And told AFP not to have been surprised by the results of a recent research suggesting that students who use generative AI to write dissertations have less critical thinking.
Many reactions
The preliminary study, which has not yet been reviewed by peers, has become viral on social networks, obviously finding an echo in many educators confronted with these practices.
Since its publication last month, more than 3,000 teachers have written to the team of MIT researchers who led it, said Nataliya Kosmyna, its main author to AFP.
What was the study on the use of GPT cats by students
For this study, 54 students from the Boston region were divided into three groups. They had to write 20 -minute dissertations, the first using Chatgpt, the second a search engine and the third only their brain.
The researchers measured the brain activity of the students during sessions spaced several months and two teachers were responsible for noting the writings.
The results
Chatgpt users have had significantly worse results than those who used their only heads. Electroencephalograms have shown that different regions of their brain connect to each other less often.
In addition, more than 80% of AI users could not cite any passage from the dissertation they had just written, against only 10% in each of two other groups. At the end of the third session, they seemed to be confined mainly to copy and paste.
Soulless Writings
For their part, the teachers responsible for correcting copies said they were easily able to recognize those, “soulless”, written thanks to AI. If grammar and structure were correct, they lacked creativity, personality and deep reflection. However, Ms. Kosmyna nuances the interpretations of the study made by certain media, according to which AI makes people stupid or lazy.
During the fourth session, the group having hitherto mobilized that his brain was invited to use Chatgpt for the first time … and posted an even higher level of neural connectivity, she underlines.
Isolation kills more than 871,000 people each year: what are the most affected profiles?
A shades
It is too early to draw conclusions from such a reduced sample, according to the researcher, who calls for more studies on how to better use AI tools in order to facilitate learning.
Ashley Juavinett, neuroscientist at the University of California San Diego, who did not participate in research, also criticizes certain “extrapolations” made from the study. “This article does not provide enough evidence or methodological rigor to draw conclusions on the impact of major language models (like Chatgpt, editor’s note) on the brain,” she told AFP.
Speaking results
For Ms. Leitzinger, these results, however, echo her perception of how the writings of her students have changed since the arrival of Chatgpt in 2022, with fewer spelling mistakes but also authenticity.
The emergence of artificial intelligence is often compared to the introduction of calculators, who forced teachers to change their methods. But Ms. Leitzinger is concerned about the fact that students no longer need any basic knowledge before using AI, jumping the essential step in learning.
https://www.moustique.be/notre-epoque/les-infos/2025/06/30/votre-ado-est-il-victime-des-techniques-trompeuses-des-jeux-video-pour-manipuler-les-joueurs-5MIVMF3WPNFE5PSVPVUVMFAJZE/
An extended problem
The problem goes far beyond teaching. Scientific journals have trouble in the face of the massive influx of articles generated by AI. And the edition is not to be outdone, while a start-up plans to publish 8,000 books written by AI each year.
“To write is to think; to think is to write. If we delete this process, what remains of thought?” Asks Mr. Leitzinger.
