In advanced economies, public authority is undergoing a silent transformation that affects the very nature of decision making power. The introduction of artificial intelligence systems into governance processes does not represent a simple technological upgrade, but a reconfiguration of the ways in which decisions are constructed, justified and made visible. The modern balance between decision, responsibility and the public nature of public action is placed under pressure by a form of governance that increasingly operates upstream of formal choice, within the domain of computation, prediction and automated classification.
The transfer of decision making toward algorithmic systems advances incrementally. It does not appear as an explicit delegation sanctioned by political acts, but as a sedimentation of tools that promise efficiency, coherence and error reduction. Risk assessment algorithms, scoring systems for access to social benefits, predictive models for the allocation of healthcare resources or for crime prevention progressively become integral components of administrative action. According to OECD data, more than 60 percent of public administrations in high income countries already use AI based decision support systems in sensitive areas. Formally, the decision remains attributed to an official or a political body; substantively, the field of options is pre structured by models that delimit what appears rational, efficient and defensible.
This shift produces a redefinition of political authority. Modern authority has historically been grounded in the capacity for judgment and in the public assumption of responsibility. Automated governance introduces a model in which authority is exercised as validation of technical outputs rather than as autonomous deliberation. The public decision maker tends to become a certifier of solutions generated elsewhere, while effective power concentrates in design criteria, in the datasets used and in the parameters that orient system behavior. Studies conducted by the European Commission on the use of AI in public administration show that, in many cases, algorithmic logics are poorly documented and difficult to audit, even for the institutions that adopt them.
The reduction of human discretion is often presented as progress. Less arbitrariness, greater coherence, standardization of decisions. However, discretion does not represent only a source of risk, but also the space in which political responsibility is exercised. It is the place where the decision maker recognizes the exception, evaluates context, and assumes the cost of a choice that is not statistically optimal but necessary on social or ethical grounds. Automated systems privilege what is measurable, replicable and predictable, leaving at the margins dimensions that are difficult to formalize, such as substantive equity, dignity or long term impact. World Economic Forum reports highlight how the indiscriminate adoption of automated decision systems can produce regressive effects on trust in institutions, precisely in contexts where social complexity would require greater capacity for mediation.
The implications for democratic legitimacy are significant. Legitimacy does not derive exclusively from the effectiveness of public policies, but from the intelligibility and contestability of decision making processes. When decisions become the product of complex and opaque models, public debate shifts. What is discussed is no longer the choice itself, but the technical correctness of the system that generated it. Political critique is replaced by procedural verification, reducing the space for dissent. In this transition, the citizen confronts a power that decides without appearing, operating through a rationality that is difficult to translate into political language and therefore difficult to appropriate as an object of democratic contestation.
This dynamic recalls Hannah Arendt’s reflections on the relationship between power, authority and responsibility. Arendt distinguished power, which arises from collective action and consent, from mere technical administration. When government is reduced to process management, she argued, the space of political action contracts. Automated governance tends to realize this contraction in a technologically advanced form, replacing public confrontation with procedural optimization. The risk is not inefficiency, but a form of hyper efficiency that empties decision making of its political content.
Similar experiences are already observable in the private sector. Firms that have adopted automated decision systems are familiar with the limits of a leadership that hides behind numbers. Analyses by Harvard Business School show that reliance on predictive models improves average performance, but increases the risk of systemic errors when the assumptions embedded in the models are not questioned. More mature managers have understood that delegating computation does not mean delegating responsibility. On the contrary, it requires a strengthened capacity to interrogate system assumptions, recognize biases and assume ownership of decisions even when they are suggested by a machine. This lesson takes on central relevance for public authority as well.
Automated governance does not constitute an inevitable destiny, but an implicit political choice. Accepting it without questioning its effects means redefining authority as a technical function separated from judgment and responsibility. Governing it instead implies recognizing that automation does not eliminate power, but redistributes it; that technical neutrality is a construction; that every system incorporates a worldview. In a context in which power increasingly operates through invisible architectures of computation, the institutional challenge concerns the capacity to make the loci of decision visible again, to clarify chains of responsibility and to preserve the space of deliberation.
Public authority in the era of artificial intelligence thus operates within a structural tension between efficiency and legitimacy, between computation and judgment, between automation and responsibility. Within this space, the boundaries of democratic governance and the role of institutions are redefined in economies increasingly mediated by systems that compute with growing precision, but cannot replace political understanding of the consequences of the decisions they orient.
Global AI Observatory