Models, Beliefs, and Agents

Series: Synthesis / Philosophy of Mind
Synthesizes: Understanding Requires Models, Control Requires Models, The Nature of Beliefs

Apparent Tension

Claim 1: Understanding and control require models (agents must embody models)
Claim 2: Beliefs are properties of models we construct OF agents (agents may lack beliefs)

Question: Do these conflict?

Answer: No. Tension dissolves by distinguishing two explanatory levels of “models”

Resolution: Two Types of Models

1. Internal Models (Structural-Cybernetic)

Function: Structure required for regulation
Nature: Internal representational structures enabling understanding/control
Basis: Good Regulator Theorem
Form: May be molecular pathways, neural circuits, dynamical couplings (not necessarily symbolic/conceptual)
Role: Functional (not propositional)
Support: Prediction, discrimination, context-appropriate intervention
Status: Constitutive of agency (agent MUST possess)

2. External Models (Intentional-Interpretive)

Function: Observer’s framework for explaining agent behavior
Nature: Interpretive attributions we make
Basis: Intentional stance (Dennett)
Form: Conceptual/linguistic (beliefs, desires, intentions)
Role: Explanatory (not constitutive)
Support: Prediction and understanding of behavior
Status: Explanatory construct (agent need NOT instantiate)

Key Distinctions

Same Word, Different Levels

Aspect	Internal Model	External Model
Where	Inside agent	In observer’s framework
What	Functional organization	Conceptual attribution
Purpose	Enable action	Explain behavior
Necessity	Required for agency	Optional interpretation
Form	Any encoding of distinctions	Propositional attitudes
Status	Constitutive	Descriptive

Example: Thermostat

Internal model: Mapping from temperature readings → actions (encodes environmental dynamics)

No belief attribution needed: We don’t need to say thermostat “believes room is cold”

Yet model exists: Functional structure regulates effectively

Biological Organisms

Sophisticated regulatory mechanisms: Often without representing internal models as beliefs

Example: Immune system models pathogens without believing anything

Regulation without belief: Possible and common

Conditionalism Integration

Conditionalism accommodates this naturally:

Truth claims about agents depend on background model:

Describing internal dynamics: Model = system’s functional organization
Characterizing in cognitive terms: Model = observer’s intentional framework

Both are models, different representational levels

Philosophical Framework

No Conflict Between Claims

Agents must have internal models (for understanding/control)
Agents need not have beliefs (observer constructs)
Both true simultaneously (different levels)

Representational Layers

Reality
    ↓
Agent's Internal Model (functional)
    ↓
Agent's Behavior
    ↓
Observer's Model of Agent (intentional)
    ↓
Belief Attributions

Clarifies Architecture of Agency

Action enabled by: Internal functional models
Action explained by: External intentional models
Agency requires: Former (not latter)
Interpretation uses: Latter (to understand former)

Implications

For Philosophy of Mind

Belief attribution is interpretive strategy (not ontological claim)
Functional organization sufficient for agency
Propositional attitudes optional
Multiple description levels compatible

For AI Systems

Can have sophisticated internal models without beliefs
Lookup tables regulate without believing
LLMs may model without propositional attitudes
Functional success ≠ mental states

For Cognitive Science

Internal regulatory structures ≠ explicit beliefs
Representation broader than propositional thought
Many levels of modeling
Intentional stance tool (not discovery)

Understanding Requires Models
Control Requires Models
The Nature of Beliefs
Good Regulator Theorem
Intentional Stance (Dennett)
Conditionalism
Levels of Explanation

Key Quotes

“Internal models are constitutive of agency; external models are explanatory constructs. An agent must possess the former but need not instantiate the latter.”

“By distinguishing representational structures that enable action from interpretive constructs that explain action, we preserve the coherence of the broader philosophical framework and clarify the architecture of agency.”