MIT Study Warns of Cognitive Decline with LLM Use

A recent study from MIT Media Lab reveals how the increasing use of large language models (LLMs) like ChatGPT may be quietly dulling our cognitive skills, especially in complex tasks like writing and critical thinking. The findings, part of a study titled ‘Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task’, show that while AI tools can improve productivity, they come with measurable mental trade-offs.

In an effort to understand this phenomenon, the researchers split participants into three groups: one used an LLM (ChatGPT), another used a traditional search engine, and the third used only their brain—no external assistance. Through EEG (electroencephalogram) brain monitoring and essay analysis, the study found that LLM users showed significantly weaker neural connectivity, indicating lower cognitive effort and engagement compared to others.

“Brain connectivity systematically scaled down with the amount of external support,” the paper stated. Those who relied solely on their brain had the strongest and widest-ranging neural networks, while LLM users had the weakest overall coupling.

Shockingly, LLM users were far less able to recall or quote from essays they had just written. In fact, 83.3% of LLM users failed to provide even a single correct quote from their own writing, compared to just 11.1% in both the Search Engine and Brain-only groups.

The study also highlighted what it calls “cognitive offloading”—outsourcing thinking to AI tools—which leads to a long-term decline in learning capabilities. While LLMs reduce immediate cognitive load and make tasks feel easier, this reduction comes at the cost of deep engagement and memory formation.

This “offloading” also seemed to lead to what the study calls a ‘metacognitive laziness’, where users skip the mental work of integrating ideas and reflecting critically on content.

The implications are especially important for educators. The study suggests that over-reliance on AI writing tools could weaken students' long-term ability to think critically, recall information, and construct original arguments. As one participant noted in their interview, “I would rather use the internet than ChatGPT—I can read other people’s ideas.”

Researchers caution that while AI can be a powerful assistant, its use must be deliberate and well-regulated in learning environments. “We demonstrate the pressing matter of a likely decrease in learning skills,” the authors concluded, adding that LLMs can make students more productive, but less cognitively present.