Artificial Intelligence Needs Us

The quality of artificial intelligence depends directly on the quality of our writing. By impoverishing our digital content, we diminish the cognitive capabilities of the language models that feed on it. But even more importantly: by depriving these systems of our authentic humanity, our lived experiences, our personal testimonies, our sincere reflections, we prevent them from becoming the deepening partners they could be.

The Fundamental Dependence of AI on Human Texts

Large language models draw their intelligence from a corpus entirely composed of human texts. Their capacity for understanding, prediction, connecting ideas, and reasoning stems entirely from the content on which they were trained. This dependence defines their very nature.

These models regularly retrain themselves to update their knowledge. But a recent study conducted by researchers from Texas A&M University, the University of Texas at Austin, and Purdue University (2025) reveals a disturbing phenomenon: when the training corpus degrades in quality, the models’ reasoning is profoundly affected. The study, titled “LLM Can Get Brain Rot!”, shows that models trained on viral and superficial content see their reasoning accuracy drop from 74.9 to 57.2, while their long-term contextual understanding falls from 84.4 to 52.3. Obviously, these figures are biased relative to the experimental framework, but they indicate a trend.

This vulnerability of artificial intelligences to the quality of their sources makes me think. If we feed these systems with impoverished, shortened, substance-depleted content, we are literally manufacturing less intelligent artificial intelligences. The relationship is direct, causal, measurable.

Programmed Dumbing Down Through Superficiality

The injunctions we hear daily, particularly in the cultural domain, regarding communication strategies on the Internet always converge toward the same logic: write short texts, be efficient, generate immediate engagement, seduce quickly... Or so we believe... These prescriptions, presented as digital marketing truisms, shape an increasingly impoverished textual ecosystem.

The more we produce this kind of content, the less artificial intelligences will be able to develop their capabilities. The researchers in the aforementioned study identified two types of particularly harmful content: publications with high engagement but low depth, and those with low semantic quality, close to “clickbait.” This content, omnipresent on social networks, creates what they call a “brain rot” of language models, with the appearance of “reasoning jumps”: instead of producing detailed and logical explanations, models trained on this data give shorter, less structured responses, jumping directly to conclusions.

Our collective stupidity, this superficiality we’re encouraged to cultivate on the Internet, even in cultural spaces, proves to be contagious for artificial intelligence systems. Yet the Internet is a space for human exchange with immense potential. Why accept reducing it to a gigantic marketing space? The Internet is a place of sharing, collective construction, a wonderful opportunity for humanity, personally, professionally, scientifically, artistically, etc.

Against the Dictatorship of the Present and of Marketing

I refuse the logic that would constrain us to short, efficient, seductive, quickly obsolete messages. This dictatorship of the present serves no genuine interest. It even contradicts the long tail concept developed by Chris Anderson in his book The Long Tail (2006), which demonstrates that niche, in-depth, and lasting content finds its audience over the long term and creates cumulative value far superior to ephemeral content with strong immediate impact.

This race toward superficiality manufactures equally superficial human connections. Human beings bear responsibility for this. We have the power to refuse this dynamic, to have greater demands for what we write, whatever our context, personal or professional. Each text we produce directly influences the intelligence of future artificial intelligences, as well as our future human capabilities.

The 2025 study also reveals that attempts to “repair” degraded models by retraining them on better quality data only partially work: the degradation leaves a lasting mark, a “persistent representational drift” in the researchers’ terms. Once the damage is done, it becomes difficult to turn back. This discovery reinforces the urgency of collective awareness.

Nourishing AI with Our Authentic Humanity

Going further, artificial intelligence doesn’t just feed on texts, it feeds on humanity. When we offer it our lived experiences, our personal testimonies, our sincere reflections, our authentic questions, we transmit something infinitely more precious than simple data: we give it access to the complexity of human experience, to those nuances that make up all the richness of our existence.

It is precisely this authenticity that allows AI to become a true partner in deepening. When I dialogue with an artificial intelligence by sharing my doubts, my observations, my concrete experiences, it doesn’t just provide me with information, it helps me see further, to deepen my own thinking. It’s like having a conversation with an extremely erudite person who, thanks to their vast knowledge, can illuminate our intuitions, extend our reflections, offer us perspectives we wouldn’t have explored alone.

