[The book is available on Amazon and the O’Reilly digital platform.]
Related readings:
- KEOPS Kernel
- Ontological Prisms for beginners
- Ontological Prisms & The Geometry of Knowledge
- The Engineering of Meanings
- Vocabularies, Conversations, Contexts
- Bounded Contexts & Semantics
Preamble
Ontologies and knowledge graphs have gradually emerged as the necessary symbolic counterparts to AI agents, which are prone to recurring errors and blunders. Ontologies, developed from domain-specific thesauri and taxonomies, are meant to be fused into knowledge graphs. The question is how to do so without creating knowledge chimeras.
Graphs & Hyper Graphs
Graphs are the Swiss Army knife of symbolic representation. They have served operational research, system management, thesauri and databases, and they now serve ontologies and knowledge graphs. Their basic structure of nodes and edges can represent objects and references, states and transitions, or processes and controls. And since a node or an edge can itself stand for a subgraph, graphs can also represent nested structures.
Knowledge Graphs
Addressing knowledge, graphs implementation can be grouped in two families, Resource Description Framework (RDF) and Labeled Property Graph (LPG). Although they differ in terminology and purpose, both build on nodes and edges; LPG adds labels to categorize them and properties to describe both. RDF can nevertheless match that expressiveness by reifying edges. Whether through LPG labels and edge properties or through RDF types and reification, graphs support layered knowledge representation.
Layers can thus be used to distinguish between enterprise level knowledge and ontologies.
Layers can thus separate enterprise-level knowledge from ontologies. With ontological prisms, further labels can characterize each node by its epistemic dimension.
Ontologies & Knowledge Graphs
The rapid advance of agentic AI has brought to light the risks of misunderstanding—an understatement—and with them the benefits of ontologies. While the path from thesauri and taxonomies to ontologies is now well established, most approaches treat the move from ontologies to knowledge graphs as a matter of semantics alone. They overlook the conceptual and organizational issues involved in turning specific knowledge into shared knowledge.
Merging Ontologies
Once semantic issues such as homonymy have been addressed, simply merging ontologies may seem the obvious option, but simple merges come with a caveat: they ignores possible overlaps. Even setting aside the complexity that comes with scale, manual curation would be a labor-intensive and endless task. Outsourcing it to AI agents would lead to a kind of inbreeding, since these knowledge graphs exist precisely to support those agents. A third option is to draw on the knowledge already built into the ontologies themselves.
Epistemic Dimensions
To keep knowledge alive, representations must be diachronic and follow changes over time. Timestamps, however, won’t do: at scale, they would turn representations into a spaghetti mess. Using the epistemic dimensions defined by ontological prisms serve two purposes: they reduce complexity by an order of magnitude, and they make the life cycle of knowledge explicit and manageable.
Anchored Identities
Ontologies define conceptual and organizational spaces in which elements are identified (#). These identities serve as anchors and must be preserved through mergers. Aspects (≈) are different: they get their meaning from what they are attached to. As a result, they can be attached to a conceptual or organizational space, but need not be.
Shared Knowledge
Unlike aspects, which are defined independently but get their meaning from the contexts in which they are used, facts, concepts, and categories are identified by ontologies. It follows that knowledge crossings allowed for functional associations between homogeneous elements, and simple references otherwise.
Alternatively, ontologies can be managed as a hypergraph, with ontological connectors linking them.
In that case, connectors defined between ontologies also apply between their elements, under the conditions set above.




