Control Requires Models

The Good Regulator and its levels

A thermostat holds a room at 21°C through a January night. The furnace fires, the room warms, the furnace cuts out, the room cools, the furnace fires again — and across hours of shifting outdoor temperature, opened doors, and the slow leak of heat through the walls, the reading stays pinned to its target. The device is doing something a rock on the windowsill cannot do: it is keeping a variable where it should be against a world that keeps pushing it elsewhere. Ask how, and the honest answer turns out to be surprising. Buried in that cheap loop of bimetal and relay is a representation of the room. Not a picture, not a sentence, but a structure that maps what the sensor reads onto what the furnace should do — and that mapping tracks how the room actually behaves. Take the representation away and the control goes with it. This is not a fact about thermostats. It is a theorem.

The Good Regulator Theorem

In 1970 Roger Conant and W. Ross Ashby proved a result they titled, with unusual directness, “Every Good Regulator of a System Must Be a Model of That System.” The Good Regulator Theorem states that any regulator capable of achieving reliable control must embody a model of the system it regulates. This is not a slogan dressed up as mathematics. It is a demonstration that control and representation are the same problem seen from two sides.

To regulate a system is to map its observed states onto corrective interventions. The regulator must distinguish states that call for different responses, and it must apply the response that will actually move the system toward its target. Those two demands, taken together, force structure. The regulator’s internal organization has to preserve the distinctions in the system that matter for choosing an action — states the system can be in that require different corrections must land on different internal states of the regulator, or the regulator will respond to both the same way and get at least one of them wrong. In cybernetic terms, effective regulation presupposes a homomorphism between system and regulator: a structure-preserving map from the states of the controlled system to the states of the controller. The regulator succeeds exactly to the degree that this map holds.

The requirement is not optional and not a matter of good engineering practice. It is what control is. A controller that fails to encode a relevant distinction will respond identically to two situations that demand different actions, and will therefore blow one of them. A controller that misrepresents how the system evolves will apply corrections that push the wrong way. Control failure is not bad luck; it is inadequate modeling made visible. Whenever a regulator reliably hits its target across a range of disturbances, you can read off from that success that it contains a model adequate to the range — the reliability is the evidence.

None of this requires the model to look like science. It need not be expressed in equations, held as symbols, or available for inspection. A model in this sense is any physical structure that differentiates among the possible states of the system and selects appropriate transitions. The thermostat’s bimetal strip is such a structure. So is an aircraft autopilot’s encoding of aerodynamic response, which is why it can stabilize flight it was never explicitly told about. So is the enzyme pathway that ramps production up or down with substrate concentration, the homeostatic loop that holds blood pH inside its narrow band, the reflex arc that pulls a hand from heat. Each realizes an implicit model of some regularity in the world, and each earns its keep by anticipating: predictive adequacy, not mechanical twitching, is what separates a regulator that works from one that merely moves.

This is the counterpart, in the register of action, to the claim that understanding is model-mediated. There is no direct grip on the world for action any more than for comprehension. To know the world you need a model of it; to steer it you need a model too, and for the same reason — the structure of what you are dealing with has to be mirrored somewhere in the thing dealing with it. A model, on the anatomy given in the previous chapter, is a structured representation that preserves task-relevant distinctions relative to a purpose. Regulation is that definition put to work: the task is control, the relevant distinctions are the ones that change what action is called for, and the purpose is hitting the target.

This is why the middle criterion of agency is the load-bearing one. An agent, on the account developed for minimal and maximal agents, is a system that is embedded in its environment, carries a predictive model of it, and biases its actions toward preferred outcomes. The Good Regulator Theorem is the formal spine of that middle criterion. It says that the predictive model is not a luxury bolted onto control but its precondition: strip the model and you do not get a simpler agent, you get a reflex — a system that reacts without anticipating, and so cannot reliably steer anything. Regulation reaches from the bacterium comparing sugar gradients to the market pricing a commodity to the nervous system holding a body in homeostasis, and everywhere it reaches, it carries a model with it.

