Discriminating Data

Discriminating data names the dual operation by which algorithmic systems both rely on discrimination (pattern-matching, categorization, correlation) and reproduce discrimination (racial, economic, geographic sorting effects). Chun traces the mathematical apparatus of machine learning—correlation coefficients, regression analysis, clustering algorithms—to their historical origins in Francis Galton's eugenic project: the scientific management of human heredity through statistical sorting. Galton developed correlation explicitly as a tool for predicting which populations should reproduce and which should not. The mathematics he invented for eugenic purposes became the foundation of modern statistics, and those methods—carrying their original design assumptions about the sortability of human populations—now power the pattern-matching engines of contemporary AI. The training data reflects existing distributions of power, recognition, and opportunity. The model learns these patterns. The outputs reproduce them. Not through malicious intent but through the statistical mechanics of pattern-matching against a biased corpus.