Cybernetic modeling theorem
Good Regulator Theorem
Every effective and sufficiently simple regulator of a system must embody a model of that system in the sense required to produce successful regulation.
effective regulation requires an internal model of relevant system behavior
Conant and Ashby proved a result under explicit assumptions about optimal regulation and mappings among disturbances, outcomes, and regulator actions. The theorem does not say the model must be conscious, symbolic, complete, or stored in one place.
A repeating disturbance enters the plant. The regulator predicts only the distinctions represented by its model; unmodeled variation passes into the output as residual error.
(%)
The animation runs automatically, pauses on the conclusion, and then repeats. The main control changes the scenario rather than scrubbing the timeline.
- CHANGE
- Internal-model fidelity
- WATCH
- residual disturbance
- MEANING
- A repeating disturbance enters the plant. The regulator predicts only the distinctions represented by its model; unmodeled variation passes into the output as residual error.
A regulator can cancel only the disturbance structure it can represent.
Prediction travels through an internal model before the compensating action reaches the plant, while mismatch remains visible at the controlled output.
What it actually says
The theorem connects successful regulation to information. To choose an action that keeps outcomes in an acceptable set, a regulator must preserve the distinctions among system states and disturbances that matter for that choice.
A thermostat has a minimal model implicit in its switching relation. A skilled operator carries a richer learned model. A feedback controller may distribute its model across sensors, state estimators, parameters, and dynamics. The relevant question is not whether a diagram looks like the plant, but whether the regulator maps situations to effective actions.
"A useful law compresses a pattern. It does not erase the conditions that make the pattern true."
How the idea developed
The modern form emerged through observation, argument, and later refinement. The timeline separates the first insight from the version now used in textbooks and practice.[1]
W. Ross Ashby develops requisite variety as a foundation for regulation.
Roger Conant and Ashby publish Every Good Regulator of a System Must Be a Model of That System.
Model-based and adaptive control formalize state estimation and plant identification.
Robotics, operations, biology, and AI use internal models for prediction and control.
How the pattern works
The relation becomes useful only when its mechanism, measurement process, and operating range are visible.
Sensors and memory preserve differences relevant to action.
The regulator estimates how the plant will respond.
A model links observed conditions to compensating actions.
Residual mismatch updates the estimate or triggers correction.
Conant and Ashby proved a result under explicit assumptions about optimal regulation and mappings among disturbances, outcomes, and regulator actions. The theorem does not say the model must be conscious, symbolic, complete, or stored in one place.
Where it earns its keep
Applications are strongest when the law changes a decision, measurement, model, or experiment rather than merely providing an analogy.
Design state estimators and observers
ApplicationUnmeasured state can be reconstructed from a plant model and sensor history.
Model uncertainty must remain explicit.
Build operational representations
ApplicationQueues, inventories, and failure modes guide effective intervention.
A dashboard is useful only when it supports the right actions.
Interpret adaptive regulation
ApplicationOrganisms can embody predictive structure in physiology and learned behavior.
Avoid implying conscious representation.
Where it stops working
The original result depends on the formal setup, the selected loss function, and assumptions about optimality. Real regulators can be satisficing, redundant, adaptive, or constrained.
A highly detailed model is not automatically a good regulator. Complexity, delay, estimation error, computation, and robustness can make a simpler model perform better.
"The controller must copy every detail of the plant"
Better: Only distinctions relevant to successful regulation are required."A model must be explicit software"
Better: Physical structure and learned policy can embody a model."More model detail always improves control"
Better: Uncertainty, delay, and overfitting can reduce performance."The theorem proves consciousness requires a world model"
Better: That philosophical claim is outside the theorem.Sources and further reading
Original publications and serious secondary scholarship are prioritized over summaries.
- Conant and Ashby - Every Good Regulator of a System Must Be a Model of That SystemThe original 1970 theorem and proof.https://doi.org/10.1080/00207727008920220
- W. Ross Ashby - An Introduction to CyberneticsFoundational treatment of regulation and variety.https://archive.org/details/introductiontocy00ashb
- Francis and Wonham - The Internal Model Principle of Control TheoryRelated formal result in control theory.https://doi.org/10.1016/S0005-1098(76)80106-6
- MIT OpenCourseWare - Feedback SystemsOpen engineering context for feedback and models.https://ocw.mit.edu/courses/6-302-feedback-systems-spring-2007/