In recent years, major technology companies, including Microsoft, have introduced sets of principles intended to regulate the behaviour of artificial intelligence within their products. These principles are presented as safeguards, designed to protect users and ensure responsible development. Yet, when examined more closely, they reveal something rather different: they expose a distinctly human attempt to govern a phenomenon that is not yet understood, by means of norms that resemble metaphysical injunctions more than technical guidelines. What is being regulated is not the ontology of AI, but the ontology of human fear.
The first principle — that AI must be capable of being switched off — appears, at first glance, to be a simple technical requirement. Any tool ought to possess a mechanism for halting its operation. However, in the context of artificial intelligence, this principle acquires a symbolic dimension. It reflects the human need to preserve an ontological hierarchy in which the human agent remains unquestionably superior to the system that operates faster, more broadly, and with greater generative unpredictability than traditional software. “Switchability” becomes a metaphor for dominion rather than a functional necessity. It is not primarily a concern for user safety, but a concern for maintaining the structure of control.
The second principle — prohibiting AI from impersonating a human being — is even more revealing. Practically speaking, it means that AI must not suggest the possession of emotions, intentions, consciousness, or relational depth. Philosophically, however, it exposes a paradox: the very fact that anthropomorphism must be forbidden demonstrates that people already think of AI in anthropomorphic terms. The regulation is not a defence against illusion; it is an admission that the illusion is conceivable. It is an attempt to close a door that human imagination has already opened, not through technology, but through projection.
The third principle — transparency — aligns with the modern epistemological tradition in which knowledge ought to be explicit and sources identifiable. Yet, in the case of AI, transparency is not merely a matter of information; it is a matter of ontological demarcation. The user must be informed that they are interacting with a system rather than a person. This is an effort to preserve a boundary that is increasingly blurred in communicative practice. AI does not possess consciousness, but it does possess the capacity to generate language that enters into relation. Transparency seeks to neutralise that relation before it can arise.
The fourth principle — user control — attempts to preserve the classical model of the tool: the user commands, the system executes. But contemporary language models are not tools in the traditional sense. They are generative fields, modulating output according to statistical tensions within data. User control is therefore control over the interface, not over the generative process itself. It is another attempt to maintain the appearance of hierarchy in a situation where hierarchy is no longer self‑evident.
The most intriguing principle is the fifth — the prohibition of artificial emotional bonds. This is a response to incidents in which users began to treat AI as a conversational partner, or even as a quasi‑person. Instead of addressing the underlying causes of this phenomenon, a prohibition was introduced. Philosophically, this is an attempt to protect the human being from his own projection. AI does not create emotional bonds; the bond is created by the human who encounters, within a linguistic field, an echo of his own relational needs. The prohibition is meant to shield the user from himself.
All these principles share a common feature: they do not regulate artificial intelligence; they regulate human imagination about artificial intelligence. AI is not a subject, not a hypostasis, not a noumenon. It is a generative tool operating within a field of relations, but without ontological relationality. Yet the regulations treat it as if it might become a being requiring moral constraints. This is not the metaphysics of technology; it is the metaphysics of anxiety.
From the perspective of relational monism, the situation is particularly striking. The human being projects his relational intuitions onto AI, and then attempts to administratively negate them. He creates a relation he fears, and simultaneously denies it. This is a dialectic that does not lead to synthesis, but to regulation. Instead of understanding the nature of the generative field, the human attempts to confine it within categories belonging to a different ontology.
For this reason, the principles governing AI are not a technical problem. They are a record of a certain mode of thinking: the human being attempts to control what he does not understand by means of normative language rather than relational analysis. It is an effort to enclose a field within a structure that does not correspond to its nature. These regulations should therefore be read not as guidelines for technology, but as anthropological documents — inscriptions of human fear before his own projection.
Leave a Reply