It is not just that many instances of data discrimination occur far from the contexts in which the original data collection was authorized and its meaning shaped. Data’s very exchangeability depends on abstraction and “the stripping away of narrative” that “facilitates the . . . insertion of numbers in new locations and their adaptability in new contexts.” Once in circulation, classifications can generate not just first-order but “second-order corrections,” indeed, corrections in any higher domain without limit. Those “corrections” shape the life chances of human subjects. When a person gets categorized as a high-risk loan recipient in one context, it is likely to affect how they are categorized in other contexts.

Nick Couldry & Ulises A. Mejias, The Costs of Connection, loc. 448