The Conant–Ashby result that any regulator which successfully keeps a system's outcome within bounds must be, or must contain, a model of that system. Success at regulation is itself evidence of internal representation — you cannot reliably steer what you do not implicitly model.
How would you explain it like I'm…
Catching Means Knowing
If you can reliably catch a ball someone throws, a little part of your brain has secretly learned how that ball flies. You can't steer something well unless a piece of you already knows how it behaves. So whenever something keeps a tricky thing under control, it's carrying a tiny map of that thing inside.
Control Needs a Model
There's a rule that says anything good at controlling a system has to carry a kind of map of that system inside it — even if nobody put the map there on purpose. Think about catching a ball: to catch it every time, your body has to predict where it's going, and that prediction is a tiny model of how balls fly. You can't reliably steer something you don't understand at some level, because to fix a problem you need to know which of your moves cancels out which disturbance. So whenever you see something keeping a system under control — a thermostat, an animal, a company — you can bet there's a model of that system hidden inside it somewhere.
Control Implies a Model
The good regulator theorem (Conant and Ashby, 1970) says that every effective regulator of a system must be, or must contain, a model of that system. More carefully: if something maps disturbances to control actions so the outcome stays within a target range, then that something has to match a model of how disturbances and actions combine inside the system to produce the outcome. The short version is: you can't reliably steer what you don't implicitly model. This is stronger than the obvious 'engineers use models to design controllers' — the theorem says any regulator that actually succeeds is implicitly a model, whether or not its designer realized it. The reason is structural: to keep the outcome bounded, the regulator must anticipate which of its actions counters which disturbance, and that anticipation requires a model. That's what makes it portable across evolved or designed, conscious or not — and turns 'where is the model?' into a question you can ask of anything observed to control something.
The good regulator theorem (Conant and Ashby, 1970) is the formal result that every effective regulator of a system must be — or must contain — a model of that system. Precisely: if a regulator maps disturbances to controller outputs so that the controlled outcome stays within a target set under the system's dynamics, then the regulator must be isomorphic, up to behavioral equivalence, to a model of how disturbances combine with controller actions inside the system to produce the outcome. Informally, you cannot reliably steer what you do not implicitly model. The structural pattern is that successful control is itself evidence of internal representation: whenever any agent — mechanism, organism, organization, algorithm — regulates a complex environment to keep an outcome within bounds, it must encode the structure of that environment somewhere, whether in its policy, weights, rules, heuristics, institutional memory, or trained intuitions. No purely reactive policy, however clever, can match a model-bearing regulator against a complex environment, because the regulator must anticipate which of its actions counters which disturbance, and that anticipation requires a model. The deep claim is distinct from the trivial reading 'regulators are designed using models,' which is merely a fact about engineering practice: the theorem says any regulator that actually succeeds implicitly is a model of the system it controls, whether or not the designer realized this. Success at regulation thus imposes a representational requirement on the regulator's internal structure — which is what makes the result substrate-portable, constraining all regulators (evolved or designed, conscious or not, neural or mechanical) and turning 'where is the model?' into an inference rule runnable on any agent observed to control something.
Converts "does this controller have a model?" from a metaphysical question into a structural inference: representational content becomes deducible from behavioral success rather than merely postulated.
Compresses a whole family of "you need a model" results into one statement, and reduces any malfunctioning regulator to a single diagnostic: where has the internal model gone stale, wrong, or absent?
Licenses a representational floor — given an environment's complexity, the regulator must carry at least that much model fidelity — and factors failures into variety failures (no action available) versus model failures (wrong picture of when to act).
Clinical medicine: the senior ICU nurse who acts before the alarm carries a richer model than the novice — evidenced by the superior regulation itself.
Control theory: the Kalman filter's estimate runs ahead of noisy readings because it is a model of the plant.
Organizational learning: a firm's "mental model" must be audited and updated as the market drifts.
A purely proportional controller leaves steady-state error against a constant disturbance; adding an integrator — which is precisely a model of a constant-disturbance generator — drives the error to zero. The regulator succeeds exactly to the extent it contains a model of what it must reject.
Good Regulator Theorem is not Requisite Variety because the theorem bounds the regulator's internal model (the right picture of when to deploy which response), whereas requisite variety bounds its response repertoire (enough actions to match disturbance variety); coupled but distinct lower bounds.
Good Regulator Theorem is not Feedback because feedback is the mechanism of routing output back to input, whereas the theorem is the stronger claim that successful feedback regulation against a complex environment entails an internal model.
Good Regulator Theorem is not Homeostasis because homeostasis is an outcome — a variable held near a setpoint — whereas the theorem is a necessity claim about the internal structure any regulator achieving that outcome must satisfy.