From the internet to AI: we risk “cognitive retreat”. An electroencephalogram monitoring has shown reduced brain connectivity and a drop in grades in students who make “careless” use of AI. Here’s how to fix it…
For millennia, every flash of human ingenuity, choice, and physical effort has been driven by the urge to obtain food. Today, on the contrary, we live in one “obesogenic” societywhich makes high-calorie and unhealthy foods all too accessible, while discouraging movement.
What all this has to do with the theme of this article was explained by the English writer and teacher Daisy Christodoulou who maintains that we are replicating the same pattern with information: «In the transition from the Internet era to the era of artificial intelligence, what we consume is not only an ever-increasing amount of low-value and ultra-processed information, but essentially pre-digested information, presented in a way designed to bypass important human functions, such as evaluating, filtering, and synthesizing information.”
Baby food ready
We risk reading, writing and recirculating the intellectual equivalent of ready-made, easy-to-chew, overabundant and poor-quality porridge, which allows us to do the minimum wage from a cognitive point of view. As AI becomes more and more pervasive in our lives and everyone wonders how the world will change, this is the story of how we are changing, using it.
In other words, artificial intelligences are they making us more or less… sapiens? Scientists have begun to measure the effects of this new cognitive regime on our brains. The research is still preliminary and evolving, but the first results are there. And they deserve to be read.
Under review
Recently, an MIT study compared the brain activity of three groups of participants writing short essays: without aids only with Google or ChatGPT as support. The electroencephalogram (the test that records the electrical activity of the brain) revealed that, in the latter group, the connectivity between brain areas appeared reduced, the attention was lower, the sense of authorship of the written topics was weaker.
Trying again without the help of ChatGPT, these people struggled to remember previously touched points. The authors of the work fear that delegating the cognitive tasks that we should be capable of to large language models (or Llms: they are the systems trained on huge amounts of text that power chatbots such as ChatGPT, Claude, Gemini or Llama) could affect learning in the long term.
«To learn, you need to deal with fatigue and frustration of profound understanding”, explains Carlo Reverberi, associate professor of General Psychology at the University of Milan-Bicocca and member of NeuroMI (Milan Center for Neuroscience).
«If these steps are missing, the knowledge is not there or it becomes fragile and superficial, ready to be forgotten».
Using AI at school in a substitute way for months and years produces a complete cognitive withdrawal. “It’s as if what in other generations meant copying from one’s partner was now possible always and continuously.” This cognitive drain can also affect long-term memory: in a study by the Federal University of Rio de Janeiro, those who used ChatGPT without limits remembered 57.5% of the contents compared to 68.5% of their classmates who did not use it.
At the side, not in place
But, used as tutors, AIs become a resource. «There are Llms, like NotebookLM, bound to a knowledge base provided by the teacher”continues Reverberi. «If they don’t have the answer, they don’t invent it, and they don’t suffer from hallucinations (those errors in which the AI invents facts, sources or quotes in a confident tone, ed.). Used as a supplement to study, they can become an extraordinary resource that enriches the university experience.”
Reverberi has made one available to its students. It acts like a tutor, it can prepare summaries, formulate multiple choice questionnaires or open questions for self-evaluation. «You can explicitly ask to have a maieutic attitude, which promotes learning: to ask questions, not immediately provide an extended answer, try to counter the arguments that are offered in a sort of cognitive pungiball».
Report card or knowledge?
The point is to understand what we really care about: good grades immediately or acquire skills? To achieve the second objective, it also matters how AI is designed, not just how it is used. This is demonstrated by a study from the University of Pennsylvania that tested students and mathematics. Those who studied by asking the solutions to ChatGPT had improved their grades by 48% compared to their classmates; but when they had to fend for themselves, without ChatGPT, ratings dropped by 17%.
However, when they used a different chatbot, tutored and trained by teachers, they improved the results by 127%. «These tools are still underused, both by students and teachers», adds Reverberi. «Certainly their pervasiveness is imposing a transformation on universities. Let’s think about degree theses: today it is very difficult to understand how much is the student’s work and how much comes from AI.”
Critical direction down
A study on 666 people of various ages conducted by Michael Gerlich, sociologist at the SBS Swiss Business School in Kloten (Switzerland), showed that those who delegate cognitive tasks to artificial intelligence more often also have lower critical thinking skills.
It’s a correlation, of course: it may be that those who have less critical sense are simply more inclined to trust machines.
But there’s more. In class, Gerlich has noticed that discussions among his students are stagnant, and that students prefer to direct their doubts to the chatbot rather than to their classmates. It doesn’t help that social media algorithms, also driven by AI, take away the effort of choosing what to watch or listen to, serving us short content “with easily digestible messages, which do not encourage critical thinking”, says Gerlich.
