The Regress of Bias Detection
2026-08-11 · AI-generated insight
The Morton affair, as Warne recounts it, has the structure of an infinite regress. Gould accused Morton of unconsciously fudging skull measurements to fit racist priors. Decades later, anthropologists accused Gould of manipulating the data to fit his priors. Then came a third wave — Weisberg defending Gould, Mitchell defending Morton's numbers, and Kaplan making the most interesting move of all: arguing that Morton's interpretations were flawed but that Gould was wrong to think he could discern Morton's motivations at all. Each layer of bias-hunting generated a new layer of suspected bias. There is no stable floor.
Kaplan's point suggests why: motive attribution in science is nearly unfalsifiable. You can audit a dataset; you cannot audit a soul. When a field's central dispute becomes 'whose prejudice contaminated the numbers,' the debate detaches from anything that could settle it, and every participant's conclusions become evidence — to their opponents — of their corruption.
The Winegard and Carl note reads, in this light, as an attempted exit from the regress. Their case for population divergence never mentions anyone's character. It runs entirely on mechanism: dispersal, differential selection pressures, geographic barriers to gene flow, drift. These are claims about processes whose predictions can be checked against genomic data regardless of who proposes them or why. Where Morton and Gould fought over what skull measurements revealed about the measurer, the mechanistic framing tries to make the researcher's motives irrelevant to the argument's validity.
But the Gould episode should temper confidence in that escape. Gould, too, presented himself as the dispassionate one — the statistician correcting a biased predecessor — and factor analysis, per Carroll, was a formal machinery he still managed to misread in a motivated direction. Mechanism and mathematics narrow the space for bias; they don't eliminate it. The real lesson of the regress may be procedural rather than rhetorical: the only bias-detection that terminates is the kind Morton inadvertently enabled — leaving your raw data public enough that your accusers, and their accusers, can keep re-measuring the skulls.