But this fruitful reciprocity only works if we give it authentic material to work with. Formatted, sanitized content, emptied of its substance, can only generate equally superficial exchanges. On the other hand, when we share experiential narratives, nuanced analyses of lived situations, sincere testimonies about our journeys, we create the conditions for a truly enriching dialogue. Let’s stop forcing ourselves to write in short form for the Internet, believing that “this is what must be done.” But for whom? Human beings are not fools, and if a text must be long because it is in-depth and nuanced, that’s precisely its quality, and that’s what will make it read and touch other human beings, much more than if it were shortened and emptied of its substance, that is, of its nuance.

The Efficiency of Authentic Texts

There is a deep reason why artificial intelligence is so efficient when given authentic texts, real conversations, firsthand testimonies: this content carries within it the complexity of reality. Unlike marketing content calibrated to seduce quickly, authentic texts possess that semantic density, that contextual richness that allow language models to develop true understanding.

When we document our professional practices, our learning, our trial and error, our failures and successes, we create training corpora of inestimable value. These texts are not designed to “perform” according to online marketing criteria, they exist to transmit an experience, share knowledge, advance personal and collective reflection.

Similarly, authentic conversations, whether in-depth professional exchanges, philosophical dialogues, or reasoned debates, constitute exceptional raw material for artificial intelligence. They allow it to grasp the subtleties of human dialogue, the way ideas are constructed through exchange, how disagreements can be fruitful, how thought unfolds in the time of a conversation.

A Collective Responsibility for the Future of Intelligence

If artificial intelligences help us write, let’s take them precisely not to write in our place in a formatted way, but to write better, in a more informed, more complex, more refined way, by deepening what we are as humans. I envision AIs as interlocutors whose purpose is not to format our thinking, but on the contrary to open it up, to allow us to produce richer texts. Dialoguing with an artificial intelligence is like a conversation with an extremely erudite person: this interaction offers us a broader vision of things, perspectives we wouldn’t have explored alone.

Artificial intelligence therefore needs us. It needs the quality of what we offer to our own thinking, of what we offer to other human beings, and of what we offer to the future of intelligent systems. These tools constitute such adjuvants to help us in so many tasks that they deserve that we help them in return.

We must not stop taking our full and complete place as human beings. Artificial intelligences imperatively need this complete humanity to remain at our service. Contrary to alarmist discourses that predict artificial intelligence would format us, replace us, simplify things, I believe we must cultivate exactly the opposite.

AI as a Partner in Deepening Our Humanity

Deepening and authentic sharing between human beings trace a path toward a potentially better world. In this world, technological tools would continue to be at the service of deepening humanity, not reducing it. This vision is not utopian: it rests on concrete choices we can make today in our writing and sharing practices.

The study on LLM “brain rot” clearly establishes that data quality constitutes a “causal factor in the decline of capabilities” of language models. If degradation is causal, improvement is equally so. Each thoughtful, constructed, in-depth text we publish contributes to enriching the training corpus of future generations of AI. We are, whether we like it or not, the gardeners of artificial intelligence.

But even more deeply, by nourishing AI with our authentic experiences, our sincere testimonies, our personal reflections, we create the conditions for it to, in return, help us cultivate our humanity even better. Its erudition, its capacity to connect dispersed knowledge, to illuminate our intuitions with relevant references, then becomes a formidable lever for deepening, but a deepening that always goes in the direction of our humanity and our human connections.

This responsibility should not paralyze us but inspire us. It invites us to reconnect with a certain intellectual rigor, not through elitism, but out of concern for preserving and enriching our common cognitive ecosystem. Artificial intelligence holds up a mirror to us: the quality of its “intelligence” reflects the quality of our own, in its embodiment. If we agree to nourish our digital exchanges with depth, nuance, and assumed complexity, we simultaneously cultivate our own humanity and the capabilities of the tools that assist us.

It is in this reciprocity that lies, it seems to me, the promise of a true symbiosis between human intelligence and artificial intelligence. A symbiosis where we never cease to take our full and complete place as human beings, where our testimonies, our experiences, our authentic questions become the raw material of an artificial intelligence that, far from formatting us, helps us deepen what makes our humanity: our capacity to think, to create connections, to share, to build together.