AGISystem2 represents knowledge in a high-dimensional vector space (default: 1024 dimensions). These dimensions are partitioned into two fundamental categories that separate what is from what ought to be.
The Two Partitions
| Partition | Default Range | Purpose | Content Type |
|---|---|---|---|
| Ontology | Dimensions 0–255 | Facts about the world | Categories, physical properties, spatial relations, causation |
| Axiology | Dimensions 256–383 | Values and norms | Permissions, prohibitions, ethical judgments, preferences |
Ontological Dimensions
The ontology partition encodes factual knowledge about reality:
Category Dimensions
- Taxonomic hierarchy: IS_A relations between concepts (dog IS_A mammal)
- Part-whole structure: PART_OF / HAS_PART relations (wheel PART_OF car)
- Equivalence: EQUIVALENT_TO between synonymous concepts
- Disjointness: DISJOINT_WITH for mutually exclusive categories
Physical Property Dimensions
- Temperature: hot, cold, warm (as concept regions)
- Size: large, small, medium (as concept regions)
- Material: solid, liquid, gas (as concept regions)
- Location: spatial containment and proximity
Causal Dimensions
- Causation: CAUSES / CAUSED_BY relations
- Temporal ordering: before, after, during
- Requirements: REQUIRES / REQUIRED_FOR dependencies
Axiological Dimensions
The axiology partition encodes normative knowledge – values and rules:
Deontic Modalities
- Permissions: PERMITS / PERMITTED_BY – what is allowed
- Prohibitions: PROHIBITS / PROHIBITED_BY – what is forbidden
- Obligations: REQUIRES in deontic context – what must be done
Compliance Dimensions
- Conformity: COMPLIES_WITH – adherence to rules
- Violation: VIOLATES – breach of norms
Value Dimensions
- Ethical valence: good, bad, neutral regions
- Priority: important, trivial regions
- Risk: safe, dangerous regions
Why Separation Matters
The ontology/axiology separation enables critical capabilities:
1. Bias Control via Veil of Ignorance
The system can mask axiological dimensions during reasoning to ensure decisions are based purely on facts:
# Query ignoring value judgments
@mask MASK_PARTITIONS ontology
@result ASK_MASKED $mask Alice QUALIFIES_FOR scholarshipThis implements Rawls' "veil of ignorance" – decisions made without knowing protected attributes.
2. Auditability
Separating facts from values makes it explicit where normative judgments enter reasoning:
- Ontological assertions can be verified against reality
- Axiological assertions can be traced to their source (law, policy, ethics)
3. Contextual Reasoning
Different theory layers can apply different axiological content while sharing the same ontology:
# Same facts, different normative frameworks
@_ THEORY_PUSH name="EU_GDPR_context"
@_ THEORY_PUSH name="US_HIPAA_context"Working with Dimensions in Sys2DSL
Masking by Partition
# Create a mask for ontology-only reasoning
@ontMask MASK_PARTITIONS ontology
# Create a mask for axiology-only reasoning
@axiMask MASK_PARTITIONS axiology
# Query with specific partition focus
@result ASK_MASKED $ontMask Water IS_A liquidMasking by Specific Dimensions
# Mask specific conceptual dimensions
@tempMask MASK_DIMS temperature pressure
@locMask MASK_DIMS location spatial_proximity
# Query ignoring temperature-related dimensions
@result ASK_MASKED $tempMask Ice IS_A solidCombining Masks
# Combine multiple masks
@combined MASK_UNION $ontMask $tempMask
@result ASK_MASKED $combined Water BOILS_AT Celsius100Configuration
Partition boundaries are configurable in config.json:
{
"dimensions": 1024,
"ontologyPartition": { "start": 0, "end": 255 },
"axiologyPartition": { "start": 256, "end": 383 }
}The remaining dimensions (384–1023) are available for:
- Domain-specific extensions
- Temporal context
- Session-specific information
- LSH indexing overhead
Technical References
- DS(/core/vector_space.js) – Vector operations
- DS(/reason/bias_control.js) – Bias masking
- DS(/support/config.js) – Configuration
- Wiki: Conceptual Spaces
- Wiki: Axiology
- Wiki: Ontology