The Language of

First-order Logic
Lecture 2
Declarative language
Alphabet
Grammar
Notation
Variable scope
Semantics
The simple case
Interpretations
Denotation
Satisfaction
Rules of interpretation
Entailment defined
Why do we care?
Knowledge bases
A formalization
Knowledge-based system
Types of Reasoning
 Deductive Reasoning
Definition: Involves deriving specific conclusions from
general rules or facts. In KRR, this means using a
formal set of logical rules to infer new knowledge
that is logically entailed by the existing knowledge
base.
Example: If the system knows "All birds can fly" and
"Tweety is a bird," it deduces that "Tweety can fly.“
Common Use: Deductive reasoning is implemented in
rule-based systems, logical programming, and
inference engines.
 Inductive Reasoning
Definition: Involves deriving general rules from
specific examples or observations. In KRR, systems
generate hypotheses or patterns based on a set of
specific data points.
Example: If a system observes that "Tweety, a bird,
flies" and "Polly, a bird, flies," it might induce the
rule "All birds can fly.“
Common Use: Inductive reasoning is used in machine
learning, pattern recognition, and data mining to
generalize from training data.
 Abductive Reasoning
Definition: This is reasoning to the best explanation.
Abductive reasoning in KRR helps infer the most likely
cause or explanation from incomplete or partial data.
Example: Given that the system observes that "Tweety
does not fly," and it knows that "Birds usually fly," it
abduces that "Tweety might be a penguin" (i.e., an
exception to the rule).
Common Use: Abductive reasoning is useful in
diagnostic systems, natural language understanding,
and hypothesis generation.
 Non-monotonic Reasoning

• Definition: A form of reasoning where adding new

information can invalidate previous conclusions. In KRR, this
allows systems to revise or withdraw conclusions when new
facts are introduced.
• Example: The system originally concludes "Tweety can fly"
based on the rule "All birds can fly." If the system later learns
"Tweety is a penguin," it withdraws its earlier conclusion, as
penguins are an exception to the rule.
• Common Use: Non-monotonic reasoning is common in
systems dealing with incomplete knowledge, such as belief
revision systems, commonsense reasoning, and dynamic
environments.
 Default Reasoning

• Definition: A type of non-monotonic reasoning where

conclusions are drawn based on default rules, which
hold unless contradictory information is introduced.
• Example: The system might assume by default that
"Tweety can fly" (since most birds can), but if it later
learns "Tweety is a penguin," it overrides the default
conclusion.
• Common Use: Default reasoning is used in situations
where systems must make assumptions about the world
in the absence of complete information (e.g., expert
systems).
 Probabilistic Reasoning

• Definition: Involves reasoning under uncertainty using

probability theory. In KRR, this means systems compute
likelihoods based on probabilistic models and make
decisions accordingly.
• Example: Given that "80% of birds can fly" and "Tweety is a
bird," the system might infer that "There is an 80% chance
that Tweety can fly."
• Common Use: Probabilistic reasoning is used in Bayesian
networks, decision-making under uncertainty, and in various
AI applications involving uncertain or incomplete knowledge
(e.g., speech recognition, recommendation systems).
 Fuzzy Reasoning

• Definition: Involves reasoning with degrees of truth rather

than binary true/false values. In KRR, fuzzy logic enables
systems to reason about imprecise or vague concepts.
• Example: Instead of saying "Tweety can fly" is either true or
false, fuzzy reasoning might assign a degree of truth, such
as "Tweety has a 0.7 likelihood of being able to fly."
• Common Use: Fuzzy reasoning is applied in control
systems, decision-making processes with uncertain inputs,
and modeling of human-like reasoning (e.g., fuzzy
controllers in appliances).
 Analogical Reasoning

• Definition: Involves reasoning by analogy, where the

system draws conclusions based on similarities
between different domains or concepts.
• Example: If a system knows that "Tweety is a bird
and can fly" and encounters a new entity "Flappy,"
which is also a bird, it might infer by analogy that
"Flappy can fly."
• Common Use: Analogical reasoning is used in case-
based reasoning systems, which solve new problems
by drawing on solutions to similar past problems.
 Commonsense Reasoning

• Definition: Involves using general, everyday knowledge to

make inferences that seem natural to humans but can be
complex to formalize. Commonsense reasoning is about
dealing with real-world knowledge that is often implicit or
assumed.
• Example: The system might infer that "If Tweety is inside a
cage, Tweety cannot fly," based on the commonsense
understanding of physical barriers.
• Common Use: This type of reasoning is challenging for AI
but is a goal of systems aiming to achieve human-like
intelligence (e.g., commonsense knowledge bases like Cyc).
 Temporal Reasoning

• Definition: Involves reasoning about events in

time, including the order, duration, and
dependencies between events.
• Example: If the system knows "Tweety was
born before Polly," it can infer "Tweety is older
than Polly."
• Common Use: Temporal reasoning is used in
planning systems, scheduling, and reasoning
about time-dependent events.
 Spatial Reasoning

