Algorithmic profiles and democracy: meta-identities, classificatory power and contestability in the digital public sphere

Abstract

The increasing delegation of decisions to artificial intelligence systems in public and private organizations and on digital platforms has produced classifications that condition access, rights, opportunities, and recognition, often without those classified knowing the criteria under which they were evaluated. This article investigates this phenomenon from the concept of meta-identity, which is the classification that a socio-technical infrastructure assigns to someone by deduction, retains as a reference, and uses to treat them differently. It raises the question of why the possibility of contesting automated decisions, formally guaranteed in administrative law, does not extend to the criteria that produced them. The argument develops on three articulated levels. On the theoretical level, four modes of existence of a classification are distinguished, based on current use, archive, reference, and documentary record. From this, it was shown that only the documentary record becomes appealable, even though, by deriving from the others, it selects what it informs, without any control instrument verifying its fidelity to what actually operated. On an empirical level, a published study comparing human and algorithmic classifications of 150 audiovisual works reveals the central mechanism: the compression of particular categories into 3 broad, reusable labels, which renders the claim of uniqueness inarticulate. On a documentary level, the reconstruction of the Brazilian case of the Cortex System (2018–2026), the largest state platform for data integration and surveillance, demonstrates, using primary sources, how the asymmetry of contestability can be produced by the legal form itself. It concludes that democratic debate over automated decisions depends less on expanding existing appeal procedures than on making the correspondence between the recorded criterion and the applied criterion auditable.