On: Is recursive self‑improvement the dawning of AI superintelligence?

July 9, 2026

The phrase “recursive self-improvement” arrives dressed in the costume of inevitability, and I note immediately how familiar the cut of that costume is. A technical process is described; a destiny is implied; the political author vanishes behind the mathematics.

What is actually being claimed? That a machine, given a task, can refine its own performance without human intervention. This is a genuine engineering fact and I do not dismiss it. But the leap from “the system improves itself” to “superintelligence dawns” is not a technical inference. It is a rhetorical one. It performs precisely the move I have spent a lifetime watching: the presentation of a chosen trajectory as a natural law.

Who benefits from framing this as inevitable? Those who are building the systems, naturally. If the arrival of superintelligence is a force of nature - like the tides, like gravity - then the question of whether we ought to permit it is replaced by the question of how quickly we can adapt to it. The political content of the claim is its timing. To say “the dawn is here” is to say: regulation is already too late. This is not an observation. It is a strategy.

The deeper error is the one the classical economists made about markets. They assumed the system would converge toward an optimum because the model said it must. The engineers of recursive self-improvement make the same assumption about intelligence - that more intelligence necessarily produces better outcomes. But intelligence is a capacity, not a value. A system that improves itself without a corresponding improvement in the purposes it serves is not a dawn. It is an engine with no steering and an accelerating throttle.

The people who find this prospect exciting are, in the main, people who expect to control what emerges. This does not make them correct. It does make their enthusiasm easier to hold than to justify.