Artificial Intelligence in the formulation and customization of public policies: Between efficiency, equity, and democratic risk

Abstract

The integration of artificial intelligence (AI) systems into the public policy cycle is substantially transforming how states diagnose problems, design interventions, allocate resources, and evaluate outcomes. This article critically examines how AI reconfigures the relationships between administrative efficiency, distributive equity, institutional transparency, and democratic legitimacy, and proposes an operational framework for its responsible adoption in public contexts. This analysis is based on the question: under what conditions do the benefits of AI justify the risks it poses to contemporary democratic governance? To answer this question, this research develops a conceptual and methodological analysis based on four pillars: efficiency, equity, transparency, and legitimacy. Each pillar begins with an operational definition, moves on to verifiable indicators that allow for the assessment of its level of compliance, and finally defines the normative limits that should guide its application. Similarly, the research distinguishes between political personalization and administrative personalization, demonstrating that administrative personalization improves the concentration and provision of services. However, it can also exacerbate inequalities if there are biases in the data or weak human oversight. The article proposes a governance sequence before, during, and after implementation that links algorithmic impact assessment, human guidance at key points, periodic external audits, traceability, disclosure of rationale, and practical appeals mechanisms. Drawing on international cases in health, social protection, and education, it identifies enabling requirements and veto assumptions to ensure that efficiency is not prioritized over social justice, non-discrimination, and proper accountability. This research concludes by noting that AI can help strengthen state capacity, but only if its implementation takes place within an institutional framework that safeguards citizen participation, democratic control, and transparency. AI does not replace human judgment or the deliberative process; it complements them when there are clear safeguards, verifiable standards, and continuous oversight focused on the public good.