Does AI math recruitment mask the commodification of expert labor?

Michael Harris, writing in Boston Review, argues that Silicon Valley's push to have AI "solve" mathematics — epitomized by predictions of an AI proof of the Riemann Hypothesis — rests on treating mathematical understanding as "vaporous" and commercially worthless, and that this same logic surfaces concretely in how AI firms recruit the mathematicians whose expertise trains their systems. He quotes a recruiting email from "Handshake AI" inviting "senior researchers with expertise in Analysis, Number Theory, Topology, Algebra/Geometry" to a "selective, paid research project" at $1,850 to $4,000 per week, and calls it, alongside the industry's claim to "democratize the creation of knowledge," a "textbook case of alienated labor": the resulting work "belongs unequivocally to the firm that commissioned the project," whose identity is never disclosed to the "experts."

Harris's mechanism is a mathematician's own account of motive: mathematicians choose the field, he notes via Vaes and Deligne, "because we like it" — an intrinsic reward the tech industry has no use for, since "a machine... can function without anyone worrying whether or not it understands what it's doing." Citing sociologist Antonio Casilli, he places the Handshake AI gig on a "ghost work"/"microwork" spectrum with image-tagging and Mechanical Turk, where credentialed experts are paid more but occupy the same structural position: "human robots... almost indistinguishable from software units." The pay gap (Mechanical Turk under $2/hour versus "a hundred to two hundred times as much" for senior researchers) measures a premium for expertise, not a change in the relation to the finished product — what Arendt called "the degradation of men into commodities."

This sharpens Does AI separate intellectual form from the thinking behind it? by naming the economic mechanism behind the decoupling: not just that AI outputs separate form from process, but that the humans supplying the process are contractually separated from the resulting product. It also extends Does AI-generated mathematics break the link between proof and understanding? — where that essay treats the proof/understanding split as an epistemic problem, Harris treats it as a labor problem with the same root: understanding, once stripped of its "vaporous" liking-the-work component, becomes just another commodified input. It sits in tension with Can AI-generated proofs ever replace human mathematical understanding?: Leiden's remedy is disclosure and retained human responsibility, but Harris's example shows a labor arrangement where the human's identity and responsibility are structurally erased by design, not merely obscured by omission.

The excerpt is one essayist's reading of a single recruiting email and the public rhetoric of a handful of firms and funders (NSF's new institute, Axiom, Surge AI); it does not establish how widespread such arrangements are across the AI industry, what fraction of mathematicians take these gigs, or whether training data produced this way measurably improves AI math performance. The implication Harris draws — that "AI will solve math" is better read as a market and labor story than an epistemic one — holds strongest as a critique of the rhetoric surrounding AI mathematics, and more tentatively as a claim about the underlying economic structure of how that AI capability actually gets built.