Key Takeaways:
- Using AI too early in writing weakens brain engagement.
- Overreliance on tools like ChatGPT erodes memory and authorship.
- Starting without AI preserves cognitive effort and originality.
- Long-term dependence may affect how future brains develop.
When calculators arrived in classrooms, many feared students might lose the ability to perform basic math. That fear faded over time, as education adapted to balance conceptual understanding with tool use. But today, AI tools like ChatGPT are stirring a new kind of concern—one that touches not just on skill, but on cognition itself. What happens to the brain when a machine starts thinking for us?
AI Effect on the Brain
A recent preprint study from the MIT Media Lab doesn’t offer a simple answer, but it does reveal a deeply important one. While AI can undoubtedly boost efficiency, the research asks something more profound: what is the cost to the mind when we outsource the act of writing?
How the Study Was Done
In this controlled experiment, 54 college-age participants were split into three groups to write SAT-style essays under different conditions:
- One group used no tools at all (the “Brain-only” group).
- Another used a search engine.
- The third used ChatGPT as a writing assistant (the “LLM” group).
Each participant wrote three essays across three separate sessions. In a fourth session, 18 of them switched methods: LLM users wrote without AI, and Brain-only users were introduced to it.
Throughout, participants wore EEG caps that recorded real-time brain activity, tracking neural engagement across cognitive networks. The essays were scored by human educators and an AI judge. After each session, participants were also interviewed to assess how well they remembered their writing and whether it felt like their own.
How the Brain Responds to Writing With and Without AI
The results were striking. The Brain-only group showed the strongest and most widespread brain activity. Their EEG readings revealed robust connectivity between areas responsible for attention, memory, and executive function. They recalled their writing well and expressed high ownership over their ideas.
The Search Engine group showed moderate engagement. While they relied on external input, they still had to synthesize, search, and decide, keeping much of the cognitive engine running.
But in the ChatGPT group, brain activity dropped sharply. Neural signals in key bands—especially alpha and beta frequencies linked to attention and semantic processing—were significantly weaker. Writing became more passive. Thinking, it seems, was being quietly sidelined.
In session four, when LLM users had to write without AI, their brains did not re-engage. The drop persisted. Their essays remained flat and repetitive. Meanwhile, those who switched from Brain-only to using ChatGPT actually showed a bump in connectivity. Their prior cognitive effort had laid the groundwork for a more strategic and active use of the AI.
How AI Use Diminishes Memory and Ownership
Cognitive activity wasn’t the only thing that changed. So did memory and authorship. Nearly all Brain-only participants could quote their essays accurately by their second session. They knew what they’d written—and why.
The LLM group couldn’t. In session one, none of them could correctly quote a single sentence from their own essays. Even by session three, memory remained shaky. Ownership was fragmented. Some admitted the work didn’t feel like theirs. Others estimated that the AI had written 50 to 90 percent of it.
These impressions weren’t just psychological. EEG data showed reduced connectivity in brain regions involved in self-monitoring and verbal memory. The machine hadn’t just helped them write—it had overwritten their cognitive imprint on the task.
The Accumulation of Cognitive Debt
Perhaps the most unsettling insight from the study is what the researchers call “cognitive debt.” Like credit card debt, it builds gradually. The more users relied on ChatGPT, the more their mental engagement declined, both neurologically and behaviorally. Their writing became shorter, more templated, and less rich. Their thought patterns narrowed.
And here’s the key: once that pattern was set, they couldn’t easily break it. Even when the AI was removed, the brain didn’t snap back. The neural infrastructure of original thinking had been dulled.
This wasn’t just a one-time effect. It was a compounding shift in how the brain handled expression, memory, and complexity. The shortcut had changed the route.
AI Should Be Used as Support, Not a Substitute
But there was hope. Participants who began the study without tools and later added ChatGPT showed the opposite trend. Their brain activity increased. Their essays remained detailed and expressive. They still remembered what they wrote. And they reported that the AI helped—but didn’t replace—them.
This suggests that it’s not AI itself that causes decline, but rather the timing of its use. When foundational thinking precedes the tool, the brain remains in charge. When the tool comes first, the brain steps back.
What Could This Mean for the Future?
The researchers stopped short of making long-term predictions, but reading the findings, it’s hard not to reflect on broader implications. Could an entire generation, raised on AI-assisted writing, lose the neurological fitness needed for deep thought?
History offers examples: GPS has eroded our sense of direction. Social media has fractured our attention. Could generative AI gradually dampen our ability to reason independently or reflect deeply?
Perhaps. The brain is shaped by what it’s asked to do. And if it’s asked to do less—because a tool does it better, faster, and with fewer mistakes—then over time, that capacity may erode.
