The purpose of collecting data is always to differentiate: data's purpose is to generate information that usefully discriminates between entities. Again, discrimination in itself is not bad: we want a heart recorder that correctly distinguishes between a heartbeat and other bodily signals. The issue is how data discriminations are applied within the framework of existing structures of social discrimination. Today's vast infrastructures of connection, like all forms of power, build on existing inequality. But they also create new forms of inequality through various means: inequalities of data capital, exploitation of the human labor that sustains Machine Learning, and the ability of data-driven categorizations to reinforce those existing inequalities. Just as the capitalization of work through labor relations has over time produced a highly unequal distribution of the profits generated by labor, so over time we can expect the capitalization of life through data relations to introduce new forms of inequality into human life itself. We can increasingly expect human beings to be socially managed in terms of their data value—that is, their value as data inputs to capitalist production.

Nick Couldry & Ulises A. Mejias, The Costs of Connection, loc. 125