A recent study from the Massachusetts Institute of Technology has uncovered compelling evidence that relying on large language models (LLMs) like ChatGPT not only reduces immediate brain effort—but also hampers mental performance in subsequent tasks.
The Experiment Setup
Researchers recruited a small group of participants to write essays across various topics, dividing them into three distinct cohorts:
· AI-assisted group (using ChatGPT)
· Search-assisted group (using Google Search)
· Unaided group (relying solely on their own minds)
Each participant’s brain activity was monitored using electroencephalography (EEG), allowing scientists to measure cognitive load and neural engagement. The findings were striking: the more technological assistance participants received, the less their brains exerted effort. The unaided group showed the highest levels of neural connectivity, followed by the search group, with the AI users exhibiting the least brain activity.
Cognitive Ownership & Essay Quality
The study also explored a concept called “ownership”—the ability of participants to recall and summarize their own work. Those who used AI struggled significantly to quote or explain what they had written. Moreover, their essays were notably uniform, lacking the diversity and nuance seen in the other groups.
Interestingly, the visual cortex was more active in the AI and search groups, suggesting a shift in focus toward tool-generated outputs rather than internal cognitive processing.
Long-Term Cognitive Effects
To assess lasting impacts, researchers reshuffled participants into two new groups:
· Brain-to-LLM: previously unaided participants now using AI
· LLM-to-Brain: AI users now writing without assistance
The results revealed a cognitive divide. Former AI users showed diminished neural connectivity and under-activation of key brain networks. In contrast, those transitioning from unaided work to AI demonstrated improved memory recall and re-engagement of critical brain regions—suggesting that AI can enhance cognition when used after deep personal reflection, but not as a starting point.
Implications & Warnings
Over the course of four months, the AI-first group consistently underperformed across neural, linguistic, and scoring metrics. The researchers caution that habitual reliance on AI tools may erode essential thinking and learning skills.
While the study’s limited sample size calls for broader research, its implications are timely. As AI becomes embedded in education and daily life, the authors warn of a “pressing matter”: the potential decline in human cognitive development when AI replaces rather than augments thought.
Final Takeaway
The study suggests a nuanced approach to AI use. When humans engage deeply with a topic first, AI can serve as a powerful enhancer. But when used as a shortcut from the outset, it may dull the very faculties it aims to support.
As search engines increasingly prioritize AI-generated results, the researchers urge caution. The long-term effects of this shift on everyday cognition remain uncertain—and worthy of further investigation.
