Are AI's math gains and human math losses really connected?

Romero's essay yokes two pieces of news from the same week in September 2026: an AI-enhanced breakthrough toward solving the Navier-Stokes Millennium Prize Problem, claimed in some form by researchers at Anthropic and/or OpenAI, and the release of PISA scores showing, in his words, that "kids have been getting dumber every year for around 15 years now and the trend doesn't seem to be plateauing." His central claim is that these are not two unrelated stories but one: "as new AI models break into the highest spheres of mathematics, new humans are falling out of the lowest ones." AI's advance into elite mathematical work and humans' decline in basic numeracy and literacy are, in his framing, the same curve measured at opposite ends.

Romero is explicit that he isn't claiming a single cause: "nothing in this universe is monocausal." He traces the blame assigned to the PISA decline across three cycles — phones were blamed when scores were bad in 2018, COVID in 2022, and generative AI now — and grants "there is merit to this conjecture" without endorsing it as the full explanation. The pairing works rhetorically rather than causally: he offers the Navier-Stokes breakthrough and the PISA release as simultaneous data points in the same week, and asks the reader to hold the asymmetry — AI absorbing the top of a capability distribution while humans slide at the bottom — as the real story beneath that week's AI-lab credit dispute over who gets to "stamp their name on the result."

This sits alongside Does AI-generated mathematics break the link between proof and understanding?, which makes a sharper, mechanism-level version of half of Romero's pairing: both treat an AI-produced mathematical result as evidence that something in human mathematical competence is being hollowed out rather than straightforwardly advanced. Romero's other half — a capability paradox in which AI's gains and humans' losses move together — echoes Does AI help individual scientists while narrowing scientific focus?, though that note documents a measured trade-off within AI-augmented science itself, where Romero asserts a juxtaposition across two separate populations (AI labs and PISA-tested students) without showing the students in question used AI at all.

The excerpt does not establish that generative AI caused, or even contributed to, the PISA decline for the cohort described — Romero names it as the current year's blame candidate among three rotating explanations, not as a measured cause, and the decline itself predates mainstream generative AI by a decade. Nor does he show any mechanism linking AI's mathematical breakthroughs to PISA-tested students' performance; the two trends are simultaneous in his telling, not shown to be connected. What the piece supports, at the strength of an op-ed rather than a study, is a provocation: that AI capability and human capability should be read as a single distribution shifting rather than two independent trends, and that readers should be suspicious of monocausal explanations for either half.