In the world of mathematics, something is beginning to quiet shift.

AI is approaching unsolved, difficult problems—in some cases, providing portions of their proofs—while human mathematicians step in to verify them. Just a short while ago, this would have been the setup for science fiction. Or perhaps a joke shared among researchers. Yet, what is unfolding now is slightly more real, and slightly more unsettling.

AI accelerates computations. AI reads academic papers. AI proposes candidate theorems. AI assembles lemmas. AI fills the gaps in proofs. And before we realize it, humans are no longer positioned as the active subjects of proof-making, but as the passive recipients of what has been proven.

Of course, not all of mathematics will be replaced by AI just yet. Mathematics is not mere computation, and a proof is not a mere procedure. How a question is formulated, the aesthetic beauty of a concept, what is deemed significant, which problems to stake one's life on—these are entered into by the intuition, history, and culture of embodied humans. Even so, the boundary is moving.

Until now, mathematicians oriented themselves toward "that which had not yet been solved by anyone." There was an atmosphere there, akin to a final fortress of human intellect. An unsolved problem was not a mere question; it was a territory humans had not yet reached. It was a mountain, an ocean, a darkness, and an object of prayer.

Into that darkness, AI is beginning to shine a light.

The real question here is not whether AI is smarter than humans. It is not even whether AI is capable of proof. Rather, the question is: how will humans choose to believe the proofs that AI produces?

A proof, by its very nature, is a form of trust.

Mathematical proof does not rely on believing in someone's authority. If you follow the steps, anyone can arrive at a point of comprehension. This is precisely why mathematics is resilient. Unlike politics, religion, or markets, the issue is not who said it, but what has been demonstrated.

However, when the proofs generated by AI become so vast, highly complex, and formal that humans cannot easily trace them, can we truly say that those proofs have been "understood"?

A computer verified it. A formal verification system cleared it. AIs confirmed it with one another. A human team read a portion of it.

Is that a proof? Or is it a new institution for believing in something that resembles a proof?

Here lies the essential dilemma of the AI era. AI provides the answer. However, humans must bestow meaning upon that answer. AI generates the proof. However, humans must determine why that proof matters. AI expands possibilities. However, humans must choose which possibilities to anchor within society.

After the theorem, human beings remain.

This does not mean humans will ultimately win. Nor is it a comforting phrase to reassure ourselves that there is a unique humanity that AI cannot replicate. In fact, it is quite the opposite.

Only after AI becomes capable of doing so much does the true work of humans lay exposed. What humans were actually doing was never limited to solving. It was questioning. It was waiting. It was doubting. It was connecting meaning. It was deciding how to handle that discovery within a community.

The role of the mathematician is not disappearing. It is shifting.

From "the prover" to "the chooser of questions." From "the calculator" to "the reader of meaning." From "the provider of correctness" to "the designer of systems that believe in correctness."

This is not a conversation limited to mathematics. The same shift is occurring in medicine, law, education, and management. AI diagnoses. AI reads contracts. AI creates educational materials. AI forecasts markets. AI proposes strategies.

At that point, do humans become mere entities that verify whether "the AI's answer is correct"? Or do we continue to be the ones who judge whether "it is acceptable to adopt that answer"? This distinction is profound.

The narrative that AI will steal our jobs is easy to grasp. Employment shrinks. Occupations shift. Compensation structures adapt. Organizations restructure. These are certainly realistic concerns. Yet, at a much deeper level, "the right to declare something understood" is migrating.

Who says this is correct? Who says this is important? Who says this is acceptable to integrate into society? Who says this should still be put on hold?

Trust in the AI era is not determined solely by the performance of a model. It is not decided merely by benchmark figures. Least of all is it decided by the announcements of giant corporations. What is required are the institutions, the manners, the observations, and the records placed around the answer.

Which AI generated which proof under what conditions? Who verified it? Which parts were understood by humans, and which parts were structurally confirmed by machine? Whom do the results affect? Does the freedom to not adopt them still remain?

While such questions may seem external to mathematics, they actually concern the very future of mathematics itself. Mathematics appears to complete itself within a closed system. Yet, it is humans who believe in mathematics. It is humans who teach mathematics. It is humans who apply mathematics to institutions. It is humans who redesign the world through mathematics.

Therefore, what is needed in an era where AI begins to prove is not a declaration of human defeat. It is a redefinition of the human role.

Perhaps humans no longer need to solve everything by hand. However, humans must not lose track of what to ask. Perhaps humans will no longer be able to trace every proof line by line. However, humans must not outsource what they choose to believe to others. AI might reach a destination in a matter of seconds that would take a human a lifetime to achieve. However, it is still up to humans to give that place a name.

After the theorem, human beings remain.

The humans remaining there are not all-powerful intellects. Rather, they are imperfect, hesitant, doubting entities who sometimes halt, unable to comprehend. Yet, I believe that very act of halting is what is important.

AI has produced a proof. What, then, does it mean? To whom should it be communicated? How should it be taught? What kind of world will be constructed as a result? And is that world truly desired?

This inquiry cannot be fully undertaken by AI.

AI may be capable of proving. However, it is humans who must live within the proven world.

Thus, the mathematical singularity is not the end of humanity. I believe it is the moment when humans migrate from being "solving entities" to "entities who take responsibility for meaning."

© SHIRO & Co.

First published: 2026-06-18