AI Glossary / Graphs and symbolic AI / Intermediate

Equality symbol

Definition

The equality symbol, usually written =, states in first-order logic that two terms refer to the same object. For example, Owner(Device7) = Maya says that the owner denoted by the first term is the very object denoted by Maya. This is a claim about identity within an interpretation, not merely similar spelling or shared characteristics.

What it means

Terms are expressions that name or describe objects. Two different expressions can identify the same object, just as a person's name and a unique description can identify one person. Equality allows a knowledge base to record that identity explicitly.
The distinction between expressions and their referents matters. Different names do not automatically denote different objects in ordinary first-order logic. A system that assumes every name is unique must state that additional assumption. Conversely, two objects sharing a label or property are not thereby identical.

How it works

An interpretation assigns referents to names and meanings to function symbols. To evaluate s = t, determine the referent of s and the referent of t, then check whether they are the same object. Equality supports replacing one equal term with the other in ordinary extensional logical contexts.
For example, if a = b and Registered(a), then Registered(b) follows under standard equality semantics. Equality is reflexive, symmetric, and transitive: an object equals itself; identity works in either direction; and identities can be chained. Automated reasoners may use specialized methods to manage those consequences efficiently, but the underlying claim remains identity.

Three practical examples

• A symbolic catalog records that two verified identifiers designate the same product entity and transfers appropriate entity-level facts between them.
• A configuration knowledge base states that the controller assigned to one subsystem is the same controller assigned to another.
• A logic exercise uses unequal terms to require two distinct participants in a relationship, preventing a person from satisfying both roles accidentally.

Think of it this way

Think of two name tags attached to the same object. The printed labels differ, but both point to one thing. This captures reference rather than textual sameness. The analogy has limits because identity in a formal model depends on its defined domain and interpretation; matching real-world records still requires evidence that the tags truly belong together.

Worked example

Suppose a fictional knowledge base contains Alias17 = Item42 and StoredIn(Alias17, Shelf3). Standard equality lets us infer StoredIn(Item42, Shelf3). If it also contains Item42 = FeaturedItem, then Alias17 = FeaturedItem follows by transitivity.
Now add Colored(Item42, Blue) and Colored(Item99, Blue). These two facts do not imply Item42 = Item99: several distinct items can be blue. A model containing two blue objects satisfies both facts without identifying them.
Finally, suppose the knowledge base asserts Alias17 ≠ Item42 as well as their equality. Those statements are inconsistent under the same interpretation and identity semantics. A system should expose the conflicting assertions rather than treat equality as a loose similarity score. The example distinguishes three different issues: aliasing, shared properties, and contradiction.

Illustrative case study

In a fictional inventory prototype, the Hazel team imports labels from two small catalogs. It initially treats matching display names as proof that records represent the same item. This merges two distinct editions whose names happen to match.
The team separates a tentative matching relation from confirmed identity. Only a verified mapping produces an equality assertion in its symbolic model. Editors can then inspect why facts propagate from one identifier to another. The invented case illustrates a modeling distinction; it does not claim that a particular record-linkage procedure guarantees correct identification.

Common misconceptions and mistakes

The equality symbol is not the same as assignment in a programming language. Writing x = 3 in mathematics asserts a relationship; some programming languages use a similar symbol to change a variable's stored value. Logical identity also differs from a numerical approximation such as “within tolerance.”
Do not infer identity from identical properties unless the model supplies a rule that establishes uniqueness. Do not assume distinct names imply inequality. And do not substitute inside quoted text or other non-extensional contexts as though a statement about the characters in a name were a statement about the named object.

Business use

For business knowledge models, equality has strong consequences because it can combine facts previously associated with separate identifiers. That makes the distinction between confirmed identity and probable similarity especially useful in integration work.
Document what the domain objects are: a product design, a physical unit, a document version, and a database row are different possible entities. Equality at one level need not imply equality at another. A reasoner can follow formal identity assertions consistently only when the surrounding data model specifies which objects those assertions concern.

People also ask

Can different names be equal?

Yes. They can denote the same object under an interpretation.

Does equality mean that the written terms are identical strings?

No. The terms may look different while referring to one object.

Can two equal objects disagree on a predicate?

Under standard extensional first-order semantics, equal terms refer to the same object, so substituting one for the other preserves the predicate's truth.

Check your understanding

A knowledge base says Manager(TeamA) = Noor and HasBadge(Noor). Can it infer HasBadge(Manager(TeamA))?

Answer: Yes, under standard equality semantics. Both terms refer to the same object. This inference depends on the asserted identity; a merely similar name would not be enough.

Related terms

• First-order logic
• Term
• Interpretation
• Unique-names assumption
• Predicate (logic)
• identity
• predicate
• substitution
• entity resolution
• equational reasoning

Learn next

Learn next about the unique names assumption. Construct one model where two different constant symbols name the same object and another where they name different objects. Identify which extra equality or inequality assertion rules out each model, then compare identity with a tentative record match.

• Unique-names assumption

More in Graphs and symbolic AI

• 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.
• Equal impact - Equal impact is a fairness objective concerned with the benefits and costs that a decision system creates for affected people.
• Equational unification - Equational unification finds substitutions for variables that make two symbolic expressions equal according to a specified collection of equations.
• Eulerian - A graph is Eulerian when it has a circuit that traverses every edge exactly once and returns to its starting vertex.
• Evidence - Evidence, in probabilistic AI, is the information on which an inference is conditioned.

Sources

Russell and Norvig, original First-Order Logic chapter, equality section: https://aima.cs.berkeley.edu/4th-ed/pdfs/newchap08.pdf Russell and Norvig, original Logical Agents chapter, interpretations and entailment: https://aima.cs.berkeley.edu/newchap07.pdf

Previous: Equality of opportunity Next: Equalized odds