AI Glossary / Graphs and symbolic AI / Intermediate

Entailment

Definition

Entailment is a relationship in logic: a set of statements entails another statement when every interpretation that makes the starting statements true also makes the conclusion true. Written KB ⊨ q, it means that query q follows necessarily from knowledge base KB under the chosen logical semantics. It is stronger than saying that q is plausible or often observed.

What it means

A knowledge base restricts which possible situations remain compatible with its assertions. A conclusion is entailed when it holds in all of those remaining situations. If even one compatible situation makes the conclusion false, the entailment does not hold.
This is a statement about meaning and truth across interpretations. An inference procedure is an algorithm that attempts to establish such relationships. Keeping those ideas separate helps explain why a correct conclusion can exist even when a particular program cannot find a proof within its resource limit.

How it works

In a small propositional problem, enumerate truth assignments, retain those satisfying the knowledge base, and inspect the query in each retained assignment. More scalable approaches use logical transformations, satisfiability solving, or proof procedures appropriate to the language.
One useful classical test asks whether KB together with the negation of q is unsatisfiable. If no interpretation can make that combination true, there is no counterexample to the entailment. A sound proof procedure derives only entailed conclusions. A complete procedure can, subject to its precise formal setting and resource assumptions, establish every entailment in its scope. These are properties to specify, not automatic features of any rule engine.

Three practical examples

• A configuration assistant derives that an adapter is required from explicit compatibility rules and the selected device type.
• A scheduling prototype checks whether every timetable satisfying its constraints places a review before publication.
• A knowledge-base editor searches for a counterexample to a proposed conclusion before adding it to an explanation shown to users.

Think of it this way

Think of considering every arrangement that fits a puzzle's clues. A statement follows from the clues if it is true in every fitting arrangement. The analogy captures necessity rather than guesswork. Its limitation is that formal interpretations can be numerous or infinite, and logical rules require exact meanings that everyday puzzle descriptions may leave implicit.

Worked example

Let P mean that a package is marked fragile, and Q mean that it requires protective packing. The knowledge base contains P and P implies Q. Consider the four assignments to these two propositions.
When P is false, the assignment fails the asserted fact P. When P is true and Q is false, it violates P implies Q. Only P true and Q true satisfies both statements. Therefore the knowledge base entails Q.
Now remove the fact P and retain only P implies Q. The assignment P false, Q false satisfies the implication but makes Q false, so Q is no longer entailed. Add Q alone instead: the knowledge base still does not entail P, because P false, Q true is compatible. This exposes the mistake of reversing an implication. Nothing in the example establishes that fragile packages are the only ones requiring protective packing.

Illustrative case study

In a fictional catalog project, the Alder team stores the rule that a premium package includes a review appointment. Its assistant sees that an order includes a review and labels the order premium. An editor challenges the explanation.
The team constructs a counterexample: a standard package with a separately purchased review. That situation fits the original rule but contradicts the assistant's conclusion. The team changes the query logic and documents the additional information needed to classify the package. This is an invented teaching case about rule interpretation, with no claimed deployment results.

Common misconceptions and mistakes

An unproved claim is not automatically false. The knowledge base may entail neither a statement nor its negation. A failed or timed-out search also does not by itself demonstrate non-entailment.
In classical logic, an inconsistent knowledge base has no satisfying interpretation and therefore entails every statement under the standard definition. That is a reason to inspect consistency before treating derived answers as meaningful; alternative logics handle inconsistency differently. Finally, entailment preserves what follows from the premises. It does not independently verify that the premises accurately describe the real world.

Business use

Entailment provides a clear standard for explanations in rule-based systems: identify the premises, their intended meanings, and the reasoning that connects them to the answer. It is especially useful when a conclusion must be reproducible from a stated policy or technical specification.
Operational teams should distinguish confirmed facts, modeling assumptions, and defaults. Defaults and exceptions may require a different reasoning framework from ordinary classical entailment. A natural-language assistant's confident explanation should not be presented as a formal proof unless an appropriate verification process actually established the relationship.

People also ask

Is entailment the same as correlation?

No. Correlation describes statistical association; logical entailment concerns necessity across interpretations satisfying the premises.

Does entailment require checking every interpretation one by one?

No. Proof procedures and satisfiability methods can establish it without direct enumeration.

Is a derived answer guaranteed true in reality?

Only if the premises and modeling assumptions correctly apply. Logical consequence alone cannot supply missing factual accuracy.

Check your understanding

A knowledge base contains only “If a file is archived, it is read-only” and “The file is read-only.” Does it entail that the file is archived?

Answer: No. A non-archived file that is read-only for another reason satisfies both statements. That counterexample defeats the proposed entailment.

Related terms

• Knowledge base
• Interpretation
• Model
• Satisfiability
• Completeness
• Implication
• logical consequence
• soundness
• counterexample

Learn next

Learn next about satisfiability. Take the package example, add the negation of the proposed conclusion, and identify whether any truth assignment remains possible. Then replace an asserted fact with an uncertain observation and explain why a probability model asks a different question.

• Probability model
• Satisfiability

More in Graphs and symbolic AI

• Empirical loss - Empirical loss is the loss measured on a finite set of observed examples, usually calculated as the average of a per-example loss function.
• End-to-end learning - End-to-end learning trains a system's learnable processing stages together using an objective tied to its final output.
• English auction - An English auction is an ascending-price auction in which participants compete through increasing bids, or remain active as a displayed price rises.
• Environment class - An environment class is a collection of possible environments used to specify, study, or evaluate an intelligent agent.
• Epigraph - The epigraph of a function is the set of points on or above its graph.
• Epsilon-ball - An epsilon-ball is a neighborhood of points within a specified distance epsilon of a chosen center.

Sources

Russell and Norvig, original Logical Agents chapter, semantics and entailment: https://aima.cs.berkeley.edu/newchap07.pdf Russell and Norvig, original algorithm collection, truth-table entailment and resolution: https://aima.eecs.berkeley.edu/algorithms.pdf

Previous: Ensemble learning Next: Enterprise bus matrix