Why AI is Used to Govern AI

Governance is a form of regulation.

Cybernetics is the study of regulation in machines and living systems. It gives us two rules that matter here.

First, the regulator must model the system it regulates. This is the Conant-Ashby theorem. Every good regulator of a system must be a model of that system.

Second, the regulator must hold at least as much variety as the system it controls. This is Ashby's Law of Requisite Variety. Only variety can absorb variety.

A set of finite controls cannot govern a system of near-infinite variability. The controls run out of responses. The system reaches states the controls never anticipated.

This is why frontier labs place AI inside their own control architecture. They need adaptive regulation. Rule-based controls cannot generate enough variety to keep pace. The approach functions, but it is neither sufficient nor reliable on its own.

So how do you regulate a system whose variety you can never match?

You stop trying to match it. Requisite variety is not reached by stacking controls until they equal the system. That path has no end.

You balance the variety equation from both sides.

Attenuate the system. Constrain the operating envelope. Bound the states the AI can reach. Reduce the variety you are required to absorb.

Amplify the regulator. Use adaptive, recursive, and self-regulating controls. Let each level regulate itself. Variety is best absorbed close to where it arises.

AI in the loop only amplifies. On its own it becomes an arms race you will lose. Paired with attenuation, the two meet at a point you can actually hold.

The real work is engineering that balance.

That is the discipline of cybernetic regulation.

A necessary practice in meeting obligations in the age of intelligence.