America's mathematicians cost about as much as bowling alleys
Even if AI can do all their research, society will keep paying them to understand it.
Noahpinion · The end of the age of heroesAll of America's academic mathematicians together earn maybe $4 billion to $5 billion a year, at most, in one rough estimate.
That figure comes from writer Noah Smith, who says the total is a rounding error on the nation's output (GDP), about what Americans spend on bowling alleys, or on Halloween candy.
Most of those mathematicians are employed mainly as teachers, and only a few do frontier research, new work that pushes the field's known limits.
That kind of research brings little direct economic payoff, beyond some cryptography and the occasional application in physics. Smith says we already keep mathematicians on staff because grasping math matters for its own sake, not because it pays.
So even if artificial intelligence can do the research, society will likely keep paying a very small sum to employ people who check AI proofs, explain them, and decide what to ask next.
Questions people ask
How many academic mathematicians are there in the United States?
Federal labor data count roughly 30,000 to 35,000 postsecondary teachers in mathematical science, plus several thousand researchers in government, nonprofits, and universities. At typical academic salaries around $100,000 to $150,000, that payroll easily lands near $4 billion to $5 billion. Most teach undergraduates as their main job.
If frontier math has little economic value, why does anyone fund it?
A small share pays off directly in cryptography, computing, finance, and physics. The rest is supported through universities because teaching requires experts, and because societies value knowledge, prestige, and training in rigorous thinking. That cultural support does not depend on each theorem earning a return.
What would checking AI math actually involve?
It would mean testing whether a claimed proof holds up, restating it so students and other scientists can use it, and choosing which conjectures to pursue next. Much of that work looks like teaching, reviewing, and explaining. It is similar to how other fields use powerful instruments and still need people who know what the output means.