The Type-Instance Confusion Problem

Both AI and human knowledge engineers struggle with a fundamental modeling decision: is this entity a type (class) or an example (instance)?

In ontology research, entities like “Fighter” or “Wizard” were often modeled as subclasses when they should be instances. AI-generated ontologies showed consistently low instance counts, they default to creating class hierarchies rather than populating them with individuals.

The distinction seems trivial but cascades through the entire knowledge structure. Getting it wrong produces ontologies that can describe categories but can’t answer questions about specific things. The competency questions that tested ontology completeness failed on AI-generated structures precisely because they asked about instances that weren’t there.

Related: 06-molecule—ontology-design-patterns