays and reported the lowest levels of “ownership” over their work.

“As the educational impact of LLM use only begins to settle with the general population, in this study we demonstrate the pressing matter of a likely decrease in learning skills based on the results of our study,” wrote the researchers. “The LLM group’s participants performed worse than their counterparts in the Brain-only group at all levels: neural, linguistic, and scoring.”