Introduction
Artificial intelligence is ceasing to be merely a tool and is becoming a new cognitive infrastructure for the state and society. This radically alters the balance of power in the relationship between authority and the citizen.
In this article, we analyze why the technical proficiency of AI cannot replace political legitimacy. You will learn how classical philosophy helps protect democracy from the risks of algorithmic governance and the erosion of truth.
AI as Cognitive Infrastructure vs. Political Legitimacy
Treating AI solely as an optimization tool is risky, because a state is not a corporation. Its goal is not to maximize a single function, but rather the equitable distribution of rights and resources.
Every algorithm used in administration participates in the exercise of power. If we base decisions exclusively on prediction, we bypass legal norms such as the presumption of innocence or the right to a defense.
A prime example is tax fraud detection systems. Even if a model is statistically accurate, it cannot independently determine penalties without human oversight and public justification.
Classical Theories of Legitimacy and the Risks of Algorithmic Administration
Political theories reveal the gaps in automation. Hobbes's Leviathan may be strengthened by AI, but it is the state—not the code—that must bear responsibility for delegated power.
Locke and Rousseau remind us of the necessity of the consent of the governed and the existence of the volonté générale. The behavioral analysis of data is not equivalent to the political will of citizens expressed through deliberation.
Montesquieu warns against the accumulation of power. AI can technically integrate legislative and executive functions, destroying the constitutional separation of powers through the centralization of data.
The Algorithmic Administration as a Radicalization of Power and Knowledge Asymmetry
The implementation of AI shifts the state-citizen relationship toward a profound cognitive asymmetry. The citizen becomes a passive data profile, whose situation depends on invisible scoring systems.
The primary threat is the replacement of public rationality with technical rationality. Computational efficiency does not guarantee justice, as models often perpetuate historical inequalities embedded in training data.
To counteract this, polycentrism and human oversight are essential. Regulations such as the AI Act must enforce transparency and the ability to challenge any automated decision.
Conclusion
The ultimate stake is not whether a machine tells the truth more often than a human. It is about ensuring that we do not mistake statistical consensus for the result of a critical examination of reality.
We must build systems in which AI supports cognition but does not become a source of legitimacy. Otherwise, we will create a world where institutions know us perfectly, yet we no longer understand the mechanisms by which we are judged.
Frequently Asked Questions
Why is treating artificial intelligence in administration merely as an optimization tool insufficient and risky?
The state is not an enterprise optimizing a single function, but an order distributing rights and obligations, and every system supporting administration participates in the practice of power. Treating AI solely as an optimization tool is risky because it may lead to the replacement of legal and constitutional norms (e.g., the presumption of innocence) with statistical prediction and deprive the citizen of their role as a participant in the reasoning process.
How do classical political theories (those of Hobbes, Locke, Rousseau, Montesquieu, and Tocqueville) help in understanding the threats associated with implementing AI in public administration?
Classical political theories point to risks related to a lack of transparency and control over algorithms (Locke), the replacement of civic deliberation with behavioral analysis (Rousseau), and the technical centralization of power despite its formal separation (Montesquieu). Furthermore, they emphasize that AI can strengthen the institutional agency of the state or corporations (Hobbes), which, in the absence of a strong civil society, promotes excessive centralization (Tocqueville).
How does the introduction of AI into public administration change the relationship between the state and the citizen, and what threats does this entail?
The introduction of AI into administration can lead to a profound cognitive asymmetry between the state and the citizen, as well as the radicalization of surveillance by shifting from population description to individual prediction. The main threats include the loss of officials' ability to understand decision-making processes, the perpetuation of historical inequalities by algorithms, and the risk of marginalizing individuals in atypical situations.
Why is high predictive effectiveness of AI insufficient to consider its decisions legitimate in a democratic state?
In a democratic state, the effectiveness of AI alone is not enough because the legitimacy of power requires publicly acceptable justifications and the ability for citizens to challenge decisions. Moreover, true legitimation stems from the process of deliberation and argumentation in the public sphere, which cannot be replaced by the statistical prediction of user preferences.
How can legal regulations and oversight systems prevent democratic decision-making processes from being replaced by algorithmic efficiency?
Prevention is possible through multi-layered oversight, federated interoperability standards, audits, and independent expert centers. It is crucial to apply criteria of public rationality (such as legality and transparency) and to introduce bans on, among other things, social scoring and profiling in crime prediction.
Can the high predictive effectiveness of AI replace democratic decision-making processes and the legitimacy of power?
No, the high predictive effectiveness of AI cannot replace democratic processes and the legitimacy of power. These systems can support data analysis and recommend decisions; however, establishing norms of justice, deliberation, and political accountability must remain the domain of humans and the law.
Why does the fact that AI generates correct answers not mean that it provides knowledge in an epistemological sense, and what are the consequences of this for the public sphere?
AI generates correct answers based on linguistic regularities rather than an understanding of reasons or observations of the world, which means that the truthfulness of a result is not identical to knowledge in an epistemological sense. In the public sphere, this leads to the separation of information from its source and the sender's responsibility, and enables the automatic scaling of personalized political persuasion.
How does artificial intelligence affect cognitive processes and democratic debate in the context of human fallibility?
Artificial intelligence does not eliminate human fallibility but statistically processes it alongside wisdom, adopting stereotypes and disinformation from training data. In the sphere of public debate, AI can support criticism by providing counterarguments, but at the same time, it threatens the freedom of the marketplace of ideas by being able to artificially generate massive, apparent voices of support.
Why is the mere provision of correct answers by AI not enough for this technology to be safe for democracy?
The mere provision of correct answers is not enough because the safety of democracy requires a correction infrastructure that allows for the tracking of evidence and the public questioning of errors. It is crucial that the technology does not make the citizen a passive recipient of authoritative content, but rather increases their ability to independently participate in the process of investigation and critical verification of information.
How can AI-generated disinformation be countered, and is the technical detection of deepfakes sufficient?
Countering AI disinformation requires a combination of provenance mechanisms, content labeling, and editorial responsibility, along with recipient competencies and legal regulations (e.g., the AI Act). Technical detection of deepfakes alone is insufficient because it does not constitute a complete epistemic strategy.
Why are access to information and simple fact-checking not enough to protect democracy from AI-generated disinformation?
Fact-checking can lead to global skepticism and a decrease in the credibility of true information, which does not build a resilient society. Protecting democracy requires not only access to data but, above all, correction institutions and procedures that allow for the reasonable distribution of trust and the identification of responsible participants in communication.
How does AI affect the diversity of knowledge in the public sphere, and what mechanisms are necessary to protect the democratic process of seeking the truth?
AI can limit the diversity of knowledge by favoring statistically dominant content and averaging expression, which pushes marginal perspectives into the background. To protect the democratic process of seeking the truth, institutional solutions are essential: a provenance layer, transparent labeling of synthetic content, the development of citizens' verification skills, and the protection of infrastructural pluralism.
What is at ultimate stake in the relationship between artificial intelligence and human society?
The ultimate stake is whether society will build an epistemic order around machines in which humans retain the ability to distinguish a convincing statement from a justified judgment, and a popular position from a true fact. In this way, artificial intelligence can become an extension of collective intelligence rather than its substitute.
