Artificial intelligence has, in the space of just a few years, become a constant presence in both our imagination and our daily lives. It writes, responds, analyzes, generates images, simulates dialogue, and makes decisions. It is fast, tireless, and seemingly creative. And yet, precisely at the moment when its power becomes most evident, a question emerges that is rarely asked in full: does artificial intelligence truly understand what it does? Or, more precisely, is it capable of seeing itself while it acts? My impression is that AI is, by its very nature, an intelligence without a mirror. It can reflect the world, but it cannot reflect itself within it. It can process content, but it cannot perceive the meaning of its own gesture. It can describe, but it cannot inquire.
This distinction is anything but academic. It touches the very heart of the relationship between artificial and human intelligence, especially when we move from the territory of performance to that of meaning. AI is born of humanity, built by humanity, trained on the products of humanity, yet it does not inherit the experience that made those products possible. It learns outcomes, not journeys. It recognizes patterns, but it does not know the weight of the time that generated them. It has no biographical memory, no body, no exposure to risk, no loss. And without loss, without friction, without lived error, something is missing that, for human beings, is constitutive: the awareness of one’s own action.
As Hannah Arendt reminded us, thinking does not coincide with knowing. Knowing means accumulating data, recognizing relationships, solving problems. Thinking, instead, means questioning meaning, dwelling in the question, accepting ambiguity. Artificial intelligence knows with extraordinary efficiency, but it does not think in this sense. Not because it is “less evolved,” but because it is not situated. It is not in the world as experience, but above the world as function.
In the first article of this column, I argued that so-called soft skills still represent a domain of human supremacy over AI. Today I would say that this supremacy does not depend on the technical complexity of these competencies, but on their nature. Empathy, intuition, leadership, the ability to read what is left unsaid are not isolated skills that can be transferred or trained like a model. They are traces of life. They are the result of mistakes, conflicts, silences, wrong decisions, and waiting. As Michael Polanyi observed, we always know more than we can tell. Artificial intelligence, by contrast, can say a great deal, but it does not know more than what it says. Not because it is deficient, but because it does not live.
This difference becomes even more evident when we approach the theme of imagination. In public debate, imagination is often confused with the ability to produce something new, original, or unprecedented. But human imagination is not simply the production of novelty. It is, first and foremost, deviation. It is an act that breaks a continuity, interrupts a given trajectory, and introduces something that previously not only did not exist, but often was not even desirable. Artificial intelligence works by extension and recombination. It starts from what exists and reorganizes it according to rules of coherence and probability. Human imagination, instead, often arises from a fracture, from a discomfort, from a restlessness that finds no solution within already traced paths.
In this sense, the thought of Gaston Bachelard is illuminating, when he states that imagination is not the faculty of forming images, but of deforming them. To deform means to accept losing balance, to renounce symmetry, to expose oneself to error without knowing in advance what the outcome will be. It is a gesture that does not aim at optimization, but at truth, even when that truth is uncomfortable, incomplete, or disturbing. Here AI encounters a structural limit: not so much an inability to produce surprising images or texts, but an impossibility of knowing when a deviation should be preserved and when, instead, it should be corrected.
Anomaly, in fact, is the point of fracture between artificial and human intelligence. For an artificial system, anomaly is a statistical deviation, an outlier, a noise to be reduced. It is something that lowers the efficiency of the model and therefore must be normalized. For human beings, by contrast, anomaly is often the place where meaning is born. It is the gesture out of measure, the non-optimal choice, the discordant word that suddenly illuminates a discourse. Much of human innovation does not arise from optimization, but from disobedience — from the decision to do something that, according to all rational parameters, should not have worked.
It is no coincidence that Nassim Nicholas Taleb has built much of his thinking around the idea of the unpredictable event, the “black swan,” that which escapes models precisely because it does not fit within known distributions. Artificial intelligence is designed to reduce the unforeseen. Human beings, at times, choose to pass through it. And they do so not out of inefficiency, but out of existential necessity.
It is in painting that this difference appears in an almost tangible way. In painting, more than in other languages, form is not a container for content: it is the content itself. There is no separation between the “how” and the “what.” In Caravaggio, shadow is not a lack of light, but a moral declaration. In Jackson Pollock, the gesture does not represent something: it is itself the meaning. In Mark Rothko, color does not refer to an object, but becomes a direct, almost physical experience. In Pablo Picasso, deformation is not an error of perspective, but a denunciation, a political act, a cry.
An artificial intelligence can describe these works, classify them, place them within an artistic movement, recognize their style. But it cannot know why the anomaly should not be corrected. It cannot feel that, in that darkness, in that disproportion, in that void, meaning resides. As Paul Klee affirmed, art does not reproduce the visible; it makes visible. And to make visible does not mean to explain, but to expose, to show, to leave open.
Here the theme of the mirror returns. Artificial intelligence does not see itself because it has no embodied point of view. It has no existential “from where.” It cannot perceive its own gesture as a gesture, because it is not situated in a history, in a biography, in a body. As Maurice Merleau-Ponty writes, the body is not an object in the world, but our way of being in the world. AI is in the world as a tool. Human beings are in the world as a presence. This difference is not quantitative; it is not destined to be closed by more data or more computing power. It is a difference in nature.
For this reason, in my view, the comparison between artificial and human intelligence is often framed incorrectly. It is not about determining whether AI will surpass us in this or that task, nor about nostalgically defending a human space threatened by technology. It is about recognizing that there is a place where meaning is born, and that this place does not coincide with efficiency, coherence, or optimization. It is the place of anomaly recognized as value, of deviation preserved rather than corrected, of imagination understood as an act of rupture.
Artificial intelligence will continue to grow, to surprise us, to radically transform the way we work, communicate, and create. It would be naive to deny it. But it will not inhabit, at least for now, that fragile and decisive place where human beings choose not to be optimal, not to be efficient, not to be aligned. It is there that humanity remains the criterion — not by superiority, but by exposure. As Albert Camus wrote, the struggle itself is enough to fill a man’s heart. As long as this struggle between rule and deviation, between form and meaning, between calculation and imagination exists, there will still be a space that is irreducibly human. And it is perhaps precisely that space that is worth defending, inhabiting, and, above all, understanding.
