Databricks and ontology
I’m repeating myself, but only because I care 😊
The genie is out of the bottle. Databricks just announced “Genie Ontology.” Spoiler, It is not an ontology.
Here’s what it actually is, per their own description: an automatic context layer that extracts snippets from tables, queries, dashboards, pipelines and connected apps, and organizes them into a graph of “how a company works.” On top of that, a PageRank-style authority score decides which definition of a term wins — based on source reputation, usage frequency, and freshness. That’s a metadata catalog with trust ranking. A very good one, probably. But it has nothing to do with what the word “ontology” means in knowledge representation.
A real ontology gives you formal semantics: classes, properties, disjointness, cardinality constraints — a logical layer that lets you infer facts that were never explicitly stated, and that lets you check consistency. Genie Ontology gives you the opposite move: when two business units disagree on what “active customer” means, it doesn’t reconcile the definitions or flag the contradiction. It picks a winner by popularity and recency.
That’s a fine engineering answer to “which definition should the chatbot use today.” It is not semantic interoperability. It’s social consensus dressed up as formal meaning, solving the trust problem, not the meaning problem. This is becoming a pattern. Google did it with the Open Knowledge Format (OKF), calling structural interoperability a knowledge graph (wrote about it earlier). Now Databricks is doing it with “ontology” for what is, functionally, a RAG grounding layer with a popularity contest attached.
I get why the word gets used, “ontology” signals rigor and sells better than “weighted metadata graph.” But precision matters here. If your system can’t tell you why two definitions conflict, only which one more people happen to use, you don’t have an ontology. You have a very well-organized opinion poll.
🧞♀️ Genie