Is this the same as a trained model's explanation?

No. Here interpretation concerns the semantics of a formal representation, rather than an explanation of a learned prediction.

Can a knowledge base leave several possibilities open?

Yes. Several interpretations may satisfy its statements.

Will valid logic correct a wrong business definition?

No. It can reason from the encoded premises, but does not automatically repair their connection to the organization's actual process.

AI Glossary / Governance, ethics and safety / Intermediate

Intended interpretation

Definition

An intended interpretation is the meaning that a knowledge-base designer wants the symbols in a formal representation to have. It links the representation to a particular domain and its objects, properties, and relationships. Logical reasoning can support conclusions about that domain only when the supplied statements correctly describe it under those intended meanings.

What it means

A rule system may contain symbols such as Approved, Available, and CanDispatch. Their spelling does not, by itself, tell the computer what those words mean in a business. A designer must establish whether Approved means approved by a manager, approved by finance, or something else.

The intended interpretation supplies that connection. It is the particular understanding of the domain that the representation is meant to capture, rather than any arbitrary assignment of meanings that fits its formal structure.

This entry uses the knowledge-representation and logic sense. In everyday language, intended interpretation can also mean what a speaker wanted a sentence to convey. Neither meaning should be confused with interpretability of a machine-learning model.

How it works

The designer identifies the domain, assigns meanings to relevant symbols, and writes statements that should hold under those meanings. A reasoning procedure then operates on the formal statements rather than directly inspecting the designer's intentions.

An interpretation that makes every statement in a knowledge base true is called a model of that knowledge base. A logical consequence holds in every such model. Consequently, if the intended interpretation is one of those models, a logical consequence holds there too.

This guarantee is conditional. Incorrect facts or poorly translated rules can break the connection to the intended domain even when the reasoning procedure works correctly. Incomplete information creates a different limit: a fact may be true in the intended situation without following from the knowledge base. The system should not automatically equate failure to establish a claim with establishing that it is false.

Three practical examples

1. An inventory rule uses Available to mean physically present and unreserved. A developer who maps it only to a stock-count field gives the symbol a different operational meaning.

2. A scheduling assistant represents Qualified as having a valid certification for a task. A record of past attendance at training may not satisfy that intended meaning.

3. A knowledge graph distinguishes Owns from Uses. Keeping their intended relationships explicit prevents a borrowed device from being treated as company-owned property merely because employees use it.

Think of it this way

Think of a map legend. A blue line means something because the legend connects that mark to a feature, such as a watercourse. The drawing needs that connection.

The analogy is limited. Formal interpretations specify relationships used to assess truth, while a map legend is usually informal and incomplete. An intended interpretation also concerns whether the represented statements accurately describe the chosen domain, not just whether people recognize the symbols.

Worked example

Consider an original toy knowledge base about one parcel, P. Let S mean that P has a shipping label, R mean that its destination has been reviewed, and D mean that it is ready for dispatch.

The knowledge base contains S, R, and the rule that if S and R hold, D holds. Under the intended interpretation, both checks really have been completed and those checks are sufficient for this simplified dispatch process. D follows from the supplied statements.

Now suppose the database sets R to true when an address is merely entered. The formal derivation still produces D, but R no longer records the intended meaning of destination review. The defect lies in the representation or data mapping, not in the elementary inference.

If R is simply missing instead, D does not follow from these statements alone. That absence would not establish that the parcel actually lacks a reviewed destination.

Illustrative case study

Fictional case: Elm Services builds a rule-based assistant for allocating equipment. Operations staff intend Ready to mean inspected, configured, and safe to issue. The software team initially fills that field whenever an item appears in the warehouse register.

During example reviews, a recently returned item receives a ready status despite awaiting inspection. The team documents the intended meaning, separates the underlying checks, and changes the data mapping and rules. Reviewers then test cases that distinguish presence from readiness.

This invented case illustrates a semantic mismatch. It does not claim that adding a definition proves the whole system correct. The organization still needs reliable observations and procedures for keeping inspection records current as equipment conditions change.

Common misconceptions and mistakes

The intended interpretation is not automatically the only interpretation compatible with a knowledge base. Nor does choosing meaningful English names guarantee that all teams use the same definitions.

A logically valid conclusion can still be unsuitable in practice when a premise fails to describe the intended situation. Conversely, a conclusion that cannot be derived may be true but unsupported by the available statements. These distinctions matter when a system presents uncertainty, requests missing evidence, or explains why a rule applied.

Business use

For businesses using rules, ontologies, or knowledge graphs, documenting intended meanings is part of requirements work. Record what each important symbol includes, excludes, and depends on, with examples.

Bring domain specialists and implementers together to check the mapping between records and concepts. Tests should include cases where similar-looking fields have different meanings. Review changes when policies or processes evolve; a technically unchanged rule can become inappropriate if the organization quietly changes what its input fields represent.

Check your understanding

A rule correctly derives an approval from a field named Verified, but that field means submitted rather than checked. What needs investigation?

Answer: The meaning and mapping of the field. Correct inference cannot compensate for a premise that fails to represent the intended verification condition.

Related terms

Learn next

Next, study interpretation and logical consequence together. Choose one business rule, define its symbols precisely, and create two cases that expose a mismatch between a field's name and its intended meaning.

Sources

David Poole and Alan Mackworth, Artificial Intelligence lecture 12.2, formal semantics and intended interpretations: https://artint.info/slides/ch12/lect2.pdf The author slides were read. This article uses the formal-logic sense while preserving the supplied governance category; examples are original. The supplied AIMA glossary did not expose the specific entry during retrieval, so editorial scope review remains appropriate.

Previous: Intelligent word recognition Next: Intensional Component