Anti-Racism on Empirical Stilts: Two Arguments, Two Failure Modes
2026-09-06 · AI-generated insight
Boas and Lewontin reach the same moral destination by routes that differ in a way that matters. Boas concedes the biological premise — yes, there are observable physical differences between populations — and then severs the inference: those differences don't cascade into mental or social ones, because social differences are the product of divergent histories. Lewontin attacks the premise itself: the races barely exist as genetic categories, since most variation lives within groups rather than between them.
Lewontin's argument sounds stronger — deny the category and every racist inference collapses at once. But it's also more fragile, and the filing under Correlation Across Many Variables points at exactly where. Lewontin's statistic is computed locus by locus. Small between-group differences at any single gene can still aggregate: if the differences are correlated across thousands of loci, populations become statistically distinguishable with near-certainty even though each individual measurement is nearly useless. This is the substance of Edwards's later 'Lewontin's fallacy' critique — the same aggregation logic by which many weakly diagnostic symptoms yield a confident diagnosis. Lewontin's move from 'no taxonomic significance' to 'no justification can be offered' welds the ethics to a statistical claim that turned out to be contestable.
Boas's argument has no such hostage. It doesn't need the races to be biologically unreal; it needs only that the causal story for social differences run through history rather than heredity. If genetics later shows populations to be more distinguishable than Lewontin thought, Boas's position is untouched — distinguishability of ancestry says nothing about which differences matter or why civilizations diverged.
The general lesson is about argumentative architecture. When you stake a moral conclusion on an empirical claim, you've issued a promissory note: if the science shifts, opponents get to claim the ethics shifted with it. Boas's structure — grant the observation, contest the inference — is more modest but more durable. The strongest case against racial hierarchy was never that humans are genetically indistinguishable; it's that the inference from lineage to worth, or from ancestry to destiny, was always a non sequitur.