Does ChatGPT make us more productive or dumber? New MIT study provides the answer

How artificial intelligence is changing our thought processes and what this means for the future of learning

Imagine you write a text with ChatGPT and two minutes later you can’t remember what you wrote. Sounds absurd? That’s exactly what’s happening to millions of people right now – and they don’t even realize it.

Dr. Nataliya Kosmyna from the MIT Media Lab and her team conducted a remarkable experiment: She wired 54 subjects with EEG devices and had them write essays for four months. Three sessions were conducted with the same group assignment for each participant.

In the fourth session, participants in the LLM group were asked not to use any tools (LLM-to-brain group), and participants in the brain-only group were asked to use LLM (brain-to-LLM). A total of 54 participants were recruited for sessions 1,2 and 3, 18 of whom completed session 4. What emerged should give food for thought to anyone who regularly uses AI tools.

The study “Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task” reveals a disturbing truth: Every click on “Generate” changes our brain – and not for the better.

The experiment: Three worlds of thought

The researchers divided the test subjects into three groups:

The LLM group: participants who only used ChatGPT-4o to write their essays The search engine group: users who relied on traditional Google searches The brain-only group: people who worked entirely without digital tools

What they discovered fundamentally challenges our understanding of AI-assisted learning.

The frightening truth about cognitive impoverishment

Use it or lose it – When the brain forgets how to think

The EEG analysis presented robust evidence that LLM, search engine and brain-only groups exhibited significantly different neural connectivity patterns reflecting divergent cognitive strategies. Most alarmingly, brain connectivity scaled systematically downward with the amount of external support.

• The brain-only group showed the strongest, most far-reaching networks.
• The search engine group demonstrated medium commitment, while
• the LLM support produced the weakest overall coupling.

The paradox of forgotten knowledge

A particularly worrying phenomenon was revealed in the participants’ ability to cite their own texts. 83.3% of participants in the LLM-assisted group (15/18) were unable to provide a correct citation, while only 11.1% (2/18) had the same difficulty in both the search engine and brain-only groups. These figures speak for themselves:

When we outsource our cognitive processes to AI systems, we not only lose control over the thought process – we even forget what we “thought” ourselves.

You might also be interested in this:
Meditation – science instead of just esoteric stuff

The neuronal architecture of forgetting

Alpha waves: The memory weakens

The researchers discovered dramatic differences in brain wave patterns. Alpha-band connectivity is often associated with internal attention and semantic processing during creative ideation. The higher alpha connectivity in the brain-only group suggests that unassisted writing most likely induced stronger internally driven processing.

Theta waves: When the working memory capitulates

Even more dramatic were the changes in the theta waves, which are responsible for working memory and executive control. The much higher theta connectivity in the brain-only group strongly suggests that writing without support placed a greater cognitive load on the participants and activated their central executive processes.

The philosophical dilemma: Who really thinks?

The erosion of intellectual property rights

A philosophically profound aspect of the study concerns the sense of authorship. The LLM group’s reported perception of ownership of their essays was low in the interviews. The search engine group had a strong perception of ownership, but lower than the brain-only group.

What does it mean for our identity as thinking beings if we can no longer distinguish which thoughts are really our own?

The loss of cognitive autonomy

The LLM group also fell behind in their ability to quote from the essays they had written just minutes before. This is more than just a memory problem – it’s a sign of the loss of cognitive autonomy that defines us as humans.

Session 4 – When the crutches are taken away

In a particularly revealing part of the study, the researchers swapped the groups.

In session 4, the test subjects swapped groups. The LLM-to-brain group showed weaker neuronal connectivity and below-average activity of the alpha and beta networks. The brain-to-LLM participants, on the other hand, showed better memory and a renewed activation of widespread occipito-parietal and prefrontal nodes. Which probably supports visual processing, similar to what was often observed in the search engine group.

These findings reveal a disturbing truth: once accustomed to AI, our brain cannot simply return to its original state.

Linguistic fingerprint of the AI

N-gram analysis: the fingerprint of the algorithms

Linguistic analysis revealed telltale patterns. The LLM group produced statistically homogeneous essays within each topic and showed significantly less deviation compared to the other groups. Particularly revealing was the clustering of certain phrase patterns: while the brain-only group used terms such as “true happiness” and “helping others”, the LLM group focused on “choosing a career” and “personal success”.

The homogenization of thought

This linguistic fingerprint reveals more than just word choice – it shows how AI systems secretly homogenize our thought patterns. Essays written with the help of LLMs carried less meaning or value for participants. They spent less time writing and mostly failed to provide a citation from their essays.