Does sharing an ontology mean sharing all the same facts?

No. Agents can use the same meanings while possessing different information about particular instances.

Is a field name enough to reveal its commitment?

Usually not. Definitions, examples, identity rules, and relationships help establish what the field represents.

Can a commitment change?

Yes, but changes can affect mappings, queries, and existing data, so their consequences should be reviewed and documented.

AI Glossary / Graphs and symbolic AI / Intermediate

Ontological commitment

Definition

An ontological commitment is a choice about what kinds of things a representation treats as existing and how its vocabulary refers to them. In AI knowledge representation, this includes decisions about objects, relations, events, or other entities the system can describe. In shared ontologies, it also concerns using agreed terms consistently with their specified meanings and assumptions.

What it means

A business system can represent a purchase as a relationship between a customer and a product, or as a separate transaction object with its own date, price, and status. These choices affect which distinctions the system can express naturally.

The commitment is not simply a list of words. It concerns the conceptual structure behind those words. Two databases may both contain a field called customer while one means a person and another means a billing account. Matching their labels does not establish that they describe the same kind of entity.

How it works

At the representation-language level, a formalism provides particular building blocks. First-order logic, for example, describes objects and relations, while different formalisms make other distinctions explicit. At the domain-model level, designers choose which categories and relationships matter for the intended application.

Definitions and axioms then constrain how the vocabulary is used. A model might distinguish a physical item from a product type, or an employment relationship from a person. Adopting the model means respecting those distinctions when making and interpreting statements.

Ontological commitment differs from what the system currently knows. Agreeing on the meaning of Shipment does not imply that two agents know the same shipments or agree about every shipment's status. Likewise, a commitment about represented entities is different from a degree of confidence in a particular factual claim.

Three practical examples

Sales integration: distinguish a customer person, customer organization, and customer account before combining records from different systems.
Asset tracking: separate a laptop model from each physical laptop so that specifications and repair histories attach to the right entities.
Event analysis: represent each delivery as an event when repeated deliveries between the same locations need separate times and outcomes.

Think of it this way

Think of agreeing on what the pieces in a board game represent before discussing a position. A piece might represent one worker, a whole team, or a task; the interpretation changes what a move means.

The analogy is limited because real ontologies can support multiple levels of detail and may evolve. The important point is that shared names are not enough without shared interpretation of the represented things and relationships.

Worked example

Suppose an initial model stores only Bought(person,product). It records Bought(Ana,Camera). If Ana buys the same product type twice, one record of this relation still cannot distinguish the two purchase occasions. Repeating the same logical assertion does not create a second transaction entity.

A revised model introduces Purchase objects p1 and p2, explicitly treated as distinct. It records buyer(p1,Ana), itemType(p1,Camera), date(p1,June1), and corresponding facts for p2 on July1. The system can now attach a separate refund status or paid amount to each purchase.

This is a change in the model's commitments: purchase occasions are represented as objects with their own properties. It does not prove that two physical cameras exist, because Camera was defined as a product type rather than an individual device. Tracking physical units would require an additional distinction and appropriate identifiers.

Nor does introducing the class Purchase assert that every imaginable purchase occurred. The actual instances and factual claims remain separate. The example demonstrates how a modeling choice enables specific questions while leaving other matters unrepresented; it is not a prescription for one universal commerce ontology.

Illustrative case study

Fictional case: a distributor combines two reporting systems and counts customer accounts as if each account were one organization. An organization with several accounts appears as several separate customers, while a shared account obscures another distinction.

The team documents the entities represented by each source and creates mappings that preserve account and organization separately. It then rewrites its count query around the intended entity. This fictional case shows why conceptual alignment precedes reliable aggregation; it does not report a measured data-quality improvement.

Common misconceptions and mistakes

An ontological commitment is not a claim that a chosen model captures every aspect of reality. It is a deliberate representation for a purpose, with assumptions that should remain visible.

It is not the same as choosing a storage format. Two systems can encode the same conceptual commitments differently, or use identical file formats while disagreeing about meaning.

Minimal commitment does not mean saying nothing useful. It means avoiding unnecessary restrictions that make legitimate uses difficult. Conversely, removing essential distinctions just to make an ontology smaller can undermine its purpose. Scope, meaning, and intended questions should guide the choice.

Business use

For business AI, reviewing commitments is valuable before connecting knowledge graphs, integrating records, or allowing agents to exchange structured claims. Many apparent technical mismatches originate in different assumptions about what a record or relation represents.

A practical review asks what each identifier denotes, which entities are distinct, whether time is represented, and which relationships have precise definitions. It should also record what the model intentionally leaves out. This helps users interpret answers without assuming the system can distinguish concepts that its representation has merged.

Check your understanding

Why does the revised purchase model not automatically prove that Ana owns two distinct cameras?

Answer: It distinguishes purchase events involving a product type. Separate physical devices and ownership would need their own representation and supporting facts.

Related terms

Learn next

Choose a familiar business word such as customer, product, or order. Write two plausible meanings and show how each changes a count or query. Then study ontological engineering to turn the required distinctions into explicit definitions and testable examples.

Ontological engineering

Sources

Russell and Norvig, First-order logic slides: https://aima.cs.berkeley.edu/slides-pdf/chapter08.pdf — representational and epistemological commitments, objects, and relations.

Tom Gruber, It Is What It Does: https://tomgruber.org/writing/bio-ontologies-2000.pdf — consistent use of shared vocabulary and the distinction between common ontology and common knowledge. Domain examples are original; the article uses the AI representation sense.

Previous: Online Store Next: Ontological engineering