AI produces better ontologies when driven by competency questions and validation loops than when given raw data structures.
Adapted from @Connected_DataShould AIs Construct Your Ontology? Bad ontologies rarely come from bad intentions. They come from starting in the wrong place: a spreadsheet, a database schema, a JSON file's field labels. Whatever structure the source data happened to have gets baked in, cardinality hacks and all, and it fossilises the moment someone builds on top of it. The alternative is starting from competency questions instead of classes. A competency question is simply something a working system should be able to answer: "Which employees report, directly or indirectly, to a given manager?" Classes and properties come afterward, as the machinery needed to answer the question, not the starting point. This is where AI earns a real role. Not as a shortcut around modelling judgment, but as a disciplined executor of it: 🔹 Generate a candidate model from the competency questions 🔹 Check whether it actually answers them 🔹 Flag what it misses, and repeat, exhaustively Kurt Cagle walks through a full worked example for a corporate training platform: six competency questions, an OWL ontology, SHACL shapes with every constraint traced back to the question that justified it, SPARQL queries run and validated, and a deliberately broken record to show what a shapes violation looks like and why it should fail. The conclusion: an AI given a spreadsheet will reproduce the same bad patterns a human would, just faster. An AI given a rigorous set of competency questions and a validation loop is a different tool entirely. The advantage was never intelligence. It's discipline, applied at a scale most teams don't have the billable hours for. https://t.co/S4U04YUMV0 #OntologyDesign #SHACL #EnterpriseArchitecture #DataModeling #KnowledgeArchitecture -- Connected Data London 2026 | 11–12 November | Leonardo Royal Hotel London Tower Bridge 🎤 Share your work with the world's most passionate data community. The Call for Submissions is open. https://t.co/LYWPyNlfy1 🎟 Tickets on sale now. Advance rate discounts up to 15%. https://t.co/A5DC4llVKv 📺 Sponsorship opportunities available. Contact info@connected-data.london for details. #KnowledgeGraph #GraphRAG #Ontology #Graph #AI #DataScience #GraphDB #SemTech
