MIT Study Finds AI Diminishes Human Brain Engagement

A study conducted by MIT (Massachusetts Institute of Technology) reveals that using a large language model (LLM) not only reduces mental effort in the moment, but also has lingering negative effects on cognitive performance in subsequent tasks.

In the experiment, researchers worked with a small number of participants—a limitation acknowledged in the study [PDF]—who were assigned to write essays on various topics. One group was permitted to use AI (ChatGPT was selected, as researchers saw little functional difference between it and similar tools), another could use Google Search, and a third group was designated as ‘brain only,’ meaning they had to complete the task without any technological assistance.

All participants underwent electroencephalography (EEG) to assess cognitive engagement and mental load. The results showed distinct patterns of neural connectivity across the groups, reflecting different cognitive strategies. The more technological support a participant received, the lower their observed brain activity. EEG analysis confirmed that the ‘brain only’ group exhibited the highest level of neural engagement, followed by the search engine group, with AI users showing the least.

The study also measured what it called ‘ownership’—the ability of participants to recall and summarize their own writing afterward. Ownership dropped significantly as technological aid increased. Very few individuals in the LLM-assisted group could accurately paraphrase their essays. Moreover, essays written with AI assistance were notably uniform per topic, showing far less variation compared to the other groups.

As expected, the visual cortex was more active among those using search engines or ChatGPT, indicating greater focus on the output of the tools themselves.

Longer-term effects

After multiple rounds of essay writing, two new groups were formed: ‘Brain-to-LLM’—participants who initially worked unaided and were later allowed to use an LLM—and ‘LLM-to-Brain’—those who began with AI support and were later required to write independently.

The researchers observed that “LLM-to-Brain participants displayed weaker neural connectivity and reduced engagement in alpha and beta frequency networks, while Brain-to-LLM participants demonstrated better memory recall and reactivation of widespread occipito-parietal and prefrontal brain regions. […] This suggests that turning to AI only after engaging deeply in a topic promotes cognitive integration, memory reactivation, and top-down control.”

In essence, people benefit from AI assistance most when they first thoroughly explore a subject using their own intellect, experience, and knowledge. However, relying on AI from the start appears to gradually reduce cognitive engagement and impairs performance when the AI support is removed.

According to the paper, “Over the four-month study period, LLM-assisted participants consistently underperformed compared to the brain-only group across neural, linguistic, and scoring metrics.”

Limited study

Given the small sample size, the authors recognize the need for broader and more diverse participant groups to produce statistically robust conclusions. Still, as AI becomes more integrated into education and daily life, the researchers emphasize the urgent concern that overreliance on AI could lead to a decline in essential learning abilities.

Conclusions

If the trend of substituting human reasoning, reflection, and synthesis with AI-generated content continues, critical thinking skills may deteriorate over time. Using AI to supplement one’s own thought process—rather than replacing it entirely—leads to better outcomes.

Search engine use occupied a middle ground between unaided thinking and AI-dependent composition. However, with tech companies like Google and Microsoft increasingly integrating AI-generated responses into search results—often placing them at the top of the page—the cognitive demand on everyday users may decrease if they rely solely on these automated summaries.

The research team concludes that further investigation is needed to understand the long-term cognitive impact of AI “before LLMs can be considered an unequivocal benefit to humanity.”

Interested in learning more about AI and big data from industry experts? Attend the AI & Big Data Expo in Amsterdam, California, or London. This comprehensive event is part of TechEx and co-located with other top tech conferences. Click here for details.

AI News is brought to you by TechForge Media. Discover more upcoming enterprise tech events and webinars here.