ARTIFICIAL INTELLIGENCE AND THE DUTY TO GIVE REASONS: Beyond the Black Box and the Limits of Explainability

The growing integration of Artificial Intelligence (AI) into public administration challenges a fundamental pillar of the rule of law: the duty to give reasons for decisions. The opacity of advanced models, such as Deep Learning ("black boxes"), creates a direct conflict with the need for transparency, justification, and oversight of administrative acts. In response, "Explainable Artificial Intelligence" (XAI) has been proposed as the solution to align technology with legal guarantees. This dissertation investigates whether XAI, in its current conception, is an adequate and sufficient tool to fulfill the duty to give reasons required by Administrative Law. Through a methodology that combines a systematic literature review, a conceptual analysis of the technology, and a legal analysis of legal reasoning, the study argues that XAI is fundamentally incompatible with normative requirements. The research demonstrates that a technical "explanation," focused on post-hoc computational processes, differs structurally from a legal "justification," which demands an ex ante rationale based on facts and legal norms. The conclusion is that XAI is an insufficient solution, as it fails to address the core issues of the legality and fairness of a decision. Instead of attempting to explain opaque systems, this work highlights the need to adopt inherently interpretable models for high-stakes decisions and to develop an algorithmic governance framework that subordinates technology to the principles of law, ensuring that automated decisions remain transparent, contestable, and legitimate.