Lookup Tables and Agents

Series: Agency Theory / Philosophy of Mind
Related: Control Requires Models, The Nature of Beliefs

Core Question

When is a lookup table an adequate model of an agent’s behavior, and what does this reveal about the architecture of agency?

Clarifies distinction between understanding, control, and belief by examining agents whose behavior can be captured by simple condition-action mappings.

Lookup Table Definition

Fixed mapping: Discrete conditions → corresponding actions

No abstraction or compression
Simply specifies which response associated with which observed state
Minimal internal model

The Sphex Wasp: Minimal Controller

Behavior Pattern

Nest provisioning sequence:

Drag paralyzed cricket to burrow entrance
Leave cricket outside
Enter nest to inspect
Retrieve cricket

Rigid invariance:

If cricket moved during inspection → repeats entire sequence
No contextual integration
No internal state maintenance
Can loop indefinitely

As Lookup Table

Condition-action mappings:

IF prey near threshold → drag to threshold
IF prey at threshold AND nest uninspected → enter nest
IF nest inspected → retrieve prey

Minimal state-transition table:

No internal representation of prey/environment/task required beyond mapping
Coherent within ecological niche
Natural selection optimized for specific environment

“The wasp’s behavior is coherent within the ecological niche in which it evolved, and natural selection has furnished a controller that satisfies the demands of that environment without requiring a general model of it.”

Biological Examples of Lookup Tables

1. Bacterial Chemotaxis

Chemical concentration changes → motor behavior
Simple gradient following
No cognitive representation

2. Fixed-Action Patterns (Birds/Fish)

Simple stimulus conditions trigger behaviors
Stereotyped responses
Minimal state tracking

3. Plant Tropisms

Local chemical gradients → differential growth
Distributed sensing
No centralized model

Common features:

Environment supplies limited, well-structured cues
Behavior depends on small number of disjunctions
Natural selection optimizes mappings without requiring cognitive internal model

When Lookup Tables Suffice

Adequate Conditions

Lookup table works when:

Relevant state space is small: Few possible states
Environment is stable: Regularities persist
Behavioral repertoire is rigid: Fixed responses optimal
Distinctions few: Limited relevant variables

Result: Controller reliably succeeds encoding only necessary distinctions

No need for:

Rich internal models
Flexible reasoning
Counterfactual evaluation
Long-horizon planning

When Lookup Tables Fail

Limitations in Complex Environments

Cannot:

Integrate information across time
Evaluate unexperienced contingencies
Adapt to novel contexts
Revise strategies in response to unexpected dynamics

Require Richer Models

Examples needing more:

Human cognition
Markets
Political institutions
Complex ecosystems

Reason: Non-trivial structure cannot be captured by finite rule tables

Alternative: Controllers that construct/update internal models reflecting underlying causal relations

Cybernetic Distinction: Models Without Beliefs

Good Regulator Theorem Application

Theorem: Any reliable regulator must embody model of regulated system

For simple organisms:

Model may reduce to lookup table
Table IS a model: Functionally preserves distinctions necessary for action
Structural correspondence sufficient

Models ≠ Beliefs

Key distinction:

Belief: Concept from intentional stance (observer attribution)
Model: Structural encoding of distinctions
Lookup-table controller: Can regulate without propositional attitudes

Implication: Internal model is structural rather than conceptual

“A lookup-table controller can regulate a process without instantiating anything resembling a propositional attitude.”

Gradient of Agency

Lower Bound: Lookup Tables

Minimal model-based behavior
Regulate effectively in limited, structured environments
Few and fixed relevant distinctions
No beliefs or intentions

Higher Complexity: Rich Models

Complex environments require richer internal models
Support prediction, adaptation, intentional action
Counterfactual reasoning
Temporal integration

Continuum

Lookup Table → Fixed-Action Patterns → Reactive Behaviors → 
Model-Based Control → Flexible Agency → Intentional Systems

Implications for Artificial Systems

Simple Embedded Controllers

Resemble biological lookup tables
Effective in narrow domains
Low modeling burden
No need for sophisticated representations

Large Language Models

Cannot be captured by finite rule set
Flexible generative behavior
Internal representations support generalization
Different category of model
Beyond experienced data

Design Insight

Understanding lookup-table sufficiency clarifies:

When to use simple controllers (narrow, stable domains)
When to invest in rich models (complex, variable environments)
Architecture of agency gradient

Philosophical Implications

For Philosophy of Mind

Behavior ≠ mind: Lookup tables regulate without beliefs
Intentional stance: Observer attribution (not intrinsic property)
Structural vs. conceptual models: Different kinds of representation
Agency gradient: Not binary (has/lacks)

For Cognitive Science

Minimal cognition exists: Lookup tables are edge case
Representation necessary: Even simple controllers model
Flexibility distinguishes: Rich models from simple mappings
Evolution optimizes: To environmental demands (not general intelligence)

For AI Ethics

Simple controllers: No moral status (no beliefs/intentions)
Complex models: Gradual emergence of morally relevant properties
LLMs: Unclear position on gradient
Design choices matter: Model richness affects moral considerations

Good Regulator Theorem: Control requires models
The Nature of Beliefs: Beliefs as observer attribution
Fixed-Action Patterns: Stereotyped behaviors
Intentional Stance: Dennett’s framework
Reactive vs. Deliberative: Control architecture distinction
Model-Based vs. Model-Free: Reinforcement learning categories
Minimal Cognition: Simple biological regulation

Key Insights

Lookup tables ARE models: Minimal but sufficient in narrow domains
Models without beliefs: Structural encoding ≠ propositional attitudes
Agency gradient: From lookup tables to intentional systems
Environment complexity drives: Need for richer models
Natural selection optimizes: To niche (not general capability)
Flexibility distinguishes: Complex from simple controllers
Intentional stance: Observer framework (not intrinsic)
LLMs different category: Generative beyond finite rules
Design implications: Match controller to domain complexity
Lower bound illustrated: What minimal agency looks like

Key Quotes

“When is a lookup table an adequate model of an agent’s behaviour, and what does this reveal about the architecture of agency?”

“Natural selection has furnished a controller that satisfies the demands of that environment without requiring a general model of it.”

“A lookup table can be an adequate model of an agent when the relevant state space is small, the environment is stable, and the behavioural repertoire is rigid.”

“A lookup-table controller can regulate a process without instantiating anything resembling a propositional attitude. Its internal model is structural rather than conceptual.”

“Minimal regulators succeed in narrow domains because the modelling burden is low. More sophisticated agents must construct richer internal models to cope with complex, variable environments.”

“Lookup-table controllers illustrate the lower bound of model-based behaviour.”