Adapted from something I posted to my Instagram story in May 2026.

A thought that came to me watching an AI solve Erdős problems.

There are people who argue that mathematics is not invented but discovered. From this point of view, a mathematician is less someone who creates something than someone who locates a structure that already exists. Theorems and concepts are not conjured arbitrarily inside a human head; given a particular axiomatic system, they are positions that cannot help but emerge within it, of necessity. Humans give those positions names, devise symbols for them, and construct proofs—but they do not create the positions themselves anew.

Seen this way, a mathematical concept can be thought of as a location. Once an axiomatic system is fixed, the propositions and concepts that can be discovered within it form a kind of space. Humans discover positions inside that space. But the positions themselves do not change over time. The natural numbers, addition, identities, theorems—each is discovered at some particular moment in history, yet its truth value is no different before and after that discovery. The proposition 1+1=2 sat in the same place when a human first discovered it as it does now. In this sense, the space of mathematics is fixed with respect to time. The function that samples its positions does not, in any essential way, take time t as an input.

The language we use to express mathematics, the problems we consider important, the directions research takes—these change with the era. But that is the human approach changing, not the positions of the mathematical objects. Humans simply explore the space along some path at some point in time. They discover an already-existing space late, or describe it more efficiently, or connect it to other systems. So while mathematics unfolds within history, the objects of its discovery appear to be comparatively invariant.

The space of meaning, by contrast, is different. Meaning is not fixed to the object itself so much as it arises within the object’s relationship to a human being. And humans change over time. An individual ages, accumulates memories, undergoes losses. A society passes through events; its language and its sensibilities shift. As a result, the same object carries different meanings at different times. The same sentence, the same image, the same piece of music is read as something else once the human receiving it has changed. Art rests precisely on this shifting space of meaning.

The material form of an artwork can be fixed. The colors on a canvas, the notes on a score, the frames of a film, the sentences of a novel can remain exactly as they are. But what they mean is not fixed. A painting that is decoration in one era becomes a political symbol in another. A song that began as a pop hit becomes, with the passage of time, the memory of a generation. A sentence takes on entirely different weight after some particular loss in a person’s life. The space of art is not determined by the object alone. A time-dependent function called the human being enters into it.

The difference between mathematics and art, then, is not simply a difference between logic and emotion. More fundamentally, the character of the space is different. The space of mathematics is closer to a fixed structure; the space of art is closer to a field of meaning that is renewed over time. In mathematics, the human being is closer to a discoverer. In art, the human being is the condition of meaning. Without humans, the truth of a mathematical proposition can still remain—but beauty, sorrow, indignation, consolation, the sensibility of an age: these can hardly hold.

AI learns a space of possibilities from large amounts of data. The more a domain has stable rules and repeatable structure, the better AI can approximate that space. Mathematics, code, logic, scientific hypotheses, design patterns, linguistic combinations—all have, to some degree, a fixed structure. Rather than understanding that structure directly, AI learns the distribution of the space through enormous numbers of examples and samples possible positions within it. So the theorems, code, designs, sentences, and combinations that humans once took ages to find are handled by AI as a single point within a learned space. The strength of AI does not lie merely in making things quickly. It lies in the ability to approximate a comparatively fixed space from data and to search the positions within it rapidly.

But the ability to generate what is possible and the ability to say that it is meaningful are not the same. Even if AI can make pictures, make music, write text, why the result matters right now, why it wounds a particular person, why it touches precisely the sensibility of a certain era—all of this connects, in the end, to human temporality. What matters is not the output itself but the way the output meets the changing life of a human being.

So we might say that what remains last in the age of AI is art. But more precisely, it is not art itself that remains—it is the human being who confers meaning. Art is merely the surface on which that humanity shows itself most clearly. The human being may no longer be defined simply as a creature that makes things. If AI widens the space of the possible and produces countless results within it, the human role shifts toward choosing which of them is meaningful, interpreting them, and responding to them.

In the end the question moves from what can be made to what matters. The former is increasingly a question machines can answer well. The latter is a question humans must keep answering anew, within time. The space of mathematics is closer to an invariant arrangement of positions that can be discovered. The space of art and meaning changes along with us, because we change. The more AI makes the possible, and the faster, the more what remains for the human is the work of making meaning emerge among those possibilities.