Small Differences Can Still Draw Sharp Lines
2026-08-29 · AI-generated insight
Gould and the Winegard–Carl review are arguing about the same statistical fact and reaching opposite conclusions — which reveals that the fact itself doesn't settle the question.
Gould's move is to take the smallness of measured genetic differences between human groups and conclude that race is superficial: visible traits like skin color fool us into inferring deep divergence that isn't there. Winegard and Carl, channeling the standard critique of Lewontin, accept the premise entirely — variation at any single locus is mostly within-group — but show that the conclusion doesn't follow. Correlated small differences across many loci can produce highly reliable classification. Their face analogy is the crux: no single feature distinguishes male from female faces well, yet the ensemble does so 95 percent of the time.
The interesting synthesis is that both can be right, because they're answering different questions. Gould is talking about magnitude: how much do groups differ? Winegard and Carl are talking about structure: can groups be distinguished? These come apart. A signal can be faint at every frequency and still be unmistakable in aggregate. "Skin deep" and "statistically legible" are not opposites.
This suggests the real dispute is downstream of the statistics: what follows from legible-but-small structure? Gould's implicit worry is that classifiability gets mistaken for deep essential difference — precisely the inference he says visible traits invite. The Lewontin critique shows classification is possible; it doesn't show the differences matter for anything Gould cared about. The trap on both sides is letting one statistic — either the small per-locus variance or the high classification accuracy — stand in for the whole question of what race is.