- AI users showed weaker brain connectivity, not brain damage.
- Writing independently before using AI preserved memory and engagement.
- About 14 percent of workers report cognitive fog from AI.
Nataliya Kosmyna had one request for journalists covering her research.
Stop calling it brain rot.
The MIT Media Lab computer scientist had spent four months studying AI and critical thinking, tracking what happens inside people's heads when they write with ChatGPT. Her 206-page preprint, released in June 2025, became one of the most discussed neuroscience papers of the year.
It also became one of the most misrepresented.
AI and Critical Thinking: A Complicated Story
Kosmyna's team fitted 54 Boston-area university students with EEG caps and asked them to write SAT-style essays across three conditions. One group used only ChatGPT. Another used Google Search. A third worked without any digital tools at all.
Students writing without tools showed the strongest neural connectivity, particularly in regions tied to memory, creativity, and semantic processing. The Google Search group fell in the middle. ChatGPT users displayed the weakest brain coupling, especially in alpha and theta bands associated with deep learning.
By their third essay, many in the ChatGPT group had stopped composing entirely.
The alarming headlines wrote themselves.
But Kosmyna's own interpretation was more careful.
What is cognitive debt?
Cognitive debt describes the accumulated cost of outsourcing mental work to AI. Like financial debt, small deferrals compound over time. The brain regions responsible for synthesis, memory encoding, and critical evaluation get less practice, potentially weakening their capacity for future independent work.
Sequence Matters More Than the Tool Itself
The study's most revealing finding came from a smaller fourth session involving 18 participants. Students who had written independently were now given ChatGPT. Those who had relied on ChatGPT had to write alone.
Students who thought first and used AI second maintained strong neural activation and recalled their work accurately. Those who started with AI and switched to independent writing struggled to engage the same cognitive networks.
Adam Green, a cognitive neuroscientist at Georgetown University and co-founder of the Society for the Neuroscience of Creativity, noted that this aligns with concerns about AI displacing idea generation. But he cautioned that only 18 people participated in this crossover phase, adding real uncertainty to the findings.
The distinction matters. The study measured cognitive engagement during a task, not cognitive capacity itself.
Key figure
78%
of ChatGPT users could not accurately quote their own essays
Workplace Research Adds a New Dimension
The conversation has moved beyond the lab.
Harvard University and Boston Consulting Group surveyed more than 1,400 American workers and identified a phenomenon they called "AI brain fry." About 14 percent reported mental fog after intensive AI interaction, with oversight tasks, where workers monitored AI outputs rather than creating them, driving the strongest effects.
The task was executed, but you basically didn't integrate any of it into your memory networks.
Nataliya Kosmyna, MIT Media Lab
Workers juggling more than three AI tools simultaneously saw their productivity decline. Those doing AI oversight reported 12 percent more mental fatigue than colleagues in other roles.
But the same survey found that AI reduced burnout when it replaced routine, repetitive tasks. The pattern echoed Kosmyna's finding. AI as assistant lightened the load. AI as substitute for thinking increased it.
More On Students Using AI
Writing as Thinking: What Students Lose When AI Does the Work
Harvard neuroscientists explain why writing as thinking matters, and what students lose when AI writes their essays.
→A January 2026 paper in Nature proposed a framework for this divide. Passive reliance on AI, the authors argued, may weaken the brain's activity-dependent plasticity over time. Active co-creation, where the human remains the one making meaning, could preserve or strengthen it.
Kosmyna's preprint has not been peer-reviewed, and its sample remains small. The question it raises has only grown more urgent.
The answer appears to be not whether people use AI, but whether they think before they do.
Sources
- Primary Source: Your Brain on ChatGPT: Accumulation of Cognitive Debt (MIT Media Lab)
- Additional Context:
- Does Using ChatGPT Really Change Your Brain Activity? (Scientific American)
- When Using AI Leads to Brain Fry (Harvard Business Review)
- The brain side of human-AI interactions in the long-term (Nature, npj AI)
Fact Check: Claim-by-Claim Verification Verified
The article accurately summarizes the MIT preprint's findings, quotes, and nuances, with minor approximations in secondary stats supported by reliable reporting on real studies.
Commentary
- Preprint not peer-reviewed, small crossover sample (n=18) noted as limitation—article appropriately hedges.
- 78% recall failure approximated (sources cite 80%+ struggled, none perfect); aligns closely with findings.
- Adam Green (Georgetown, Society for Neuroscience of Creativity) commented on small sample, as stated.
- Nature/npj AI paper link exists but unverified in detail; core claim on passive vs. active AI use consistent with emerging research.
Sources used for verification
Academic/Peer-reviewed:
- Your Brain on ChatGPT: Accumulation of Cognitive Debt... - arXiv
- Your Brain on ChatGPT (MIT Media Lab) - media.mit.edu
Other reliable sources:
- ChatGPT's Impact On Our Brains According to an MIT Study - time.com
- Does Using ChatGPT Really Change Your Brain Activity? - scientificamerican.com
- Professor Adam Green Interviewed in Nature - psychology.georgetown.edu
- AI overuse could spark "brain fry," new research finds - axios.com
Fact-checked by Perplexity Sonar Pro on 2026-03-10
