AI doesn't replace us, it weakens us... if we let it

Last March, during my talk at the Belgium Cyber Security Coalition, I asked a simple but unsettling question:

What if AI didn't make us stupid... but forced us to become smarter?

Six months later, an MIT study proves me right. In black and white. Brain wired up. Measured evidence.

The Brain On LLM project, led by a team at the MIT Media Lab, has just demonstrated what I've been hammering home for months: AI is not neutral. It shapes the way we think, learn, remember and create.

And if we don't use it consciously, it gradually turns us into mere consumers of generated text.

What does MIT's Your Brain on ChatGPT study really show?

The study is fascinating, and its title already sets the tone: "Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task". Cognitive debt. Led by Nataliya Kosmyna and her team at the MIT Media Lab, it followed fifty-four participants, an electroencephalogram on their heads, across four sessions spread over four months. Three groups:

• Those who write on their own (Brain-only),
• Those who use a search engine (Search),
• And those who write with the help of a large language model (LLM such as ChatGPT).

The results are clear:

• The Brain-only group had the most activated brain networks, the richest connections, the sharpest memory.
• The Search users sat in between: taxed, but still cognitively engaged.
• The LLM users showed the lowest brain connectivity. Fewer connections. Less memory. Less sense of ownership over the content.

Worse still: LLM users struggled to quote a single sentence from the essay they had written only minutes earlier. And when, in the fourth session, those who had leaned on AI were asked to pick up the pen on their own, their brains stayed in low gear: weakened connectivity, under-active alpha and beta networks. That is cognitive debt. You don't see it in the moment. You repay it later, with interest.

One caveat, and it matters: this work is a preprint, not yet peer reviewed, on a small sample. It is not definitive proof. It is a signal, and it points in exactly the direction I have been seeing in the field for months.

AI as an intellectual prosthesis: the escalator syndrome

Earlier this year I was explaining that AI is a tool for augmentation, not substitution. But the temptation is strong: why push yourself when you can click? Why think when a machine can churn out an acceptable piece of text?

This is the escalator syndrome: the vast majority of people take the escalator, a minority take the stairs. The escalator means a little comfort right now, a lot of weakness tomorrow. The stairs mean a little pain now, but far more strength tomorrow.

AI is today the biggest cognitive escalator of our time.

And in a world that prizes speed over depth, the result is predictable: we become less attentive, less critical, less creative.

Is AI really destroying critical thinking?

I've said it, and I'll say it again: AI does not destroy critical thinking. We destroy it ourselves, through surrender. The technology is not the problem. Our relationship with effort is.

And here again, recent research confirms the intuition. In early 2025, researchers from Microsoft Research and Carnegie Mellon University surveyed 319 knowledge workers about their daily use of generative AI. Their conclusion fits into a single sentence: the higher the confidence in the tool, the less critical thinking the user applies; the higher the confidence in oneself, the more they apply. AI does not remove intellectual effort, it shifts it, from producing to verifying and integrating what the machine puts forward. Provided, once again, that we take the trouble to verify.

How do you stay intellectually sovereign in the face of AI?

If you want to stay relevant, powerful, strategic in this world transformed by AI, you don't need to flee the technology. You have to frame it with strong human practices:

1. Read: to feed your mind, cross-reference ideas, confront complexity.
2. Write: to structure your thinking, step back, create meaning.
3. Debate: to sharpen your vision, listen, respond, rise higher.

These are training exercises. Disciplines. Like sport for the body, they are the bare minimum for the brain.

AI is a magnifying glass, not a brain

What Brain On LLM shows precisely is that LLMs are not artificial intelligence in the strict sense.

They are amplifiers of our mental routines. They pick up our ways of speaking, our dominant ideas, our biases, and recycle them. Fast. Well. Without a soul.

So no, this is not man or machine.

It's man AND machine, on condition that man stays in control, and that the machine serves to push us further, not to lull us to sleep.

MIT confirms it, but the responsibility is ours

MIT put an electroencephalogram on the question. I had put my intuition, my experience, and my deep conviction on it: AI is our mirror. If you become lazy, it amplifies your laziness. If you stay engaged, it amplifies your power. The choice is always in our hands. The technology is not the problem. Our relationship with effort is.

So ask yourself the question: do you want the escalator... or the stairs?

---

Sources

• Nataliya Kosmyna et al., "Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task", MIT Media Lab, preprint of 10 June 2025 (revised 31 December 2025): arxiv.org.
• Presentation and results of the Brain on LLM project: brainonllm.com.
• Hao-Ping Lee et al., "The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers", Microsoft Research and Carnegie Mellon University, CHI 2025: microsoft.com.
• The author's talk at the Belgium Cyber Security Coalition, March 2025.