I asked ChatGPT to “explain like I’m 5” re the mathematics kerfuffle I’ve read about but don’t really understand. A long discussion between the human (me) and the AI followed. We got all the way to the question, “What then IS a human?” but by that time I was hungry and went off to find lunch. ChatGPT doesn’t need lunch. Ever. This is the first part of its answer, with citations it supplied.
Now, your question about AI and mathematics is especially interesting, because the excitement is real—but some of the headlines make it sound as though mathematicians have suddenly been rendered obsolete.
They haven’t.
What’s happening in mathematics? The five-year-old’s explanation
Imagine that mathematicians have spent centuries building an enormous castle.
Some rooms are finished. Other rooms contain doors that nobody has managed to open. Above each locked door is a question.
For example:
Is this mathematical statement always true, no matter how large the numbers become?
A mathematician might spend twenty years trying to answer that question. Sometimes the answer is yes. Sometimes someone discovers a counterexample—a single case that proves the statement isn’t always true.
Now imagine an AI system that has studied enormous amounts of mathematics and can try thousands of approaches, connect ideas from distant branches of the subject, and sometimes produce an answer that human researchers haven’t found.
That’s the breakthrough. AI is beginning to do genuinely new mathematical research, not merely calculate faster or repeat things mathematicians already know.
A real example you can understand
In 1946, mathematician Paul Erdős asked a question about dots on a flat surface.
Suppose you scatter a large number of dots across a sheet of paper. How many pairs of dots can be exactly one inch apart?
It sounds like a children’s puzzle. But mathematicians discovered that the answer becomes astonishingly difficult as the number of dots grows.
For about 80 years, they worked on understanding the possibilities.
In May 2026, OpenAI announced that one of its AI models had found a mathematical construction that disproved a longstanding conjecture connected to this problem. It used ideas from an unexpected area of mathematics to do so.
(https://www.google.com/s2/favicons?domain=https://openai.com&sz=32)
Think of a locksmith who has tried every key on a ring, only to have someone arrive and say, “Have you considered that this door might open from the other side?”
That’s the sort of surprise mathematicians are experiencing.
What happened this week?
This is probably the news you’ve been hearing.
On October 7–8, OpenAI released a huge collection of AI-generated mathematical manuscripts—more than 700 documents reporting hundreds of mathematical results across numerous fields.
Some address problems that mathematicians have worked on for decades. Others claim advances in questions that have resisted generations of researchers.
The mathematical community has reacted with a mixture of astonishment, excitement, frustration, and alarm.
(https://www.google.com/s2/favicons?domain=https://www.theverge.com&sz=32)
But there’s an important distinction between an AI producing a proposed proof and mathematicians establishing that the proof is correct.
In mathematics, a proof is not accepted merely because the conclusion sounds plausible. Every step must follow rigorously from what came before.
One way to check that is with specialized software, such as Lean, which can verify formal mathematical arguments. Many of the newly released results have not yet undergone that level of verification or independent scrutiny.
So the headline “AI solved hundreds of problems overnight” describes the scale of the reported results, not hundreds of discoveries that have all been conclusively certified.
(https://www.google.com/s2/favicons?domain=https://www.theverge.com&sz=32)
Why are mathematicians excited—and frightened?
Imagine that you’ve devoted fifteen years to understanding one particularly difficult question. You’ve made progress. You’ve developed new techniques. You’re beginning to see a possible solution.
Then, one morning, somebody delivers a manuscript claiming to have solved your problem—and hundreds of other people’s problems as well.
You might be thrilled that the answer has finally been found.
But you might also feel that fifteen years of your intellectual life have suddenly been overtaken.
Some mathematicians are worried about exactly that. They’re also concerned about the difficulty of checking so much work, the possibility of mistakes, the concentration of powerful AI systems in private companies, and what happens to younger researchers who need opportunities to develop their own mathematical abilities.
(https://www.google.com/s2/favicons?domain=https://www.businessinsider.com&sz=32)
There’s another concern that I think you, of all people, will appreciate.
Mathematics isn’t only about getting the right answer. It’s about the intellectual journey that makes the answer understandable.
A proof can establish that something is true without necessarily providing the most illuminating explanation of why it’s true.
And mathematicians love understanding why.
Suppose an AI produces a correct solution to a problem that has occupied mathematicians for fifty years. That’s an extraordinary achievement.
But suppose the solution is so complicated, or so unlike familiar mathematical reasoning, that nobody can see the beautiful underlying idea.
The problem may be solved, but the human intellectual appetite isn’t necessarily satisfied.
That’s one reason researchers are interested in AI systems that collaborate with mathematicians, rather than merely delivering enormous piles of completed proofs.
And there are encouraging results from that approach, too. In one independent research test reported in June, AI systems produced acceptable solutions to seven of ten difficult research problems, judged by expert mathematicians. They were impressive, but far from infallible.
(https://www.google.com/s2/favicons?domain=https://simons.berkeley.edu&sz=32)
And here’s what I find particularly interesting
Remember your therapist’s office, with the armload of puzzle pieces scattered across the floor?
[TL note: I had said that when I go to a therapist, I feel like I’m bringing a pile of puzzle pieces, dumping them out on the floor, and I want the therapist’s help sorting them and putting the puzzle together.]
Imagine having a collaborator who can examine those pieces and say:
“Look at this one. It might fit with something discovered in an entirely different branch of mathematics eighty years ago.”
Then imagine that the collaborator can try a thousand possible arrangements before lunch.
That is part of what AI is beginning to offer mathematics.
But the human mathematician can still ask a different question:
“Why is this particular arrangement beautiful? What does it tell us? What new questions should we ask?”
And the mathematical community is now confronting the possibility that the balance between those two activities—finding answers and understanding their significance—is going to change profoundly.
I don’t think anyone yet knows exactly what that change will look like.
One of the world’s leading mathematicians, Fields Medal winner Alessio Figalli, has argued that AI will not eliminate the need for mathematicians, while expressing serious concern about how the next generation will learn mathematics.
So, to answer your original question as simply as I can:
AI has begun discovering mathematical truths that human beings hadn’t discovered, sometimes in remarkably little time. Mathematicians are excited because this could open whole new territories of knowledge. They’re dismayed because it could transform the work they’ve devoted their lives to—and because producing an answer isn’t always the same as understanding it.
…
that could be converted into math-muscle *)