First-order logic

Table of Contents

Propositional logic is limited in a very significant way: for every individual object or idea we want to describe, we have to create a new variable. We have no way of saying all objects of some type share certain properties.

We'll introduce a few new logical constructs, and arrive at what is called first-order or predicate logic.

Constant symbols

An constant symbol is a single object, like "my cat Tony" or "President Obama." Note that propositional logic does not have constant symbols that represent objects; instead, propositional logic only has statements like "President Obama is a Democrat."

We use upper-case for constant symbols: my cat Tony is \(Tony\), President Obama is \(Obama\), etc.


We'll also add new relations among objects that are called predicates. A predicate is true or false. So, the predicate \(IsACat\) is true when apply to Tony: \(IsACat(Tony) = T\). On the other hand, \(IsACat(Obama) = F\).

Predicates may have more than one argument: \(Eats(Tony, CatFood) = T\), \(TastesLike(YellowCandy, Banana) = T\), or \(ParentsOf(Isabelle, Penelope, Rene) = F\).


Sometimes we want to find out for what values of, say, the second argument of a predicate, makes the predicate true. For example, for what values of \(x\) make this true: \(Eats(Tony, x)\)

It turns out that when \(x\) equals any of \(CatFood, Bugs, CatNip, MashedPotatoes\) (etc.) then the predicate is true.


We have the same operators as before: \(\wedge\), \(\vee\), \(\neg\), \(\rightarrow\), \(\leftrightarrow\). Thus, we can write \(Fish(Jenny) \rightarrow HasGills(Jenny)\). If this is true, then if Jenny is a fish, then she has gills.


We can combine variables, predicates, and relations to make statements like "all fish have gills" and "at least one cat eats mashed potatoes." Two new symbols are introduced to build these statements: \(\forall\) (read "for all") and \(\exists\) (read "exists").

For example, \((\forall x)(Fish(x) \rightarrow HasGills(x))\) is read "for all \(x\), if \(x\) is a fish then \(x\) has gills." That statement may be true or false; it depends on whether it is true that all fish have gills (we'll see how to answer that question in the next section).

Another example: \((\exists x)(Cat(x) \wedge Eats(x, MashedPotatoes))\), which is read "there exists an \(x\) such that \(x\) is a cat and \(x\) eats mashed potatoes." Again, we need to ask, how do we know if this is true?

Possible worlds (semantics)

How do we know if \((\forall x)(Fish(x) \rightarrow HasGills(x))\) is true? It depends from what "world" we pull our \(x\) values. A "world," or "possible world," is simply a set of objects and truth-values for predicates. In one possible world, the set of objects may be \(Tony\), \(Jenny\), \(Obama\), \(Isabelle\), \(Penelope\), \(Rene\), \(YellowCandy\), \(Banana\), etc. The truth values for the predicates are \(Cat(Tony) = T, Fish(Tony) = F, Cat(Jenny) = F, Fish(Jenny) = F, Eats(Tony, Jenny) = F, Eats(Jenny, Tony) = F\), etc. Every predicate must be true or false for every object or combination of objects (in other words, no predicate's truth for any inputs is "unknown").

Now if we want to know if some statement is true, we just use the following rules:

  • If the statement has a \(\forall\) in front, we just check if every object makes the statement true.
  • If the statement has a \(\exists\) in front, we just check if there is at least one object that makes the statement true.

Notice that the negation of a \(\forall\) statement is an \(\exists\) statement and vice versa:

$$\neg (\forall x) (P(x)) \equiv (\exists x) (\neg P(x))$$

$$\neg (\exists x) (P(x)) \equiv (\forall x) (\neg P(x))$$

Building a knowledge database

Here is a very simple (and common) example of a database of first-order logic statements:

