ChatGPT Is Dumbing Us Down

Never has our relationship with intellectual effort been so disrupted that since 2022, when ChatGPT arrived on the market, bringing in its wake the forced opening of the AI sector to the general public. LLMs (Large Language Model) are powerful tools, able to relieve us of essential cognitive tasks, constituting the very exercise of thought.

There’s no denying that AI saves us time and can make us more productive, but no effort is delegated for free. The more efficient a tool is, the more tempting it becomes to give it an ever greater share of our efforts, even more so when we realize that they are no longer necessarily required. to obtain acceptable results. A societal subject that is becoming increasingly hot, that the MIT Media Lab wanted to explore.

On June 10, 2025, the research laboratory shared a 206-page study on the arXiv platform, summarizing four months of observation work. Four months during which researchers measured the impact of using ChatGPT in an academic context to understand what intellectual effort becomes when it is possible to do without it. Unsurprisingly, the bill is already too higheven though we are only at the beginning of the use of this technology.

ChatGPT: an intellectual shortcut that costs us dearly

The researchers thus invited 54 participants to participate in the experiment, students and young professionals, whom they divided into three different groups in order to have them produce dissertations. The first group was instructed to use only ChatGPT as their sole source of information and writing assistance. The second had access to the internet via a search engine, and the third was deprived of any digital tools. While they were at work, their brain activity was recorded by electroencephalography (EEG)in order to measure the effort they were making.

Participants who worked without digital assistance are those in whom brain activity appears the densest. Their brain remains constantly challenged, forced to maintain attention over time and organize their ideas: it may not be the most comfortable, but the exercise was much more educational for them.

Among those who used the search engine, the effort dropped a notch. If the tool could relieve them in certain stages, he left intact the intellectual effort required for the work that the researchers asked them. This is normal: they still had to understand what they were reading, make choices, rephrase their sentences and integrate this information into their writing. The brain remains active, but it is simply less used than the first group, without this becoming problematic.

EEG analysis data from the group using only ChatGPT showed a decrease in neuronal activity related to attention and cognitive effort. Not because of a deficiency in these functions, but because the use of the chatbot allowed them not to request them. The brain learns that an acceptable text can be produced without it being necessary to fully mobilize its resources: it therefore adapts accordingly, reducing cognitive investment to the bare minimum required.

Cognitive debt: the Price OpenAI doesn’t make you pay

This is what researchers called this cognitive : a state in which the reduction of the “ charge cognitive pertinente “, permitted by the use of ChatGPT replaces the ” costly cognitive processes » necessary for learning to give rise to rapid production of dissertations, but far too superficial.

At no point do the researchers call for banning AI or ChatGPT, which would be a misreading of their work. They simply warn of the fact that by making excessive use of it, the tool loses its assistant status and becomes a substituteprofoundly modifying the intellectual functioning of its user.

When we are given such a tool, allowing us to achieve objectives that we ourselves deem admissible while they never required the effort that should underlie them, the value we place on them necessarily declines. We have been accustomed to cultivating our thinking, since it is through it that we apprehend our reality. A process that takes place, most of the time, by making mistakes or doubting, and if this discomfort disappears, we detach learning from any personal appropriation. A thought that we no longer need to confront with doubt will always end up depend on what thinks for him : whether ChatGPT or another model, saving effort can make the error unnecessary and accustom our mind to ease. Everything that can save us from thinking too long is already stamped with the term “ innovation » ; iIt is not certain that we need to add a layer.