Those Who Study with ChatGPT Almost Forget Everything: The Worrying Phenomenon
The MIT has measured with EEG what happens in the mind of someone writing an essay using ChatGPT. The findings are clear: connectivity between brain areas has decreased, attention has waned, and the sense of ownership over the produced text has weakened compared to those who worked, for example, with just a search engine.
The study compared three groups of participants engaged in the same task: writing short essays without any aids, using Google, or relying on ChatGPT. Those who used the chatbot, when asked to write again later without its help, struggled to even remember the ideas they had previously expressed. For researchers, the systematic delegation of cognitive tasks to large language models (the LLMs powering ChatGPT, Claude, Gemini, or Llama) risks undermining long-term learning.
Memory Fades When AI Writes
"To learn, one must confront the effort and frustration of deep understanding," explains Carlo Reverberi, cited by Focus, associate professor of General Psychology at the University of Milan-Bicocca and a member of NeuroMI. Without this step, he adds, knowledge remains fragile and superficial, destined to be forgotten quickly.
The numbers confirm this: in a study at the Federal University of Rio de Janeiro, those who used ChatGPT without limits remembered 57.5% of the content studied, compared to 68.5% of those who abstained. When used as a tutor bound to a knowledge base provided by the instructor, however, an LLM does not invent answers and becomes a resource: this is the case with NotebookLM, which Reverberi provides to his students for summaries, quizzes, and a "Socratic" method, focusing on questions rather than pre-packaged answers.
A study from the University of Pennsylvania tested the effect of ChatGPT on mathematics: those who queried it for solutions improved their grades by 48%, but when tested alone, without the chatbot, their performance plummeted by 17% compared to peers. With a different chatbot configured by instructors to act as a tutor, the improvement rose to 127%. In short, the difference lies not only in usage but in how AI is designed from the outset.
An investigation of 666 people conducted by Michael Gerlich, a sociologist at SBS Swiss Business School, links more intense use of AI to lower critical thinking skills. It is a correlation, not proof of causation, but Gerlich also notes more stagnant discussions in class and students preferring the chatbot to their peers for resolving doubts.
On the professional writing front, research from MIT shows that ChatGPT increases text quality by 18% and cuts writing time by 40%, with the greatest benefit for those starting with weaker skills. Those who were already skilled and begin to delegate everything, however, give up the very excellence that distinguished them.
According to Cristina Becchio, a professor of Neuroscience at the University of Hamburg cited by the source, some next-generation models like GPT-5 already surpass human averages in divergent thinking tasks, but no AI reaches the creative peaks of the top 10% most talented individuals in the population. For this reason, Becchio describes it as a team game rather than competition and cites China's decision to introduce artificial intelligence as a school subject from elementary school as a model to follow, not to fear.