How much AI can a constitutional state take? Where authorities rule on wind farms, building projects, or other applications, every step has to remain verifiable – and procedures are expected to get faster. Berlin-based Rulemapping Group aims to bring both together: its method translates laws into visual decision logic, so-called rulemaps, and uses AI only selectively. The approach is already in use in the first districts.

Behind Rulemapping is Ina Remmers, who previously co-founded the neighborhood platform nebenan.de. For her, the same question runs through both ventures: how can living together succeed – in the neighborhood just as in the relationship with the state? In the interview, she explains how AI can be used in administrative procedures without making decisions opaque, what would change if laws were created as code from the outset ("Law as Code") – and what Berlin needs as a GovTech location for good approaches to reach everyday administration.

Ina, with nebenan.de you worked on bringing people in neighborhoods closer together. Rulemapping is now about the relationship between people, the state, and rules. What connects these two ventures for you personally?

What drives me is the question of how social cohesion actually works. That's also what connects nebenan.de and Rulemapping for me.

nebenan.de was about coexistence in very concrete, everyday terms — civil society, essentially: Who actually lives next door to me? Who can I borrow a drill from? And who might I start a book club with?

Rulemapping is about the foundation that makes this coexistence possible in the first place: a state that's actually able to apply its own rules reliably. Because social cohesion needs trust. And that trust also grows when people experience that the state works, that its decisions are understandable, and that they can rely on it.

The core idea behind Rulemapping predates the current AI boom: legal logic gets broken down into individual, traceable decision paths. When did you realize that modern AI could give this exact method a completely new significance?

Law professor Stephan Breidenbach developed the method more than 20 years ago to teach his students how to think like lawyers: how do you break down a legal provision into individual steps of examination and make the underlying logic visible? Since then, Rulemapping has been used in many projects — among other things, to automate mass procedures at the federal level. Because once a law has been translated into machine-readable logic, many clearly defined steps can already be automated without AI at all.

The limits used to lie mainly where procedures involved unstructured information or discretionary decisions. With the rise of powerful language models, it became clear to us how much more would become possible. AI can extract information from extensive documents, support the evaluation of discretionary questions, and make transparent which aspects are relevant to a decision. And we can do all of that in an AI-agnostic way.

At the same time, this lets us create rulemaps faster. The visual logic already reveals ambiguities or contradictions in laws during the mapping process itself. That means we can apply existing rules better — and even help design new laws to be more consistent from the very start.

Generative AI works with probabilities, while legal decisions demand reliability and sound justification. How does Rule AI bring these two worlds together without losing traceability?

That's exactly the central challenge of using AI in public administration and the justice system: a probability-based tool is being asked to operate inside a rule-based environment. But in a constitutional state, an answer that's merely probably correct — and that could be distorted by hallucination — simply isn't good enough. It also has to remain traceable and legally sound at all times how a decision actually came about.

That's why Rule AI combines two different systems. Our rulemaps are the visualization of rules and laws. They lay out the legal examination path, with its prerequisites and dependencies. So we're not asking the AI, "Should this building permit be approved?" Instead, it's only brought in at individual, clearly defined nodes. There, it answers specific sub-questions or supports a discretionary decision.

For the human in the loop, it always remains visible which sub-question the AI answered, how it answered it, and what effect that has on the rest of the decision path. The assessment can be reviewed, edited, or corrected. That's how we use the strengths of generative AI without handing it the entire decision-making process as a black box.

Wind power permits and other complex procedures show just how many rules, agencies, and individual steps have to interact today. Where do you see the biggest leverage for AI in practice — and where would automation be the wrong answer, in your view?

We see the biggest leverage wherever very large volumes of information meet clear rules. With wind power permits, for example, application documents running thousands of pages have to be checked against numerous legal requirements. In the Neckar-Odenwald district, our Rule AI can now pre-check such documents in minutes instead of weeks. That cuts the review workload by up to 75 percent.

It's similar with building permits. In two districts in Thuringia, the entire procedure was translated into a machine-readable decision tree with more than 500 branches. Here, AI can capture information faster and prepare clearly defined review steps. Staff can then use the time they gain for complex cases, communication, and genuine deliberation.

The wrong answer would be full automation that obscures human discretion. A faster black box is still a black box. Responsibility, control, and the ability to object all have to stay with people.

You're now thinking not just about how laws are applied, but about "Law as Code": rules that are machine-readable from the moment they're created. What would that actually change for public administration, businesses, and citizens?

Our state isn't built for the digital age, because the rules and laws that its processes and decisions run on are still stuck in the analog era. Laws today are still drafted as plain text. For every digital procedure, they then have to be laboriously translated into software. That often takes months — and by the time the software is finally finished, the law has likely (very likely, in fact) already been amended again, and the programming has to start over. That creates a permanent backlog, and the familiar frustration people feel with our bureaucracy.

Law as Code would eliminate this translation step. Every law would get a digital twin from the very beginning, one that maps the rules in logic that's readable by both humans and machines. Administrations and businesses could integrate it directly into their applications. Changes to the law could be updated quickly across the affected processes, without having to reprogram every system individually.

That's the crucial point. Only machine-readable rules create the foundation for automating procedures quickly and at scale. Administrations and businesses could implement rules much faster. For citizens, that means faster procedures and decisions. Automation would be built in from the start, instead of chasing after a new law for months or years.

Rulemapping was founded in Berlin and works at the intersection of law, technology, public administration, and AI. What role does Berlin's ecosystem play in your development — and what potential do you see for the city to move GovTech and AI solutions out of pilot projects and into everyday administrative practice faster?

Berlin is an important place for Rulemapping because law, administration, technology, academia, and a vibrant startup scene all come together here directly. We need exactly that mix for the modernization of the state too. This proximity gives us the chance to work together on concrete solutions and test new approaches under real conditions.

To me, a functioning administration and reliable digital infrastructure have long been part of basic public service. Berliners experience every day, through an application, a permit, or an official notice, whether their city actually works.

During the election campaign, the leading candidates all agreed that Berlin needs more speed. What matters now is the "how." We need structural solutions that bring successful approaches into standard operations and make them usable beyond individual agencies.

The new Berlin government can lay the groundwork for that. It should make administrative digitalization a priority, establish shared standards, and make it easier to work with GovTech companies. Berlin has already been a pioneer in many areas. I'd like to see the city take on that role in digital administration too, going forward. Every process that works in a way that's understandable, fast, and reliable for people ultimately strengthens their trust in the state as well.

Thank you for the great conversation.

Note: This interview was originally conducted in German and subsequently translated into English language.