Control Requires Models

Series: Epistemology / Control Theory
Companion: Understanding Requires Models

Core Thesis

Control presupposes representation. Any regulator capable of reliable control must embody a model of the system it regulates. Not rhetorical—precise demonstration that control and representation are inseparable. Unless controller preserves distinctions present in system, cannot consistently select appropriate actions.

Source: Good Regulator Theorem (Conant & Ashby, 1970)

“Regulation, in this sense, is an applied form of model-based cognition.”

The Good Regulator Theorem

Statement

Good Regulator Theorem (Conant & Ashby, 1970):

Any regulator capable of achieving reliable control must embody a model of the system it regulates.

Formal Insight

If system occupies range of distinguishable states:

Regulator succeeding across that range must encode at least the distinctions relevant to deciding actions
Must contain representation sufficient to map states → interventions
Mapping must preserve system’s causal structure

Not optional: Inadequate internal modeling → control failure

Control Failure Mechanism

When regulator lacks proper distinctions:

Responds identically to states requiring different interventions
Applies inappropriate corrections
Cannot reliably achieve goals

Cause: Inadequate internal modeling

Regulation and Representation

What Regulation Requires

To regulate a system:

Map observed states → corrective interventions
Distinguish states calling for different responses
Predict consequences of actions

Requirements = structured internal mapping mirroring relevant system structure

Cybernetic Formulation

Homomorphism required:

Regulator’s internal organization must preserve distinctions necessary for reliable action
Between system and regulator
Structural correspondence

Model Definition (Broad)

Not necessarily equations or scientific model

Any representational structure that:

Differentiates among possible system states
Identifies appropriate transitions
Maps states to interventions

Examples:

Enzyme pathways (biological)
Homeostatic loops
Neural reflexes
Scientific models
All exhibit implicit environmental/internal dynamics models

Control as Representational Task

Core Process

Regulator constructs/maintains:

Mapping: perceived conditions → actions
Actions expected to bring system closer to target
Presupposes predictions about system evolution under interventions

Key Distinction

Predictive adequacy (not mechanical reaction) distinguishes effective regulation

Mechanism:

Internal structure reflects system regularities
Anticipates responses
Selects appropriate interventions

Domain Examples

1. Aircraft Autopilot

Must encode aerodynamic behavior
Maps sensor readings → control surface adjustments
Stabilizes flight via model of aircraft dynamics

2. Thermostat

Maps temperature readings → heating/cooling decisions
Based on environment dynamics model
Simple but embodies system structure

3. Central Nervous System

Integrates sensory data with internal states
Maintains homeostasis
Complex distributed model of body/environment

4. Markets (Distributed Regulation)

Encode decentralized information
Preferences, constraints, opportunities
Emergent regulatory model

5. Political Systems

Falter when regulating without sufficient informational structure
Need adequate model of governed system
Control failure from model inadequacy

6. AI Agents

Must construct/acquire models of human behavior
To act coherently with respect to human values
Alignment = adequate human model

Pattern: Successful control derives from internal structure reflecting regulated system regularities

Axio Integration

Agency as Model-Based Regulation

Axio characterization:

Agency = capacity to select actions based on expectations
Expectations = structured anticipations of world response
Good Regulator Theorem aligns perfectly

Requirements for Agency:

Model of environment
Model of task
Model of own possible interventions

Without model: Action reduces to blind reaction (lacks intentional behavior structure)

Conditionalism Connection

Parallel structure:

Understanding	Agency
Empirical truth conditional on interpretive background	Agency conditional on representational structure
Knowledge model-mediated	Regulation model-based
Coherence of model determines understanding	Coherence of model determines regulation

Unified principle: Both comprehension and control require models

Framework: Regulation occurs within model; goals, errors, corrections defined relative to structure

Broader Implications

For Biology

Evolution of internal models: Survival depends on regulation
Organisms encode environment/self-dynamics
Natural selection for modeling capacity
Homeostasis = regulatory success

For Economics

Markets as distributed regulators: Function by encoding decentralized information
Price signals = model components
Market failure = model inadequacy
Information aggregation essential

For Politics

Governance requires adequate models: Cannot regulate without informational structure
Failed states = inadequate system models
Centralized planning problems = model complexity limits
Distributed knowledge coordination

For AI Alignment

Must model human values: To regulate toward them
Alignment problem = adequate human model problem
Cannot control what you can’t model
Coherent action requires representation

For Philosophy of Mind

Intentionality requires models: Blind reaction ≠ intentional action
Purpose presupposes anticipation
Beliefs/desires as model components
Mental representation necessary for agency

Companion Thesis

Understanding Requires Models (Previous Essay)

Comprehension: Necessarily model-mediated

No direct world access
Representations organize perception, guide expectation

Control Requires Models (This Essay):

Effective action: Necessarily model-based

Regulation inseparable from representation
Control = applied model-based cognition

Combined: Both understanding AND action require models

Unified Framework

Reality (inaccessible directly)
    ↓
Models (representational structures)
    ↓
Understanding (model-mediated comprehension)
    ↓
Control (model-based regulation)
    ↓
Effective Action (intentional behavior)

Key Insights

Control inseparable from representation: Not optional
Models enable regulation: Encode relevant distinctions
Homomorphism required: Internal structure mirrors system
Predictive not reactive: Anticipation distinguishes effective control
Broad definition: Even simple mechanisms embody models
Universal requirement: Biology, markets, politics, AI all need models
Agency presupposes models: Intentional action = model-based
Conditionalism applies: Regulation conditional on framework
Control failure = model failure: Inadequate representation → regulation breakdown
Applied cognition: Regulation is model-based cognition in action

Good Regulator Theorem: Formal basis (Conant & Ashby)
Cybernetics: Study of regulation and control
Homomorphism: Structure-preserving mapping
Understanding Requires Models: Companion thesis
Conditionalism: Framework dependence
Control Theory: Mathematical regulation framework
Predictive Processing: Brain as prediction machine
Homeostasis: Biological regulation
Agency: Model-based action selection

Key Quotes

“Any regulator capable of achieving reliable control must embody a model of the system it regulates.”

“Control and representation are inseparable: unless a controller preserves the distinctions present in the system, it cannot consistently select appropriate actions.”

“A model, in this context, is any representational structure that differentiates among possible states of the system and identifies appropriate transitions.”

“Control failure is therefore a consequence of inadequate internal modelling.”

“Predictive adequacy, not mechanical reaction, is what distinguishes effective regulation.”

“Without such a model, action reduces to blind reaction, lacking the structure necessary for intentional behaviour.”

“Just as empirical truth is conditional on interpretive background, agency is conditional on representational structure.”

“Control presupposes representation. To influence a system reliably, an agent must encode the distinctions pertinent to that system’s behaviour.”