In every major technological transition, what changes is not just production but the very architecture of social power. From the industrial revolution to the digital age, value was transferred from manual labor to machinery and later from matter to information.
Today, with artificial intelligence, we are experiencing an even deeper shift: the transfer of knowledge as a form of power, as Foucault aptly described, to new agents that no longer belong to society but to uncontrolled technological systems. Artificial intelligence is not fueled only by electrical power and innovation algorithms. These are necessary, but not decisive.
The critical “fuel” is one: data. In Bourdieu’s terms, AI is not simply educated with information – it is educated with symbolic and cognitive capital, not data as statistical units, but as living representations of human experience, language, judgment, and creativity. Without them, no technology, no matter how advanced, can “learn.” That’s why, over the past three years, we’ve seen a global race to acquire, acquire, or create data. It’s the largest transfer of value in history, and it’s happening silently: from people to the models that teach them.
With every question, every correction, every evaluation of an answer, millions of people are feeding the evolution of these systems. And they do it without perceiving it as “work,” even though they’re generating value far greater than the value of the time they spend. It’s no coincidence that leading AI companies are expanding their data collection and retention, investing in vast content bases, and—when real content isn’t enough—turning to the creation of synthetic data. Synthetic data, generated by the models themselves, is growing exponentially and already constitutes over 60% of the data used to train new generations of LLMs.
However, there is a critical limit: to what extent can a system train itself without reproducing its mistakes? Synthetic data accelerates evolution, but it also carries a risk: the “self-recycling” of machine knowledge, which gradually moves away from real human experiences. This is the great contradiction of our time: AI needs more and more data, at the same time as the available data is being exhausted.
People, users, content creators are becoming, without realizing it, the most valuable source of production of the future. And for this reason, the value is now transferred from human intelligence to machine learning. Thus, in Habermas’ terms, the public sphere risks being replaced by a technical sphere of algorithmic discourse production, where social experience gives way to computational representations of it. The equation is clear: a model that learns from millions of users is more valuable than a model that relies solely on computational power.
Human feedback is worth far more than any optimization algorithm. It’s no coincidence that big companies are now investing billions not in hardware but in vast human review teams. Humans are still the main enablers of machine learning – but the knowledge they produce no longer belongs to them. AI doesn’t just replace jobs; it reshapes how value is produced, and therefore who controls it.
In the era of platforms, the price we pay for “free” access was the transfer of our personal data. Today, we pay with something much deeper: our participation in shaping machine knowledge. The question is not whether AI will dominate. The question is who will determine the data that shapes it, and who will benefit from this enormous transfer of value. Will it remain concentrated in a few companies, or will it be shared more equitably across the ecosystems that help educate it? Can society claim some of that value? The answer will shape not only the economy of the next decade, but also how we understand knowledge itself. As synthetic data proliferates and machines learn from past versions of themselves, human input becomes even more important.
The future of AI will be determined not by algorithms, but by the quality—and stewardship—of the human data we allow it to use. Perhaps the greatest transfer of value in history will not be economic. It will be cognitive. And the real challenge for societies will not be to protect their data, but to the creative capacity that produces it.
