The Self-Driving State: Why Public Administration Is the Ideal Candidate

When "Das selbstfahrende Unternehmen" was published in 2021 (English edition: "The Self-Driving Company", Springer 2023), the same question kept coming back in our consulting work for ministries and public administrations in Austria and Germany: Does this apply to us as well? Together with Patrick Pils and Philipp Seper-Ambros, I have worked this question into a book of its own, which Springer Gabler will publish this year: "Der selbstfahrende Staat". The answer is clearer than many expect: Public administration is not a latecomer to autonomy but its ideal candidate, and precisely because of the two traits that are otherwise held against it as weaknesses. It is bound by rules, and it must justify and document every decision.

The starting point: more applications, fewer people

Two curves are diverging. The number of interactions between citizens, companies, and the state grows with every simplification: Those who can submit applications digitally submit more of them, and every new service generates new inquiries. At the same time, the workforce is shrinking. In Austria, around 45 percent of federal civil service staff will retire by the mid-2030s, and in the federal administration's IT departments this wave starts even earlier (Die Presse, 2022). The labor market cannot supply skilled staff in these numbers, even if every graduate of a relevant program chose a career in public service.

What happens when an inherently simple process meets this gap was documented by Austrian public broadcaster ORF on August 17, 2021, using the example of Vienna's Municipal Department 35: Of roughly 450 incoming emails, at most 120 could be processed, and telephones went unanswered because every answered question triggered new inquiries. Yet extensions of residence permits are largely a legal formality, while for those affected they are existential. The process is not complicated; it simply occurs too often for the hands available.

The classic way out, having more staff work according to uniform guidelines, runs into the same limit. That leaves the second path: software that takes over the rule-bound share of the work. This is exactly where the observation at the heart of this article begins.

Four levels: from the analog to the self-driving state

Our model transfers the evolutionary levels of the self-driving company, which are modeled on the autonomy levels of vehicles, to the state. The 80 percent thresholds are target marks of the model, not measured values; they mark the point at which a level shapes the character of an organization.

The last level is decisive: Laws continue to be written in prose, but they are additionally published as digital twins of law in a formal, machine-readable form. The legislature enacts both, the judiciary reviews and certifies the formal rulebooks, and the executive implements them in case-processing systems and platforms. The remaining roughly 20 percent of decisions, such as discretion, hardship cases, and everything political, stay with people.

Three traits that make public administration the ideal candidate

In "Economy and Society" (1922), Max Weber described bureaucracy through characteristics such as rule-boundness, written records, and fixed jurisdictions. A century later, these characteristics read like a requirements catalog for software. Three of them are decisive.

The rules are already written

A company that wants to become self-driving first has to find out which rules it actually decides by. A large part of this knowledge sits in the heads of experienced employees, in habits that have grown over time, and in exceptions nobody has written down. Before software can decide, this knowledge has to be made explicit, and in practice that is the most laborious step.

Public administration has this step behind it before it even starts. The principle of legality binds it to the law: Under Article 18 of the Austrian Federal Constitutional Law, the entire public administration may be exercised only on the basis of the laws, and Article 20(3) of the German Basic Law binds the executive to law and justice. Every official act has a legal basis: a statute, a regulation, a decree. These rulebooks are public, versioned in the official gazette, democratically legitimized, and apply equally to everyone. At its core, bureaucracy is nothing other than a system of explicit rules designed to minimize arbitrariness and ensure equal treatment.

Software can take over tasks reliably when the requirements are clearly defined and predictable. Nowhere are they written down as completely as in the law. What is missing is only the form: Laws exist as prose, not as formal rulebooks. The digital twin of law closes this gap. For the book, we translated the Austrian IT collective agreement of 2023 into such a machine-readable form in XML, together with the ERP vendor BMD, the Upper Austrian Chamber of Commerce, and a legal expert. The lesson from this: The salary table with minimum base salaries and advancement rules could be formalized directly, whereas procedural provisions on negotiating and the validity of the agreement could not. The real challenge lay not in the technology but in the hierarchical dependencies between constitution, statutes, and regulations; that is why we recommend not starting with a wholesale conversion but taking every amendment as the occasion for a twin. The rulebook itself already existed.

