AGISystem2 – Semantic Libraries

Research Overview

Goal: Create a package ecosystem for domain knowledge—like npm for facts and rules. Enable knowledge reuse across applications.

Key insight: Just as software libraries enable code reuse, semantic libraries enable knowledge reuse with formal guarantees.

1. The Vision: Knowledge as Packages

Core Idea: Package domain knowledge (concepts, relationships, rules) as versioned, distributable units that applications can depend on and extend.

2. Library Structure

Each semantic library contains: type hierarchy declarations, core concepts, relationship definitions, inference rules, integrity constraints, and validation tests.

2.1 Package Manifest

{
  "name": "@sys2/biology",
  "version": "1.2.0",
  "description": "Biological concepts, taxonomies, and rules",
  "keywords": ["biology", "taxonomy", "species", "genetics"],
  "dependencies": {
    "@sys2/core-relations": "^2.0.0",
    "@sys2/units": "^1.0.0"
  },
  "sys2": {
    "entryPoints": ["00-types.sys2", "01-concepts.sys2"],
    "exports": {
      "concepts": ["Species", "Gene", "Protein", "Cell"],
      "relations": ["isA", "partOf", "encodedBy", "expressedIn"]
    },
    "testSuite": "tests/biology.test.sys2"
  }
}

3. Example Libraries

🧬

@sys2/biology

Biological concepts and taxonomies

Taxonomic hierarchies (Kingdom → Species), genetic concepts (Gene, Protein, DNA), cellular biology, physiological systems.

📊 2,400 concepts 🔗 5,800 relationships 📜 340 rules ✅ 98% test coverage

⚖️

@sys2/law

Legal concepts and reasoning rules

Legal entities (Person, Corporation), rights and obligations, contractual relationships, jurisdictional rules.

📊 1,200 concepts 🔗 3,400 relationships 📜 890 rules ✅ 95% test coverage

🏥

@sys2/medicine

Medical knowledge and diagnostic rules

Symptoms, diseases, treatments, drug interactions, diagnostic pathways, anatomical relationships.

📊 8,500 concepts 🔗 24,000 relationships 📜 2,100 rules ✅ 92% test coverage

4. Composition and Extension

Extending a Library:

// my-veterinary-app.sys2

// Import base libraries
import @sys2/biology
import @sys2/medicine

// Extend with domain-specific concepts
isA VeterinaryMedicine Medicine
isA DomesticAnimal Animal

// Add specialized rules
@r_vet1 Implies (And (isA ?animal DomesticAnimal) (hasSymptom ?animal ?symptom))
                (suggestCheckup ?animal Veterinarian)

// Customize existing rules
@r_vet2 Override @r_med_dosage
        (scaleDosage ?drug ?animal (bodyWeight ?animal))

5. Versioning and Compatibility

Version Change	Type	Compatibility
1.0.0 → 1.0.1	Patch	Bug fixes, no API changes
1.0.0 → 1.1.0	Minor	New concepts/rules added, backwards compatible
1.0.0 → 2.0.0	Major	Breaking changes (renamed concepts, modified rules)

Compatibility Guarantee: Libraries with same major version guarantee:

Concept names remain stable
Existing relationships preserved
Rule semantics unchanged
Query results consistent

6. Quality Assurance

Every semantic library must pass quality gates:

Check	Requirement	Tool
Syntax validation	All files parse without errors	`sys2 lint`
Consistency	No contradictions detected	`sys2 check --consistency`
Completeness	No undefined references	`sys2 check --references`
Test coverage	≥90% concepts tested	`sys2 test --coverage`
Documentation	All public concepts documented	`sys2 doc --check`

7. Distribution and Discovery

# Install a semantic library
sys2 install @sys2/biology

# Search available libraries
sys2 search "medical diagnosis"

# Publish your own library
sys2 publish --access public

# View library contents
sys2 info @sys2/biology --concepts

8. Research Challenges

Ontology alignment: How to merge overlapping concepts from different libraries?
Conflict resolution: What happens when two libraries define contradictory rules?
Namespace management: How to handle concept name collisions?
Incremental loading: Load only needed portions for efficiency
Versioned reasoning: Reason over facts from different library versions

9. Implementation Roadmap

Phase	Features	Status
Phase 1	Basic package structure, local loading	Implemented
Phase 2	Dependency resolution, versioning	In progress
Phase 3	Package registry, discovery	Planned
Phase 4	Conflict resolution, merging	Research

10. Related Resources

Learning from Text – Populating libraries from textbooks
DSL Syntax – Language for defining knowledge
Guides – Creating custom libraries