• Definition: Involves reasoning about spatial

relationships between objects. This kind of
reasoning allows systems to infer properties about
physical space and object interactions.
• Example: The system might reason that "If Tweety
is in the kitchen and the kitchen is next to the living
room, then Tweety is near the living room."
• Common Use: Spatial reasoning is used in robotics,
geographic information systems (GIS), and
navigation systems.
Each of these reasoning types plays a crucial role
in how intelligent systems make decisions,
solve problems, and interact with the world
based on the knowledge they have. By
integrating different reasoning approaches,
KRR systems can better handle complex,
dynamic, and uncertain environments.
Model-Finding and Satisfiability
In Knowledge Representation and Reasoning
(KRR), model-finding and satisfiability are
fundamental concepts used to determine
whether a set of logical statements or knowledge
is consistent and whether specific conclusions or
goals can be achieved within the system. These
concepts are essential for ensuring that the
knowledge base can be effectively used for
inference and decision-making.
 Satisfiability (SAT)

• Definition: Satisfiability is the problem of

determining whether there exists an interpretation
(or model) that makes a given logical formula true.
In KRR, a formula is said to be satisfiable if there is
at least one assignment of truth values to its
variables that makes the entire formula true.
• Importance in KRR: Ensuring that a knowledge base
is satisfiable is crucial because an unsatisfiable set
of statements (i.e., a contradiction) would make it
impossible to draw meaningful inferences.
 Key Aspects of Satisfiability

• Boolean Satisfiability Problem (SAT): This is a

central problem in logic and computer science,
where the goal is to determine whether a Boolean
formula can be satisfied. A Boolean formula is
satisfiable if there is some assignment of truth
values (true/false) to its variables that makes the
entire formula evaluate to true.
– Example: Given a formula like (A∨B)∧(¬A∨C), is there
an assignment of true/false values to A, B, and C that
makes the formula true?
• First-order Satisfiability: In the context of first-
order logic (used in KRR), satisfiability is about
determining whether there is an interpretation (a
model) that satisfies a given set of logical formulas
(e.g., relationships and properties defined over
objects).
Example: In a knowledge base, you might want to
check if there exists a way to interpret "All humans
are mortal" and "Socrates is a human" in a way that
doesn't lead to contradictions.
 Application of Satisfiability in KRR

• Knowledge Base Consistency: A knowledge

base is consistent if it is satisfiable, meaning
there are no conflicting or contradictory
statements.
• Planning and Problem Solving: In AI planning,
systems check whether there exists a
sequence of actions that leads from an initial
state to a goal state. This can be modeled as a
satisfiability problem.
 Model-Finding

• Definition: Model-finding is the process of finding a

concrete model (an interpretation or assignment of
values) that satisfies a given set of logical formulas
or constraints. A model is an interpretation of the
variables and relations in the knowledge base that
makes all the statements true.
• Importance in KRR: In KRR, model-finding helps
identify valid interpretations of the knowledge base,
showing how the abstract concepts and rules apply
in specific cases or scenarios.
 Key Aspects of Model-Finding

• Model: A model is a structure that interprets the symbols in a

logical formula and makes the formula true. For example, a model
for first-order logic would define what the objects in the domain
are and how predicates and functions apply to those objects.
– Example: If the knowledge base contains "All humans are mortal" and
"Socrates is a human," a model would define an interpretation where
"Socrates" is part of a domain of humans and "mortal" is a predicate
that holds for all humans in that domain.
• Model-Finding Algorithms: These are algorithms used to search
for models of logical systems. In propositional logic, model-
finding corresponds to finding truth assignments that satisfy a
formula. In more complex logics (e.g., first-order logic), models
include interpretations of predicates, functions, and constants.
 Application of Model-Finding in KRR

• Automated Theorem Proving: Model-finding is used to verify the

validity of logical statements by finding counterexamples or models
that satisfy certain premises.
– Example: If you want to prove that a conclusion follows from a set of
premises, model-finding techniques can search for models where the
premises hold. If no model can be found where the premises are true but
the conclusion is false, the argument is valid.
• Constraint Satisfaction Problems (CSPs): Model-finding is also
widely used in CSPs, where the goal is to find assignments of
variables that satisfy a set of constraints (e.g., scheduling, Sudoku).
• Knowledge Base Querying: Model-finding is involved when querying
a knowledge base. The system searches for models where the query
holds true given the current facts and rules in the knowledge base.
Relationship between Satisfiability and Model-Finding in KRR

• Satisfiability is a precursor to model-finding. Before trying to find a

model, a system must determine whether a model can exist, which
means checking the satisfiability of the knowledge base or logical
formula.
• Model-finding goes a step beyond satisfiability by not only determining
whether a solution exists but also by identifying one or more specific
models (interpretations) that satisfy the logical constraints.
• Examples in KRR:
• Satisfiability: If you have a rule-based system and want to check if the
rules are consistent (i.e., they don't lead to contradictions), you check for
satisfiability.
• Model-Finding: If you want to see how the rules in the knowledge base
apply in a specific scenario or case, you use model-finding to discover a
valid interpretation that satisfies the rules.