The Tension

Here the argument runs into what looks like a wall. I have just insisted that any regulator embodies a model of what it regulates. But I have also argued, in the theory of what beliefs are, that beliefs are not things agents contain — they are features of models we construct of agents. Agents don’t have beliefs; models of agents do.

Put the two claims side by side and they seem to collide. If every regulating agent must embody a model, and belief is what lives in a model, then surely the thermostat, which embodies a model of the room, believes the room is cold. Either the first claim proves too much — dragging beliefs into every furnace and enzyme — or the second is wrong, and beliefs really are inside agents after all. Something has to give.

Nothing gives. The collision is an equivocation on the word model, and once the two senses are pried apart the tension disappears completely.

Two Levels

There are two entirely different things the word “model” is doing in these two claims, and they operate at two different explanatory levels.

The first is the cybernetic-structural sense. This is the model the Good Regulator Theorem is about: the internal organization that lets a system discriminate states, anticipate outcomes, and intervene appropriately. It is realized in the machinery — in bimetal, in enzyme kinetics, in neural circuitry, in weights and activations. Its role is functional, not propositional. It does not represent the world to anyone; it just preserves the distinctions that make control work. This model is constitutive of the agent: it is part of what the agent physically is, and the agent could not regulate without it. The thermostat has one. So does the bacterium, and so do you.

The second is the intentional-interpretive sense. This is the model an observer constructs of an agent in order to explain and predict its behavior — the intentional stance. Beliefs live here. When I say the mugging victim believes the weapon is real, I am not naming a component that neuroscience could locate in his skull; I am deploying a representation of him as a creature with goals, expectations, and information, because that representation predicts what he will do cheaply and well. This model is not part of the agent at all. It is attributed — an explanatory construct that sits in the interpreter, justified when it earns accurate and economical predictions.

An agent must possess a model in the first sense and need not instantiate one in the second. The thermostat has the constitutive model — that is why it regulates — and lacks the attributed one, because no interpreter gains anything by saying it believes the room is cold rather than simply noting that it maps temperature to switching. The belief-attribution would be idle: it predicts nothing the bare mechanical description does not predict better. So the thermostat models temperature without believing anything, and there is no paradox in saying so. It has structure that mirrors the room’s dynamics; it has no place in an intentional model where beliefs could appear, because nothing is gained by putting it there.

This is why belief-talk switches on somewhere up the scale of agents and not at the bottom. The bacterium and the thermostat regulate, and so they model in the constitutive sense — but a human deliberating over whether to trust a stranger is worth modeling as a believer, because the intentional description of him buys enormous predictive power that no tractable mechanical description could. Belief does not turn on because a threshold of internal complexity is crossed inside the agent. It turns on when an interpreter finds the intentional stance the cheapest accurate handle on the agent’s behavior. The property is relational, and it lives on the interpreter’s side of the relation.

Conditionalism makes the division principled rather than convenient. Since every truth claim depends on the model used to interpret it, the question “does this system have a model?” has no answer until we fix which model we mean. Relative to the system’s own functional organization — the background that asks what physically enables its behavior — the thermostat has a model and the claim is true. Relative to an intentional framework — the background that asks what the system believes and wants — the thermostat has nothing, and the belief-attribution is false, or rather empty. Both answers are correct because they are answers to different conditional questions. The word “model” was carrying two conditions at once, and the tension was the shadow of the collision.

So the architecture of agency has two layers that must not be flattened into one. Internal models are constitutive: an agent is, in part, the model it carries, and without it there is no control. External models are explanatory: beliefs are what we write into our picture of an agent to make its conduct intelligible, and they exist in that picture and nowhere else. Understanding and control require the first. Belief arises only in the second. The claims never touched, because they were never about the same thing.

There is a payoff waiting further on. If regulation requires an internal model, and the richest regulator we know is a mind regulating itself, then consciousness ought to fall out as an engineering consequence of that self-regulation — the system building and consulting a model of its own states in order to control them. That is exactly the bet made by the Modeler-Schema account of consciousness, where the Modeler is a regulator turned on the agent itself. The theorem that forces a model into every thermostat is the same theorem that, followed up the ladder of agents, arrives at a mind watching itself.