Semantic Gap

The Semantic Training Gap: Ontology-Grounded Tool Architectures for Industrial AI Agent Systems

This paper identifies and formalizes the "semantic training gap" in LLM-based AI agents deployed in manufacturing, where statistical fluency lacks grounded operational understanding. This gap causes incorrect outputs and compounding failures like semantic drift, which the proposed ontology-grounded architectures aim to address.