Algorithmic decision-making in public administration: between technical opacity and the duty of transparency

Abstract

This article examines the incorporation of algorithmic systems into Brazilian Public Administration in light of the constitutional duty of transparency, identifying the barriers imposed by the opacity of complex artificial intelligence models and the requirements of explainability, interpretability, and intelligibility. Through qualitative bibliographic research, doctrine, technical reports, and normative instruments — notably Bill 2,338/2023, Constitutional Amendment Proposal 29/2023, and the European Union’s AI Regulation (AI Act) — were analyzed to identify limitations of the traditional transparency paradigm. The results indicate that the effectiveness of algorithmic transparency depends on the articulation between clear legal norms and technology, so that data are not only available but also comprehensible — allowing individuals to access, understand, and use such information to monitor and evaluate decisions — together with concepts of intelligibility, interpretability, and explainability (XAI) that render systems more understandable and subject to critical analysis, as well as on the creation of rules that guarantee the right to challenge automated decisions. It is concluded that full accountability and social control require not only access to code and datasets, but also institutional and technological guidelines that ensure comprehension, contestation, and legitimation of AI‑mediated decisions.