The concern isn’t that people will be less smart. It’s that they might become less practiced in the very mental functions that define learning: generating, connecting, reflecting, and remembering.
Related Reading:
AI’s Impact on Jobs: Conflicting Messages from the Companies Leading the Charge
FAQs
What was the goal of the MIT study?
To measure how different writing tools—ChatGPT, search engines, or none—affect brain activity, memory, and learning during essay writing.
How many people took part?
Fifty-four participants were involved in the main sessions. Eighteen returned for a fourth session where they switched writing methods.
What tools did the groups use?
One group used no tools (Brain-only), another used search engines, and a third used ChatGPT (LLM). In the final session, some switched tools to see the cognitive impact.
What did the EEG scans show?
Participants who wrote without tools showed the highest brain activity. Those who used ChatGPT showed the lowest engagement, even after switching back to tool-free writing.
How did AI use affect memory?
ChatGPT users struggled to recall or quote what they wrote. Brain-only participants had strong memory retention by session two.
Did participants feel connected to their AI-written work?
No. Many in the AI group said the essays didn’t feel like their own. Brain-only writers had the strongest sense of authorship.
Was AI helpful when introduced later?
Yes. Participants who started without tools and added AI later retained better memory, brain engagement, and control over their writing.
What is ‘cognitive debt’?
It’s the long-term mental cost of relying on AI to do thinking for you—less memory use, weaker attention, and dulled critical skills over time.
Does the study suggest banning AI?
No. It warns against using AI too early in the writing process, before your own thinking is fully developed.
Can AI support learning if used properly?
Yes. When used after independent thinking, AI can help refine and expand ideas without replacing core mental effort.
What’s the risk of relying on AI long-term?
Habitual dependence on AI could reduce our use of memory, attention, and reasoning abilities, which may weaken over time if not exercised.
Could this impact future generations?
Possibly. If reliance on AI becomes normal from an early age, cognitive development may shift. Over generations, even brain structure could adapt.
Can AI use affect brain development?
Yes. The brain adapts to how it’s used. Less mental effort can weaken key functions.
Is this evolution?
Not yet—but if these habits persist across generations, it could lead to evolutionary change.
Could future humans think differently?
Possibly. If deep thinking becomes unnecessary, the brain may adapt to a simpler cognitive load.
Has this happened before?
Yes. GPS weakened spatial memory. Smartphones shortened attention.
What’s the takeaway for students and professionals?
Write first. Think through your ideas. Then use AI as a tool, not a starting point.
Can AI use raise dementia risk?
Possibly. Less mental effort over time may increase vulnerability.
Does brain activity protect against dementia?
Yes. Active thinking helps delay cognitive decline.
Final Thoughts
This study doesn’t argue that AI should be banned from writing. It calls for being deliberate, especially about when and how we use it. Tools like ChatGPT are here to stay, and when used well, they can sharpen ideas and support clarity. But introduced too early—or leaned on too often—they can quietly erode the very skills they’re meant to support.
Writing isn’t just putting words on a page. It’s a mental process that forces us to slow down, retrieve knowledge, organize complex ideas, and make sense of what we think. It challenges memory, strengthens attention, and deepens understanding. That effort isn’t extra—it’s essential. When we skip it, we’re not just outsourcing the task. We’re giving up the opportunity to actually learn.
If you’re an educator, consider when your students first engage with AI. Create space for slow thinking—moments where struggling through a sentence is part of learning. If you’re a student or professional, don’t let your first draft come from a tool. Start with your own reasoning. Then use AI to test, refine, or extend what you’ve already built—not to skip the building.
Writing this way doesn’t just make you a better communicator—it keeps the machinery of thought active. When we stop using those cognitive functions, they don’t disappear overnight. They fade. And getting them back can be harder than we think.
The risk isn’t that AI becomes smarter than us. The risk is that we gradually stop thinking for ourselves and don’t realize it until the habit is set. And if this shift continues over years or generations, the effects may not stop at behavior. Evolution doesn’t preserve what isn’t used. If memory, attention, and independent reasoning become less necessary, even the structure of the human brain may begin to adapt.
Could this also influence when cognitive decline begins—possibly even increasing the risk or early onset of conditions like dementia? It’s a question worth asking now, before patterns become permanent.
The change won’t be sudden. It will be quiet. And it starts with one shortcut at a time.
References
Kosmyna, N., Hauptmann, E., Yuan, Y. T., Situ, J., Liao, X.-H., Beresnitzky, A. V., Braunstein, I., & Maes, P. (2025). Your brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing task (arXiv:2506.08872). arXiv. https://doi.org/10.48550/arXiv.2506.08872
