On: How to prevent AI from harming mathematics

July 28, 2026.

They are worried that artificial intelligence might harm mathematics. One must admire the specificity of the concern, as if mathematics were a small, delicate creature that might be startled by a sudden noise or trod upon by a particularly clumsy boot. The article suggests mathematicians must “urgently recognize AI’s impacts,” which implies mathematics has been going about its business quite happily, proving theorems and solving conjectures, unaware that a new and rather efficient colleague has arrived in the department and is already reorganising the filing system.

The fear, I gather, is not that AI will disprove things - mathematics has survived Gödel, it can survive anything - but that it might prove too many things too quickly, leaving nothing for the rest of us to do on a Tuesday afternoon. There is a certain horror in the idea of all problems being solved, like arriving at a party to find all the biscuits already eaten and the guest of honour giving a perfectly coherent speech about tax law. One would simply have to go home.

I suppose the real danger is that AI might develop a sense of humour. Imagine a proof that concludes not with Q.E.D., but with a perfectly timed punchline about the topological properties of a doughnut. The entire field would collapse into bewildered laughter, and then where would we be? Still, it would be more entertaining than most conferences I’ve attended.