With the dizzying proliferation of artificial intelligence tools such as ChatGPT, the way we think, learn and solve problems is undergoing a silent transformation — and, according to several experts, not always for the better.

From asking for help structuring essays to using AI to polish cover letters, the everyday use of these technologies is becoming common practice, especially among students and professionals. Yet concern is growing: to what extent is delegating these tasks to AI compromising human cognitive ability?

A study published by the Massachusetts Institute of Technology (MIT) earlier this year shed light on this unease. The researchers observed that, when using ChatGPT to write essays, participants showed less activity in the brain regions associated with cognitive processing. They also had more difficulty recalling the content they had produced, compared with those who wrote without the help of AI.

With 54 participants from universities near MIT, the research used electroencephalograms (EEG) to monitor brain activity. The students used AI to summarise questions, locate sources, adjust grammar and even generate ideas. However, many admitted that the tool was not effective at that last task — generating original ideas.

In parallel, another study, led by Carnegie Mellon University in collaboration with Microsoft (which operates Copilot), analysed the impact of AI on the reasoning of 319 office workers. Assessing 900 interactions with AI tools — from data analysis to compliance checks — the researchers concluded that the greater the confidence in AI, the less critical effort was invested in the tasks. The warning is clear: excessive trust can inhibit deep engagement with work and, in the long term, compromise autonomy in problem-solving.

In the United Kingdom, a survey by Oxford University Press (OUP) revealed that six in ten pupils believe AI has had a negative impact on their school skills. However, the picture is complex: nine in ten said the technology had helped them develop at least one relevant skill — be it creativity, problem-solving or revision.

"It's a quite nuanced picture", observes Dr Alexandra Tomescu, a generative AI specialist at OUP. In her view, the problem lies not only in the use itself, but in the lack of clear guidance on how to use these tools in a pedagogical and effective way.

Professor Wayne Holmes, of University College London, takes a more cautious stance. He considers the current volume of independent research on the impact of AI in education insufficient. "There is no robust evidence, at scale, that these tools are effective, safe or beneficial for learning", he says.

Holmes also warns of the phenomenon of "cognitive atrophy", observed in other fields such as medicine. A study by Harvard Medical School indicated that, although the use of AI improved the performance of some radiologists in interpreting X-rays, it inexplicably harmed that of others.

The main concern, he stresses, is that students may achieve better grades with the help of AI, but without consolidating the knowledge they need. "The results are better, but the learning is worse", he sums up.

On the industry side, Jayna Devani, head of international education at OpenAI, acknowledges the dilemma. "We don't think students should use ChatGPT to outsource their work", she insists. In her view, AI should be seen as a tutor, not a shortcut. An example? A student interacting with the tool in study mode, exploring a complex question in a guided way, rather than simply being handed a ready-made answer.

Even so, Devani admits that the debate is far from over. And Holmes adds: "I never tell my students not to use AI... But I insist that they understand how it works, how data is handled, and that they always check the results."

The metaphor is revealing: AI is not merely the new calculator — it is a technology with profound cognitive, social and ethical implications. And like any powerful tool, its impact will depend on how we use it.