For Lodder, AI is not a new topic. As early as the 1990s, he conducted legal research within a computer science project. What is changing now is mainly the practical impact. “For a very long time, it was mostly predictions and theory. Now it also works in practice,” he says. This makes the discussion about law and responsibility more urgent than before.
AI systems are developed by people and organisations that make choices. Which data are used? What goals is a system given? And on what basis is an outcome produced? These choices are not neutral. They can have major consequences for citizens, for example when AI is used in education, employment, or healthcare.

That is precisely why, according to Lodder, the use of AI must be considered at an early stage. He too often sees organisations joining in simply because a technology is new, without clearly formulating what problem they are trying to solve with it. “There is not enough critical thinking at the outset about what you actually think you can achieve. I find that concerning.”
Lodder therefore does not think it is a good idea to leave ethics solely to developers: “Ultimately, companies that use technology want to make a profit. The moment ethics does not contribute to that profit, they will be less interested in it.” That is exactly why enforceable rules are needed. Not to stop innovation, but to set clear boundaries.

The black box

An important problem with AI is that systems are often difficult to understand. It is not always clear how an outcome has come about. This is also referred to as the “black box.” For Lodder, this is not only a technical problem, but also a social and legal issue. “It is often not even clear why it happens in that way. I find that black box quite concerning in AI applications, also from an ethical perspective.” Without explanation, it is difficult to assign responsibility.

According to Lodder, that responsibility does not lie with one party alone. Developers, organisations, users, and governments all play a role. Still, he observes that regulation often focuses heavily on the user of a system. “In general, regulation pays too little attention to the developer. A great deal of responsibility is placed on the user.” Yet it is precisely developers who make important choices at the start. They determine which data are used, which assumptions are embedded in the system, and which goals are technically built in.

Critical thinking

Biases also pose a risk. AI systems are often trained on existing data. If those data contain inequality or stereotypes, a system can adopt or even reinforce those patterns. Lodder illustrates this with the example of car insurance in the United States. In some cases, the premium was partly determined by the way people paid their credit card bills.

As a result, people with less financial flexibility could receive a higher premium, not because they drove less safely, but because their payment behavior was used as a risk signal. According to Lodder, this is problematic: “That has nothing to do with your driving behavior as such.” The example shows that an AI system can appear to calculate neutrally, while in fact translating social inequality into financial disadvantage. The problem therefore does not only arise at the moment AI is used, but much earlier: in the design of the system and in the selection of the data on which decisions are based.

For that reason, responsible AI, according to Lodder, begins with critically examining the foundation. Which data are we using? Why are we using those data? Which groups might be missing? And who checks whether the outcomes are correct? AI should not be treated as a system that automatically provides objective answers. Lodder therefore emphasises the importance of the “why question”: “Why do you want to use a particular form of AI?” According to him, organisations must “think more critically about what they want to achieve” before deploying AI.

Human oversight and critical questions are essential in this process, because otherwise there is a risk that systems will be used without it being clear what problem they are actually solving.

European AI Regulation

The European AI Regulation attempts to gain more control over this. It classifies AI systems into risk categories. Lodder considers this a positive development, because it requires organisations to think in advance about the consequences of their technology. Still, he also sees limitations. The law focuses strongly on sectors and applications. An AI system in education, for example, may be considered high-risk, while some applications in that sector are fairly harmless. Conversely, systems outside those categories may still pose significant risks.

Lodder therefore argues for a broader approach. It should not only be the sector that determines how strict the rules are, but above all the potential impact. “As far as I am concerned, you should approach this in a more generic way and not based on a specific type of application,” he says. What happens if the system makes mistakes? Who could be affected by them? And how dependent are people on the outcome?

Another important point is supervision. Rules are only valuable if they are properly enforced. According to Lodder, this is where a challenge lies. Supervisory authorities must have sufficient knowledge, capacity, and access to genuinely assess AI systems. Otherwise, there is a risk that organisations will mainly be occupied with forms, procedures, and documents, while little changes in practice. “Then you get a kind of paper reality,” he says, “without really thinking about what it should actually be about.”

Lodder’s message is clear: AI requires not only smart technology, but also clear rules of the game. Ethics helps determine what is desirable and responsible. But when voluntary agreements fall short, the law must set boundaries. Only in this way can transparency, responsibility, and fundamental rights also be protected in practice.

Interview by Bodiel Siebers