What is an Ontology?

If you've been following along in my 'What Is' posts, we’ve covered knowledge graphs and tackled triples – so naturally, the next question is: What’s an ontology?

It’s a term that's becoming increasingly common, so let’s break it down practically and clearly.

🔵 Schemas: The Starting Point

Let’s begin where most data folks are comfortable: schemas.

In a database or spreadsheet, a schema is like your column headers. It defines your data structure – what you can store and how it fits together. Imagine capturing information about people: you might have columns like "name," "age," and "height." A schema is essentially a blueprint: precise, defined, and – crucially – closed.

Why closed? Traditional databases follow the Closed World Assumption: if something isn’t explicitly in the schema, the system assumes it doesn’t exist. If there’s no column for "favourite sci-fi author" (mine’s Iain M. Banks, by the way), then the database assumes no one has one.

🔵 Ontologies: Schemas That Think for Themselves

Now, flip that logic. Instead of assuming the data is complete, imagine a system accepting data as evolving – just because something isn’t known yet doesn’t mean it isn’t true. This is the Open World Assumption.

That’s where ontologies come in.

Ontologies are like supercharged schemas. They don’t just describe data structures; they define what the data means and how those meanings relate.

🔵 Properties That Float

In traditional schemas, tables own their columns. A row in a "Person" table has a "name" column.

In an ontology, this is reversed. Properties are free-floating. Facts such as "makes a quacking sound", "lives in a pond", or "waddles while walking" help infer what the type is – as the saying goes: if it walks like a duck and quacks like a duck, it's a duck!

🔵 Hierarchies, Inheritance, and Description Logics

Ontologies also include built-in structures like inheritance:

🔹 A `Doctor` is a kind of `Person`
🔹 A `Person` is a kind of `Animal`
🔹 A `Startup` is a kind of `Company`

These aren’t just labels – they’re logical constructs. If animals have a respiration rate, then dogs, cats, and doctors all inherit this property. A doctor is a person who is an animal and, therefore, has a respiration rate.

This structure is formalised through description logics – formal languages used in AI and knowledge representation. These logics power ontology reasoning, enabling systems to make logical inferences from partial information. While LLMs are 'natural' models operating in continuous vector spaces, ontologies are 'formal' models working in discrete graph spaces.

🔵 Why Ontologies Matter Right Now

Data isn't just data anymore – it’s increasingly powering AI systems that need to understand the world, not merely index it.

Ontologies bridge the gap between raw data and intelligent behaviour. They allow machines to reason about categories, infer missing facts, all while preserving meaning in a formal way that you can trust.