The duty to give reasons is the specification of traceability

The second trait is felt even more often as a burden. A high share of working time in the analog state goes to documentation and record-keeping, because almost every decision has to be justified and recorded in writing. In companies, comparable effort counts as overhead to be rationalized away. In public administration it is required by law, and rightly so.

Because what the law demands of an administrative decision is exactly what must be demanded of an automated decision. Section 60 of the Austrian General Administrative Procedure Act requires the statement of reasons to summarize the results of the investigation, the considerations decisive for weighing the evidence, and the assessment of the legal question based on them; Section 39 of the German Administrative Procedure Act requires the essential factual and legal grounds. Translated into the language of software: data basis, derivation, legal basis. A system that cannot show, for every decision, which data and which legal basis it rests on is not acceptable in the public sector. This criterion existed long before anyone talked about artificial intelligence.

The duty to give reasons thus acts as a filter for the architecture. It pushes toward decision cores that can be defined, validated, and certified in advance, in our model the digital twins of law, with complete logging and a full history of every decision. Learning methods have their place where patterns need to be recognized, for instance in fraud detection; but they must not build solely on historical experience data, because old biases would be carried forward in them, and their results must withstand the same checks. And the duty turns around: What costs working time today emerges as a by-product in a self-driving procedure, because every transaction is automatically certified and traceable. The duty to document remains; the documentation effort disappears.

The separation of powers is a built-in quality management system

Anyone who transfers decisions to software in a company first has to build the governance for it: Who defines the rules, who reviews them, who operates the system, who hears the complaint? The state has had this structure for centuries. In our model, the legislature defines the formal rulebooks and publishes them openly, the judiciary audits and certifies the rulebooks and the executing software, and the executive provides platforms and case-processing systems. Added to this are control bodies that no company has in this form: administrative courts, courts of audit, ombudsman institutions, and parliamentary oversight.

This produces a feedback loop: Wherever an inconsistency becomes visible, it can be traced back to its origin using the logged decision, right down to the question of whether the rulebook itself needs to be corrected. The existing separation of powers is therefore not an obstacle to digitalization but its ideal precondition. This applies all the more since the European Parliament adopted the AI Act on March 13, 2024: Many applications in the public sector will be classified as high-risk and will require documentation, transparency, and human oversight. No other sector already brings the institutions for this with it.

What makes public administration appear slow is at the same time its most important precondition for autonomy: It acts according to written rules and gives reasons for every decision.

What "ideal candidate" does not mean

So that the thesis is not misunderstood, four clarifications:

• No administration without people. The target mark is around 80 percent of decisions, not 100. Discretionary decisions, hardship cases, and all political determinations remain with people. Whether 80 or 100 km/h applies on a country road is decided by politics; the software supplies the analyses for it.
• No software that interprets laws. What gets automated are the rule-bound parts of a procedure. Generative AI systems such as ChatGPT can help with understanding and drafting; the decision itself rests on reviewed and certified rulebooks.
• No perpetuation of old discrimination. Data and algorithms must be checked for bias before deployment and continuously thereafter. The advantage over human decision-making: This check is possible at all, because every decision is logged. Transparency toward those affected is what distinguishes the transparent state from the surveillance state.
• No lighthouse projects. Individual apps and portals do not change the level. What is needed is a foundation of registers, the EU's once-only principle (Regulation (EU) 2018/1724), and digital base services, on which each level is built step by step.

The next step

The book "Der selbstfahrende Staat" will be published by Springer Gabler in 2024; we already describe the evolutionary levels and the model behind them on the page The Self-Driving State. How we support ministries, federal states, and municipalities from digitalization strategy to independent project controlling is described under Public Administration. In the coming articles of this series we will go deeper into individual building blocks of the model; if you would like to know beforehand which level your organization is at, get in touch with us.