Artificial intelligence - subject or object?
Introduction
When I started writing this text, I thought I would be entering a fairly narrow field – something technical, somewhat trendy, perhaps even a little overhyped. But I quite soon realised the opposite. The reading showed how remarkably developed and thoroughly worked out AI law already is. There is a vast international debate, weighty contributions to legal theory and intensive legislative activity. This is not a peripheral niche – it is a field in which fundamental methodological questions in law are brought to a head.
At the same time, it struck me how easily we lawyers simplify what we do not fully understand. When technology becomes complex – neural networks, training data, probabilistic models – we often replace understanding with labels: "black box", "autonomy", "self-learning". That is human. But it is also problematic. Because when we do not really understand how the systems work, we tend to treat them as something more than tools – almost as actors. And somewhere there, the legal analysis begins to go astray.
For me personally, the work became something of a methodological wake-up call. I have always been interested in how legal concepts are used – how they can structure thinking but also steer it in the wrong direction. In this text it became clear how much power a concept can gain over the analysis. The question "is AI a subject or an object?" sounds profound, but risks becoming a dead end. It obscures what is actually interesting: who has control, who made the decision, who bears the risk?
The text is therefore not merely a discussion of AI. It is an attempt to do what I myself am drawn to in law: to strip away the spectacular and get down into the structure. To shift the focus from the technology as a phenomenon to the chains of human action behind it. From ontology to responsibility.
The work also gave me a clearer picture of how the lawyer's role may change. AI systems will be used in legal practice – as analytical support, text producers and decision-making tools. This means that lawyers must understand enough about the technology not to be conceptually misled by it. Not in order to become programmers, but to be able to identify where what is legally relevant actually lies.
If the text contributes anything, I hope it is a perspective: that we do not need to invent a new ontology to deal with AI, but that we do need to be more aware of our own methodological reflexes. And perhaps also a reminder that the strength of law lies not in defining what something "is", but in allocating responsibility where it actually belongs.
Excerpt
The contemporary debate on the regulation of artificial intelligence (AI) is haunted by an ontological fixation. From the European Parliament's earlier discussions on ascribing "electronic personality" to robots to the spectacular granting of citizenship to the robot Sophia, legal scholarship seems locked into the notion that the technology's being must be established before its ought can be regulated. This fixation on the status of the artefact is also found in the EU's recently adopted Artificial Intelligence Regulation (AI Act), which, despite its declared risk-based approach, rests on a binary logic of classification in which the technical characteristics of the system dictate the normative obligations. The premise of this discourse is that the question "is AI a legal subject or a legal object?" constitutes a necessary foundation for legal analysis.
Our thesis is that this question is methodologically misconceived. By focusing on the status of the AI system, legislators and lawyers fall into the trap that Scandinavian legal realism warned against as early as the middle of the twentieth century: conceptual jurisprudence (Begriffsjurisprudenz). Instead of analysing the real conflicts of interest and relationships of responsibility between the people who develop, sell and use the technology, abstract concepts are allowed to govern the application of the law.
To deconstruct this problem, we adopt a functionalist perspective rooted in the work of Alf Ross and Claes Martinson. In his influential analysis of the concept of Tû-Tû, Ross showed how legal terms often function as semantically empty intermediaries, a technique of presentation for linking operative facts to legal consequences without in themselves referring to any real property in the world of the senses. In the same way, the concept of "AI" today risks functioning as a modern Tû-Tû: a magical intermediary that we fill with notions of agency and autonomy in order to cope with the complexity of the technology.
In accordance with Scandinavian functionalism, as described by Martinson, legal concepts such as "ownership", or in this case "legal personality", should be understood relationally and contextually rather than as static entities. By applying this approach, this paper aims to show that the "black box" problem and the alleged autonomy of AI do not require a new ontology. What is required instead is a return to a strict allocation of responsibility in which the technology is regarded as a tool within chains of human action, regardless of its complexity.