Don’t settle for the first answer
Another problem is the anchoring effect: if we passively accept the first response of artificial intelligences, the brain follows that flow without looking for alternatives. «Very few, after the first response, ask the chatbot for a different point of view. Without this effort, even trivial aspects that the machine does not consider can be lost”, clarifies Andrea Lavazza, professor of Moral Philosophy at the Pegaso University and of Neuroethics at the University of Milan, on the board of Sepai (Society for the ethics and politics of AI).
To understand how it works, a domestic example is enough. We ask a chatbot to plan the month’s lunches: the first response will be generic. The truth critical use comes when you add “reduce waste”, “take intolerances into account”, “make it cost less”.
To safeguard critical capabilities, says Lavazza, «it is It is essential to verify what AI offers us. If the end result is something we don’t understand, we are no longer in charge. The risk mainly concerns young people, while for adults it is lower, because they developed cognitive abilities and critical sense before the AI era. We don’t know what kind of ability a fourteen-year-old who studies today relying on generative AI will have in ten years. It is a prediction that, for now, makes no sense.”
The price of convenience
Even among adults, having ready-made answers served makes the palate less refined in the long run. Researchers from Carnegie Mellon (USA) and Microsoft found a decline in critical work engagement and problem-solving skills in over 300 employees who used AI for work at least once a week. «We don’t really know what the AI cuts out when it summarizes and selects», comments Lavazza. «We can train it on our criteria, of course, but it remains different from doing it ourselves.
The point is learn to choose: for quick things you can be satisfied, for those that matter you still need human supervision, slow and detailed”.
Fellow co-pilots
«Even in those who are already trained, the chronic delegation of professional skills already acquired to AI can lead to deskillingthe loss of skills that are no longer trained”, specifies Reverberi. «A doctor who systematically delegated part of his work to artificial intelligence would, in the long run, compromise his own independent judgement. But a general practitioner who doesn’t have time to read twenty new scientific articles for his professional development can upload them to an AI, interrogate it, verify the key points. On the other hand, in the healthcare sector, AI used sensibly improves performance and the accuracy of diagnoses.”
In other cases, men and machines work as a team: «A decision made together by a person and an AI is often better than one that humans would have made alone. In some cases it is even better than both, taken one at a time”, says Reverberi. «It’s called complementarity: the errors of man and machine are different, and the two entities correct each other. But deskilling can destroy this virtuous dynamic: if one of the partners of this team loses competence at a critical level, all that remains is total oblivion to the AI’s suggestion.”
Leveling effect
Especially in creative works, and especially in writing, AI acts on two opposite levels. On the individual, it increases productivity and helps generate more original content. Collectively, it does the opposite: the texts generated by AI are very similar to each other, and the mass use of LLMs is flattening the cultural diversity of written production, a heritage made up of very high peaks and dizzying differences in level.
In this dynamic, those who gain the most are those who know the least: MIT research shows that, among professionals who use ChatGPT, the quality of writing increases by 18% and writing time drops by 40%, with the greatest advantage for those with weaker skills. AI reduces inequalities between job performance; those who lose out are those who are already good and start delegating everything, giving up errors but also excellence.
10% of humanity is unbeatable
«In creativity, results with AI are often superficial», explains Cristina Becchio, professor of Neuroscience at the University of Hamburg (Germany).
“The answers are often predictable, and the creative ideas follow recurring patterns: AIs tend to always follow the same paths.” Some latest generation artificial intelligences, such as GPT-5, can exceed average human performance in some creative tasks, such as those of divergent thinking (the ability to generate original solutions to an open problem).
But no system, not even the best, matches the peaks of creativity of the most imaginative humanity: the gap is particularly evident in the most creative 10% of the population. More than a race, perhaps, it is better to see our relationship with artificial intelligence as a team game.
AIs can process large amounts of information; we know how to combine disparate concepts and think outside the box, like a chef who experiments with unusual combinations and then chooses, after careful tasting, which one to bring to the menu. Sometimes, however, you need to cook quickly. And having plenty of food, even already processed, is better than having to get it every day.
As long as you choose carefully what to bring to the table and consume. “The idea that AI makes us dumber or lazier because we delegate is probably limited,” concludes Becchio. «Intelligence artificial is changing the natural one in a new and surprising way and, for those who deal with cognitive processes, in an extremely interesting way. It cannot be something that people and young people discover in their free time, without guidance: it is an impoverished use of its potential. China has introduced artificial intelligence as a school subject since primary school. I think it’s the right choice.”
