For all the good it brings to our lives, artificial intelligence (AI) has plenty of shortcomings as well, including our increasing overreliance on large language models (LLMs) when it comes to writing, which has proven to not just make us lazier but also harm our brains and thinking skills.
As it happens, a new study by the Massachusetts Institute of Technology (MIT) has explored the neural and behavioral consequences of LLM-assisted essay writing, showing clear neurological evidence that it may reshape how we focus, create, and critique.
Specifically, the researchers had 54 students write SAT-style (Scholastic Aptitude Test taken in the US to measure students’ abilities before they go to college) essays across four sessions while high-density electroencephalogram (EEG) tracked information flow among 32 brain regions.
They divided the students into groups according to the tools they used for writing – no aid (‘brain only’), Google search, and GPT-4o. After the initial testing, they flipped the groups, where students who had written unaided rewrote with GPT (brain -> LLM), and those using GPT had to write solo (LLM -> brain).
Results discover cognitive debt
Researchers discovered that the neural networks dimmed when the participants offloaded their creativity to the LLM, with pure GPT use producing the weakest fronto-parietal and temporal connectivity among all conditions, indicating lower executive control and poorer semantic processing.
The results differed when it came to order, as well. In students who tackled ideas on their own and then revised with GPT, brain-wide connectivity soared and exceeded every earlier GPT session. However, those who started with GPT and then worked without it showed the lowest coordination and relied on GPT-favored vocabulary.
Moreover, none of the AI-assisted writers could later quote a sentence they had just written, whereas nearly every unassisted author could, and the deficit persisted even after practice.
Finally, the study’s authors realized that cognitive debt accumulates, with repeated GPT use reducing topic exploration and diversity, and writers struggling to recover the breadth and depth of their earlier human-only work without AI assistance.
They defined this cognitive debt as a “condition in which repeated reliance on external systems like LLMs replaces the effortful cognitive processes required for independent thinking.” Furthermore, they said:
“Cognitive debt defers mental effort in the short term but results in long-term costs, such as diminished critical inquiry, increased vulnerability to manipulation, decreased creativity. When participants reproduce suggestions without evaluating their accuracy or relevance, they not only forfeit ownership of the ideas but also risk internalizing shallow or biased perspectives.”
Elsewhere, it seems like LLMs are becoming more advanced each day, raising copyright concerns as the likes of Llama 3.1 recall 42% of the first Harry Potter book, and fake bands and artificial songs flood streaming platforms like Spotify and YouTube.
ReadNext