  • \((\forall m)(\forall c)(Mother(m, c) \leftrightarrow (Female(m) \wedge Parent(m, c)))\)
    • (mothers are female parents and vice versa)
  • \((\forall f)(\forall c)(Father(f, c) \leftrightarrow (Male(f) \wedge Parent(f, c)))\)
    • (fathers are male parents and vice versa)
  • \((\forall x)(Male(x) \leftrightarrow \neg Female(x))\)
    • (males are not females and vice versa, and there are only two sexes; this statement recalls our discussion about the "violence" of knowledge engineering)
  • \((\forall x)(\forall y)(Brother(x, y) \leftrightarrow Brother(y, x))\)
    • (being a brother is a symmetric relationship)
  • \((\forall x)(\forall y)(\forall c)((Parent(x, c) \wedge Brother(x, y)) \leftrightarrow Uncle(y, c))\)
    • (uncles are brothers of one's parent and vice versa)
  • \((\forall c)((\forall x)(\forall y)((Parent(x, c) \wedge BiologicalParent(y, c)) \rightarrow x \neq y) \leftrightarrow Adopted(c))\)
    • (adopted children all (both) of their biological parents different from their "regular" (everyday) parents)

And so on…

We also need to know about our objects and predicates. Our world is:

  • \(\{Frank, John, Kathy, Susan, Gabrielle, Male(Frank) = T, Male(Susan) = F, \dots\)
  • \(BiologicalParent(John, Gabrielle) = T, Father(Frank, Gabrielle) = T, \dots\)
  • \(Parent(Kathy, Gabrielle) = T, BiologicalParent(Susan, Gabrielle) = T, \dots \}\)

Inferencing with a knowledge database

Let's make a query: Is Gabrielle adopted? \(Adopted(Gabrielle)\)? How do we answer this question using our knowledge database?

First we look at the statements in the database and ask, which ones refer to adoption? Only the last statement listed above refers to adoption; it says that for Gabrielle to be adopted, all objects \(x\) and \(y\) where \(x\) is Gabrielle's parent and \(y\) is Gabrielle's biological parent, must have the further property that \(x \neq y\) in order for Gabrielle to be adopted. If, after checking all objects in our world, the relations described hold, then Gabrielle is adopted.

Declarative vs. procedural knowledge

First-order logic provides a way of writing declarative knowledge. But when we ask a query, we need some kind of procedure for determining if the query is true, or alternatively, finding objects (to put in place of the variables) to make a query true. There are several procedures that can perform this task. They are essentially all search procedures. Actually, we can apply depth-first search (say), to evaluate queries. Prolog will do this search for us, as long as we properly provide the declarative knowledge we care about.

As an example, consider the following knowledge database:

  • \((\forall x)((List(x) \wedge Empty(x)) \rightarrow Sorted(x, x))\)
    • (an empty list is a sorted list; the \(Sorted(a, b)\) predicate means \(b\) is the sorted version of \(a\))
  • \((\forall x)(\forall y)(\forall z)((List(x) \wedge FirstElement(y, x) \wedge RestOfList(x, z) \wedge Empty(z)) \rightarrow Sorted(x, x))\)
    • (a list with one element is a sorted list)
  • \((\forall x)(\forall x')(\forall y)(\forall y')(\forall z)(\forall z')((List(x) \wedge Permutation(x, x') \wedge FirstElement(y, x')\)

    \(\wedge RestOfList(z, x') \wedge Sorted(z, z') \wedge FirstElement(y', z') \wedge LessThan(y, y'))\)

    \(\rightarrow Sorted(x, x'))\)

    • (\(x'\), a permutation of a list \(x\), is the sorted version \(x\) when the first two elements of the list \(x'\) are \(y\) and \(y'\), and \(z'\) is the sorted remainder of the list (which includes \(y'\) at the front), and \(y < y'\))

This is the declarative knowledge about sorted lists. The query \(Sorted([5, 2, 6, 3, 4], x')\) should find the value for \(x'\) that makes the query true. Finding this \(x'\) is finding the sorted version of \([5, 2, 6, 3, 4]\), i.e., sorting the list. An inefficient way to find this \(x'\) is to try all possible permutations of \(x\), that is, consider all \(x'\) that satisfy the predicate \(Permutation(x, x')\) (we are assuming the world contains the sorted permutations of all known lists, and may well contain many more permutations of those lists).

Intro to AI material by Joshua Eckroth is licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported License. Source code for this website available at GitHub.