Why MIT’s Brain-Scan Study on ChatGPT Misreads Human Cognition: A Psychextrics Rebuttal
By: OMOLAJA MAKINEE
MIT’s recent brain-scan study has made headlines by warning that ChatGPT “dulls” human cognition. The results are striking: reduced memory recall, diminished neural connectivity, and what researchers describe as “soulless” writing. Yet from the standpoint of psychextrics—a behavioural science that examines how neurotype genes link to emotional and expressive variants—this conclusion reflects a category error. Rather than showing that ChatGPT weakens cognition, the study demonstrates the mismatch between inherited neurotype spectrums of perception–expression and the narrow metrics used to measure cognition.
1. Misinterpreting Memory: Offloading vs. Weakening
The MIT study places heavy weight on the finding that 83.3% of users could not recall sentences they had written with ChatGPT minutes earlier. This is framed as “cognitive weakening.” Under psychextrics, however, this is better understood as cognitive offloading, a natural phenomenon observed whenever humans externalise memory into tools. Writing itself, since antiquity, was criticised by Socrates in Plato’s Phaedrus for eroding memory—yet we now understand writing as a memory-encoding amplifier, not a cognitive disease.
What the scans reveal is not dulling, but a shift in neural pathways. When individuals write with AI, the thalamic nuclei responsible for speech-expression (ventral anterior and ventrolateral nuclei) and writing-expression (medial dorsal nucleus) redistribute their cognitive load. The brain stores less verbatim recall because it is outsourcing surface structure to the machine, but it retains conceptual encoding—the deeper scaffolding of meaning. To judge this as “weakening” is to confuse where cognition is happening, not whether it is happening at all.
2. The Fallacy of Productivity Metrics
MIT’s study admits a paradox: ChatGPT increases productivity by 60% while reducing “mental effort” by 32%. But from a psychextrics standpoint, this is not a paradox at all. Human cognition evolved to minimise unnecessary strain while maximising adaptive reasoning. In terms of the thalamic spectrum, ChatGPT accelerates low-level linguistic assembly (words, syntax) so that higher-order reasoning, imagination, and memory integration can be given more bandwidth.
Measuring “brain activity” by raw connectivity misses this crucial nuance. A reduction in distributed neural chatter can signal greater efficiency, not decline—similar to how expert pianists show less cortical activation than novices when playing complex passages, because their brain has streamlined the skill. The MIT interpretation risks mistaking efficiency for erosion.
3. The Question of “Soulless” Writing
Educators in the study flagged AI-assisted essays as “robotic” or “soulless.” This critique again reflects stylistic bias, not cognitive deficit. Psychextrics demonstrates that eloquence and intelligence fall on different thalamic spectrums: the medial dorsal nucleus underpins stylistic fluency, while the anterior nuclei privilege substantive reasoning. A student with average eloquence but deep reasoning may appear “flat” on the page, yet their intelligence remains intact.
AI reveals this fissure by flattening stylistic differences into uniform outputs. But rather than proving that users are “weaker,” it shows that individuality in expression has historically been overvalued as a proxy for intelligence. The charge of “soullessness” is therefore not a diagnosis of dull minds, but a mirror reflecting educational systems that prize rhetorical sheen over conceptual clarity.
4. Why the “Later Users” Performed Best
The most intriguing finding of MIT’s study is that participants who began without AI and later adopted it outperformed all other groups. Psychextrics predicts this outcome: the neurotype spectrum thrives when the brain first encodes raw expression in its natural style (literal or abstract, oral or written), and only then integrates augmentation. This sequencing allows the medial dorsal and pulvinar nuclei to encode both the self’s native expressive style and the AI’s efficiency gains in parallel.
This finding does not show that AI use “dulls” cognition; rather, it underscores the importance of sequencing AI as augmentation, not replacement. Just as calculators work best when users first understand arithmetic, ChatGPT sharpens cognition when used after initial encoding of human expression.
