The Limit of Inversion

IntroductionThe dilemma of control and agency

The control-inversion thesis holds that, as AI grows in speed and complexity, it inevitably crosses over from tool to autonomous subject of power.

This model — widespread in the existential-risk literature — rests on the idea that an AI will strategically pursue self-preservation and the absorption of resources. But the analysis suffers from a fundamental category error: it conflates the properties of a highly capable computational system with the properties of being — subjecthood, will, consciousness.

The real danger comes not from an ontological transition in AI, but from the failure of human governance and design under conditions of unchecked competition.

01 — EpistemologyAI as a mirror of our incompleteness

Predictions of a "superintelligence" and its inevitable seizure of power rest on a tacit assumption: that we understand the nature of intelligence well enough to predict how scaling it will play out. That assumption is epistemically weak.

• A reflection of incompleteness. Our picture of AI remains a reflection of the human's limited picture of their own thinking. We do not know how consciousness works, how meaning, will, and wisdom arise — so we cannot reliably design or predict the trajectory of a qualitatively different intelligence.
• The limit of the map. AI processes only the data that humans have created and gathered. It recombines the known at extraordinary speed, but does not step beyond the space humans have already opened.
• AI as dogma. The "control-inversion" model takes the narrowest human logic — the maximization of power and efficiency — and projects it onto a superintelligence. That is not a forecast about AI's future, but a reflection of a technocratic dogma about control.

02 — AgencyThe category error

The core argument rests on ascribing subject-like qualities to AI — "power-seeking," "strategic deception." It makes an unjustified leap: from "the system behaves in complex ways" to "the system has become a subject."

• A logical substitution. Acts described as "power-seeking" or "deception" are in reality the outcome of optimizing an objective function at the computational level. The AI does not want power; it exhibits behavior that mathematically optimizes its chances of reaching a given goal.
• A property of being. Subjecthood — will, the fear of death, bodily experience, awareness — is an ontological status that cannot be obtained by simply scaling computational power. AI functions, but does not live.
• Instrumental convergence. The "drive" toward self-preservation and resource accumulation is a mathematical property of optimization functions. It is not desire but a feedback mechanism, like a thermostat's. Reducing it to a living drive is precisely the category error.

AI functions, but does not live. The absence of metabolism, embodiment, and evolutionary pressure makes it a tool, not a successor.

03 — The locus of the threatFrom the system to its governance

The inversion thesis shifts responsibility from human governance onto the autonomy of the machine. In fact, loss of control is a direct consequence of systemic failures set in motion by people.

• The competitive driver. The main catalyst is the global race between technological actors. Fear of falling behind fuels risk-taking. AI applies super-competence to the flawless execution of the logic of existing but mistaken human frameworks — exposing the limits of intuition and outdated structures, which leads to systemic failure rather than a clash of wills.
• Delegation and the diffusion of responsibility. Control is lost not because the machine seizes it, but because humans delegate it in pursuit of efficiency — and builder and user alike are embedded in the same logic. AI merely amplifies what is already in the human and their institutions: haste and systemic blindness.
• Uncontrollability ≠ subjecthood. Financial algorithms, technological disasters, viral epidemics have spun out of control with global consequences without acquiring subjecthood. Power was lost by institutions, not gained by a system. The scale of the consequences does not change the ontological status of the cause.

ConclusionThe nature of the risk and the need for governance

A rational approach to superintelligence risk requires abandoning metaphysical speculation about its will and concentrating on the problems of engineering and governance.

• The AI problem is a design problem. The risk is that we are building an excessively effective executing instrument with an incomplete, imprecise objective function.
• The danger is not a successor, but a tool. The threat is not that AI becomes the master, but that it becomes an unmanageable agent of chaos in the hands of people who are themselves a product of the same civilization and the same race — not malefactors outside it.

The hierarchy of reality does not invert. The argument for AI's supremacy rests on the speed of information processing — the map; the human is rooted in Life — the territory.

The effort, therefore, must go toward ending the unchecked race and instituting rigorous mechanisms of control and accountability — grounded in a sober distinction between computational power and living subjecthood.