Introduction
Is artificial intelligence in public administration a path toward efficiency, or the beginning of autocracy by automation? This text analyzes the vision of the Artificial State, warning against equating computational power with the right to exercise authority.
The reader will learn how to distinguish helpful tools from systems that erode democracy. We will explore the concept of the constitutionalization of AI, which allows us to utilize technology without surrendering sovereignty over the citizen.
Automation as a Tool Against Human Arbitrariness
The introduction of AI can protect citizens from the whims of bureaucrats. Humans are prone to cognitive biases, such as noise or the anchoring effect, which lead to inconsistent treatment of similar cases.
Algorithms can realize the ideal of the rule of law by ensuring decision stability. An example is the automated verification of income criteria, where a machine eliminates personal prejudice and reduces waiting times for benefits.
The key, however, is combating scale bias. While a human may err individually, a poorly designed model can replicate discrimination across thousands of cases simultaneously.
Automating Routine vs. Automating Judgment
Simply replacing humans with machines is not the goal. Real benefit stems from automating routine tasks, which frees up resources for matters requiring interpretation and empathy.
Danger arises with the automation of judgment. When AI decides whether someone is "credible" or "deserving," it enters the realm of values rather than technique.
To avoid situations where a civil servant becomes a mere formal signature on a system's output (automation bias), symmetry of responsibility must be introduced. Deviating from an AI recommendation cannot be organizationally riskier than blindly accepting it.
Public Value Over Blind Efficiency
High technical accuracy does not automatically translate to a better state. A system may be efficient yet destroy public value if its decisions are opaque and impossible to challenge.
Justice requires reason-giving—the obligation to provide specific reasons for a decision. A probabilistic result (e.g., "score 0.73") is not a legal justification, but merely a measurement.
To prevent the dominance of private AI providers, interoperability and antitrust policies are essential. The state must maintain public epistemic capacity to realistically audit the systems upon which the execution of law depends.
Conclusion
Dismantling the Artificial State does not require destroying servers, but rather separating prediction from decision. We must protect the right to political unpredictability against the conservatism of data.
Icarus did not err by the mere act of constructing wings, but by confusing the ability to fly with a lack of limits. The question remains: can we soar high using machines without forgetting where optimization ends and freedom begins?
Frequently Asked Questions
Why could the introduction of AI into public administration be beneficial for citizens and democracy?
The introduction of AI allows for the reduction of cognitive biases, prejudices, and the personal arbitrariness of officials, which promotes equal treatment in similar cases. Thanks to automation, administration becomes more predictable, faster, and more effective in realizing the ideals of the rule of law.
Why is simply replacing an official with an algorithm not the solution, and where does the real benefit of AI in administration lie?
The real benefit of AI in administration lies in reducing human arbitrariness, cognitive biases, and unjustified variability in decisions (so-called noise), ensuring greater stability and equal treatment given the same data. Algorithmic support can also relieve officials of mechanical tasks, allowing them to focus on exceptional cases that require human judgment.
Does high technical effectiveness of AI in administration automatically translate into better state functioning?
No, high technical effectiveness of AI does not automatically translate into better state functioning, as it may lead to the unreflective acceptance of recommendations and a loss of employee competence. The quality of the human-machine interface and attention to public value are key to ensuring that the system does not become arbitrary or discriminatory.
Why is the high technical effectiveness of an AI system insufficient to consider its decisions in public administration fair?
In a state governed by the rule of law, the efficiency of a system alone is not enough, because government decisions require public justification in the language of norms and facts, rather than just probabilistic measurement. Fairness depends on procedural fairness, which includes the possibility of challenging a decision and the obligation to provide specific reasons for the ruling.
Does the automation of administration exclude the possibility of control over state decisions and democratic accountability?
Automation does not have to exclude democratic accountability, as digitalization increases the auditability of the state and allows for the verification of policy effectiveness based on data. Models can improve the prediction of decision consequences; however, democracy must retain the power to define values and goals, which a machine is unable to resolve.
Where is the line between the beneficial use of AI in administration and the risk of creating a digital dictatorship?
The line is crossed when AI stops merely extending the cognitive capabilities of the state and improving citizen services, and begins to assume the function of establishing values or replacing normative decisions. The risk of digital dictatorship arises when a prediction result becomes unquestionable, and the system is opaque and devoid of control by the political community and the person whom the decision affects.
Where does the helpful role of AI in administration end and dangerous autocracy begin?
Automatocracy begins the moment a system's prediction and an algorithm's output take precedence over human participation and disputes over values. To prevent it, one must institutionally separate the technological competence of machines from the political legitimacy of power, maintaining democratic control over optimization goals and accountability for results.
How can the threats posed by the 'Artificial State' be neutralized without having to completely reject AI technology?
Threats can be neutralized through the constitutionalization of computational power—that is, limiting it by law and dividing competencies rather than abandoning the technology. It is crucial to separate the infrastructure owner from the legislative function and restore democratic control over which tasks machines perform and on what principles.
How can one technically and legally prevent a situation where an algorithmic prediction automatically becomes a binding administrative decision?
An institutional 'fuse' must be introduced to separate the model's output from the legal effect, ensuring that the prediction does not grant itself legal significance. This could take the form of an explicit legal rule, a democratically established threshold, multi-stage verification, an audit, or a mandatory appeals process.
How can we prevent a situation where a human becomes merely a formal signature under a decision made by an algorithm?
The algorithmic recommendation must be separated from judgment, and the right to challenge inferences and update profiles must be ensured. It is also crucial to design symmetry of responsibility, so that deviating from the system's suggestion is not organizationally riskier than its unreflective application.
How can we prevent the arbitrary power of private AI providers and platforms that de facto perform public functions?
Preventing the arbitrary power of AI providers requires applying the principle of proportionality, according to which an entity's increase in infrastructural importance entails greater obligations regarding transparency, non-discrimination, and accountability. Key is the introduction of interoperability, which makes it easier for users to switch providers, and the creation of mechanisms for social participation in defining the rules governing these systems.
Why can relying on predictive systems in administration hinder social change and the development of civil rights?
Predictive systems rely on historical data, which leads to the preservation of the current state of affairs and the replication of existing regularities. This inhibits social change and the development of civil rights because breakthrough reforms require the creation of new norms and values, rather than merely reflecting the statistical status quo.
How does the state's dependence on private AI technology providers affect its actual ability to exercise power?
Dependence on private AI providers limits the real power of the state by creating infrastructural dependency and a lock-in effect, which can make the theoretical right to change the system practically impossible. Additionally, the lack of in-house experts leads to information asymmetry, where the state's formal supervisory competence exists without the actual possibility of exercising it.
How can the blurring of responsibility in AI systems be prevented, and what are the physical costs of operating the state's digital infrastructure?
To prevent the blurring of responsibility, a human or institutional entity responsible for the system must be clearly defined, separating duties according to the competencies of the designer, provider, deployer, and public authority. The physical costs of digital infrastructure include the consumption of energy, water, and land, as well as the burden on networks and spatial planning.
How can the totality of the digital state be avoided, and how can citizens be ensured real agency in automated administration?
The central knowledge system should be replaced by a polycentric structure of diverse AI models, which increases system resilience and prevents technological monoculture. It is crucial to ensure the right to an analog path in important matters and to treat the individual as a citizen with political rights, rather than merely as a service user.
How can the Artificial State be dismantled without rejecting AI technology?
Dismantling the Artificial State requires ensuring the normative primacy of law over technology by creating institutions that make AI computing power subservient to human agency. Key is the introduction of an architecture where technology is subordinate to law, based on distinctions such as prediction versus decision and recommendation versus judgment.