MIT Study vs. Psychextrics Reinterpretation
| MIT Finding / Claim | Interpretation in MIT Framework | Reinterpretation under Psychextrics |
|---|---|---|
| Memory loss: 83.3% of users couldn’t recall a single sentence written with ChatGPT | AI “weakens” memory; users fail at short-term recall. | Cognitive offloading, not weakening. Memory shifts from verbatim recall to conceptual scaffolding. Similar to how writing historically externalised memory. |
| Brain connectivity dropped by 47% during AI use | Reduced neural activity = cognitive dulling. | Reduction reflects efficiency, not decline. Experts in any field show less activation as skills become streamlined. ChatGPT reduces low-level linguistic strain, freeing higher-order reasoning. |
| AI essays described as “robotic” or “soulless” | Evidence that AI output degrades depth and creativity. | Reflects stylistic bias. Eloquence (fluency) ≠ intelligence. “Soullessness” highlights how systems overvalue rhetorical sheen and undervalue substantive reasoning. |
| Even after halting AI use, under-engagement persisted | Long-term dulling effect; AI makes people dependent. | Misreads neurotype dynamics. Once the brain reorganises expression pathways, retracing to old modes takes time. This is restructuring, not damage. |
| Highest performance from those who started without AI and later added it | Proof that AI only helps after cognitive foundations are built. | Confirms psychextric sequencing principle: self-expression must be encoded first (oral/written style), then AI augments it. This is augmentation, not dulling. |
| AI boosts speed (60%) but reduces mental effort (32%) | Speed = gain; effort = loss; paradox of productivity vs. learning. | No paradox. Cognition evolved to reduce unnecessary strain. AI redistributes load: less effort on syntax, more capacity for reasoning and imagination. |
| Overall takeaway: AI offloads thinking and reduces engagement | Cautionary: use AI sparingly or risk dullness. | Correction: AI is a harmoniser of fragmented neurotypes. Risk lies only in premature overuse (before self-expression encodes). Potential = giving all minds an equal voice. |
5. Toward a Psychextrics Framework of Augmented Expression
The MIT study’s most fundamental flaw is treating cognition as a uniform process. Psychextrics shows that human expression is non-unison: speech, writing, and memory encoding occupy different spectra, each governed by overlapping but distinct thalamic nuclei.
- Speech spectrum: fast, automatic, “tap-like” flow mediated by ventral nuclei.
- Writing spectrum: slow, deliberative, line-by-line encoding via medial dorsal nuclei.
- Memory encoding spectrum: foundational scaffolding modulated by anterior nuclei.
AI does not replace these processes but mediates their gaps, harmonising literal and abstract thinkers, slow writers and fast speakers, fragmented and fluent neurotypes. The supposed “dulling” effect is, in fact, the brain restructuring where cognition happens when external supports are introduced.
6. Why This Matters for AI’s Future
The real danger lies not in “weakened brains” but in misframing human intelligence through outdated metrics. Standardised tests, brain-scan connectivity scores, and stylistic evaluations privilege certain neurotypes while suppressing others. By declaring ChatGPT “dulling,” MIT risks reinforcing the very inequalities of expression that psychextrics seeks to redress.
A psychextrics view suggests a different conclusion:
- AI use is not a degradation of intelligence but a redistribution of cognitive load.
- The risk is not dullness, but dependency without scaffolding—using AI before self-expression is encoded.
- The promise is harmonisation—equalising expressive capacity across neurotypes, giving the “equal voice” that conventional systems distort.
Conclusion: Rethinking the MIT Findings
The MIT study reveals less about ChatGPT and more about the limits of current cognitive science. To treat reduced memory recall or neural connectivity as decline is to mistake form for function. Psychextrics argues instead that AI, when integrated wisely, expands the range of human expression by mediating the fragmented non-unison of perception and expression across neurotypes.
The real future lies not in warning that ChatGPT makes us forgetful, but in recognising that forgetting verbatim lines while retaining conceptual scaffolds is how intelligence evolves. ChatGPT is not dulling human minds—it is rebalancing them toward collective intelligibility, the very condition of human progress.
Reference
Kos’myna, 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. Retrieved from: https://arxiv.org/abs/2506.08872
MIT Study
- Title: Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task
Authors: Nataliya Kos’myna; Eugene Hauptmann; Ye Tong Yuan; Jessica Situ (MIT Media Lab & collaborators).
Sample: 54 participants aged 18-39, divided into three groups (LLM / ChatGPT-assisted writing; Search Engine aided; Brain-only with no tool). Retrieved from: https://www.boston.com/news/local-news/2025/06/20/new-mit-study-brings-potential-downsides-of-chatgpt-use-to-light/ - Method: 54 participants (for first 3 sessions), 18 participated in session 4; writing essays under three conditions (ChatGPT/LLM, Search Engine help, Brain-only) over 4 months. EEG used to measure brain connectivity.
Methodology: EEG (electroencephalography) to measure neural connectivity during essay writing tasks over four sessions spanning approximately four months. In the fourth session, groups switched tool-usage (LLM → Brain only; Brain only → LLM) to assess carry-over effects. Retrieved from: https://www.media.mit.edu/projects/your-brain-on-chatgpt/overview/
Key Findings:
- Brain-only group exhibited strongest, most distributed neural connectivity.
- LLM (ChatGPT) users showed weaker connectivity, lower memory retention (quoting their own writing shortly after writing), reduced sense of ownership over their essays.
- When participants originally writing without AI then used AI (Brain-to-LLM), they showed better performance and neural re-engagement vs. those who used AI from the outset.
Back to: 👇