1
/
12
Family Resemblances: A Solution for Knowledge Interoperability
By G. Sawatzky, embedded-commerce.com
Sept 2025
A concise deck that connects a classic idea from philosophy with practical AI-driven interoperability, while preserving internal rigor.
Read the full article
The Problem with Perfect Categories
Real-world concepts often resist rigid taxonomies.
Precision enforced early can create paradoxes, costly mappings.
Enterprise integration rarely finds one-to-one schema matches.
Classic Examples
Tomato: fruit, vegetable, or both, context matters.
PhD student: employee, student, or both, policy matters.
Category A
Category B
Context-dependent overlap
Philosophical Grounding
Wittgenstein: family resemblance, overlapping similarities.
No single trait defines the set, similarity patterns do.
Member
Trait A
Trait B
Trait C
Member
Related Perspectives
Goguen: inexact concepts, degrees of membership.
Sowa: human-centric KR, flexible, yet precise modeling.
AI for Practical Interoperability
LLMs connect terms by usage and context.
Map similar concepts across systems without forced sameness.
Useful for external integration, preserves internal rigor.
Term A
Term B
context + usage
Example: Patient Identifier vs Member ID
Different labels, similar roles in context.
LLM aligns by descriptions, usage, and constraints.
Proposed mapping is explainable, subject to review.
Term mapping
Context-driven mapping
Patient Identifier
Member ID
Similar usage and constraints
Precision Inside, Flexibility Outside
Internal: ORM for unambiguous, human-readable rules.
External: AI similarity for cross-system matching.
Integrity preserved, interoperability improved.
Internal (ORM)
External (AI Similarity)
Unambiguous rules
Constraints
Verbalizations
Context mapping
Similarity scoring
Human review
Share ORM Models as JSON
Publish conceptual elements and verbalizations.
Let AI derive family resemblances across models.
Keep authoritative constraints at the source.
Suggested Workflow
Model precisely with ORM, export JSON + verbalizations.
Use AI to propose mappings, review with experts.
Codify accepted mappings, monitor drift and change.
ORM Model
Export
JSON + Verbalizations
AI Mapping
Proposals
Review
Benefits and Guardrails
Faster integration, fewer brittle mappings.
Explainable proposals, auditable decisions.
Safeguards: constraints, test cases, provenance.
References (Selected)
Wittgenstein, L. (1953). Philosophical Investigations.
Goguen, J. A. (1968). The logic of inexact concepts. Synthese.
Sowa, J. F. (1984). Conceptual Structures. Addison-Wesley.
Prefer prose?
Open the full written article
.
Previous
Next