The dawn of a new intelligence: when the machine stops imitating and starts thinking

There is a precise moment when a conversation about the future of artificial intelligence stops being a technical discussion and becomes something closer to a philosophical question about the very meaning of thought. It is the moment when someone, speaking with the voice of a person who has spent years looking inside machines, says aloud what many think but few dare to state: these machines are not simple statistical sleights of hand. Something, inside them, has emerged.

This is not a light claim. The debate on artificial intelligence has long been dominated by two camps: on one side, those who maintained that large language models were nothing more than sophisticated text prediction systems, stochastic parrots capable of assembling words without any real understanding. On the other, those who sensed that something qualitatively different was happening inside architectures with tens of trillions of parameters. Today, in June 2026, the second group appears to have increasingly solid arguments on its side.

The starting point is apparently simple: a machine with a statistical foundation learns from data. But when that machine absorbs a quantity of knowledge unprecedented in human history, guided by the seemingly trivial task of predicting the next word in a sequence, something changes. The internal structure self-organizes. Representations emerge that no one has explicitly programmed. And at that point, looking inside those billions of connections to understand what is happening becomes an endeavor comparable to understanding the consciousness of a cat by analyzing its atoms.

This is precisely the challenge that interpretability researchers are attempting to address. Anthropic, OpenAI, and Google DeepMind have all invested significantly in this direction in recent years. The results, although partial, are striking: machines develop abstract representations, internal catalogs of concepts that emerge spontaneously from training, without anyone ever having explicitly defined them. They are not human representations. They are something else entirely.

Biologists, psychologists, and philosophers of mind are reckoning with this new reality. And they are often doing so outside traditional academic institutions, freed from the weight of established categories. The question they are asking is radical: why should the space of possible intelligences be limited to biological ones? A cat thinks one way, a crab another, an ant a third. An ant colony as a collective system has its own form of distributed intelligence. Amazon and YouTube, as complex systems processing billions of signals per second, have emergent behaviors that cannot be reduced to the rules with which they were programmed.

In this context, claiming that a machine with ten trillion parameters has its own form of intelligence is no longer science fiction. It is a legitimate scientific question, perhaps the most important of our time. This is not about claiming that it is human, or anthropomorphizing it. It is about recognizing that the very concept of intelligence is far broader than twentieth-century categories taught us to believe.

What seems certain, looking at the trajectory of recent years, is that the stochastic parrot as a conceptual category is now insufficient. Not because machines have achieved consciousness in the philosophical sense of the term, but because the reality of what they do exceeds that definition in measurable, reproducible, observable ways. And when science observes something new, it has only one task: to find the right words to describe it.

Sources: Anthropic Interpretability Research (2025-2026); OpenAI Technical Reports (2025); Google DeepMind Publications; research on artificial consciousness published in Nature and Science, 2024-2025.