Cognitive debt is not permanent damage

An MIT Media Lab preprint links LLM essay help to lower engagement and recall. That is a warning about practice, not proof of permanent injury.

The trap

Headlines jump from “brain scans looked quieter” to “AI permanently damaged your brain.” That leap is not what careful reading allows. Clear eyes means you take cognitive cost seriously without inventing permanence.

What the evidence shows

Kosmyna et al. (2025) reported a preprint study (MIT Media Lab project Your Brain on ChatGPT) in which participants wrote essays with an LLM, with a search engine, or with no tools. Across sessions, the LLM group showed weaker neural connectivity patterns and poorer ability to quote their own essays than the brain-only group, a pattern the authors discussed as cognitive debt. The Media Lab’s own explainer (MIT Media Lab, 2025) stressed preliminary status: as of the preprint release, the work was not peer-reviewed, the sample was limited, and results should be treated with caution.

This sits beside older, peer-reviewed evidence that people offload memory when they expect external access. Sparrow et al. (2011) found that expecting future access to information reduced recall of the content itself and increased memory for where to find it.

Neither line of work proves irreversible brain injury. Both support a craft rule: if you never struggle with the first draft, you may not own the knowledge.

What this means for people

Students and knowledge workers who outsource every draft can feel fluent and still fail a closed-book ask. The fix is practice design, not fear theater.

Practice (15 minutes)

Pick one real writing task this week.
Draft the outline or first 150 words with no model.
Only then use AI for critique or expansion.
Close the chat and write three sentences you can still defend from memory.

Reflection

Where are you collecting cognitive debt because convenience feels like competence?

Skeptic check

Kosmyna et al. (2025) is a preprint, small sample, essay task in one age band; not “permanent damage.”
EEG connectivity differences are not the same as clinical injury.
Sparrow et al. (2011) studied search-era memory, not ChatGPT specifically.

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 [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2506.08872

MIT Media Lab. (2025). Is using ChatGPT to write your essay bad for your brain? New MIT study explained. https://www.media.mit.edu/articles/is-using-chatgpt-to-write-your-essay-bad-for-your-brain-new-mit-study-explained/

Sparrow, B., Liu, J., & Wegner, D. M. (2011). Google effects on memory: Cognitive consequences of having information at our fingertips. Science, 333(6043), 776-778. https://doi.org/10.1126/science.1207745