Discriminação algorítmica e proteção de grupos em situação de vulnerabilidade pela jurisdição constitucional : dever de proteção estatal à luz da lei material da ponderação.

Abstract

This research is linked to the Postgraduate Program in Law at the University of Santa Cruz do Sul, within the Concentration Area of Social Rights and Public Policies and the Research Line “Instrumental Dimensions of Public Policies,” with an emphasis on Judicial Control in the enforcement of fundamental rights, by articulating constitutionalism and fundamental rights with the challenges of artificial intelligence in a new discriminatory context. The thesis investigates the reproduction of structural discrimination by algorithmic systems and proposes a normative–hermeneutic framework for their protection, with an emphasis on safeguarding groups in situations of vulnerability. It initially asks to what extent, and by which interpretive parameters, constitutional adjudication should protect groups in situations of vulnerability against algorithmic discrimination, calibrating transparency/explainability and trade secret/algorithmic secrecy. To that end, it begins by mapping the phenomenon and its causes—variable selection, learning from biased datasets, model and governance opacity—to show that technical opacity is not an accident but an expression of technocratic rationality that calls for specific legal responses. At the constitutional level, it articulates the State’s duty to protect with the renewed dimensionality of equality and non-discrimination in an environment of automated decision-making; at the infra constitutional level, it mobilizes the principles and mechanisms of the LGPD (Arts. 6 and 20) and the requirement of adequate reasoning (CPC, Art. 489, § 2) as the basis for standards of prevention, transparency, and review. First, a phenomenological approach is employed, through a literature review, to apprehend, in its entirety, the emergence of a legal-technological vulnerability within digital decision structures – avoiding, a priori, its reduction to established legal categories; next, a deductive method is adopted, starting from normative-constitutional premises to derive specific conclusions regarding the legally enforceable contours of judicial action in the face of violations committed by automated agents. Methodologically, the material law of balancing is adopted, drawing on Robert Alexy’s elaborations, to construct a decision model that grades judicial intervention, integrating the formula of weight in light of its concern with the degree of empirical and normative certainty (C), and allocating dynamic burdens of proof in contexts of informational asymmetry. The procedure enables the calibration of the collision between explainability/transparency and trade secret/algorithmic secrecy by means of reproducible and auditable criteria, allowing sufficient access to the logical-decisional itinerary without undermining the developer’s rights. From this follows a decision rule: the greater the interference (I) and the abstract weight (PeA) of the principle that yields, the greater must be the C required of the justifying premises and the broader the informational openness, tiered by A1–A3. The contribution lies in offering operational and verifiable criteria for judicial decisions on algorithmic discrimination—reducing informational asymmetries, strengthening the effectiveness of substantive equality, and consolidating a culture of public justification for automated choices.