Knowledge based Agents

Humans claim that how intelligence is achieved- not by purely reflect

mechanisms but by process of reasoning that operate on internal representation
of knowledge. In AI these techniques for intelligence are present in
Knowledge Based Agents.

Knowledge-Based System in Artificial Intelligence

 An intelligent agent needs knowledge about the real world to make
decisions and reasoning to act efficiently.
 Knowledge-based agents are those agents who have the capability of
maintaining an internal state of knowledge, reason over that knowledge,
update their knowledge after observations and take action. These agents can
represent the world with some formal representation and act intelligently.

Why use a knowledge base?

 A knowledge base inference is required for updating knowledge for an
agent to learn with experiences and take action as per the knowledge.
 Inference means deriving new sentences from old. The inference-
based system allows us to add a new sentence to the knowledge base. A
sentence is a proposition about the world. The inference system applies
logical rules to the KB to deduce new information.
 The inference system generates new facts so that an agent can update the
KB. An inference system works mainly in two rules which are given:

o Forward chaining
o Backward chaining

Various levels of knowledge-based agents

A knowledge-based agent can be viewed at different levels which are given

below:
1. Knowledge level
Knowledge level is the first level of knowledge-based agent, and in this level,
we need to specify what the agent knows, and what the agent goals are. With
these specifications, we can fix its behavior. For example, suppose an
automated taxi agent needs to go from a station A to station B, and he knows
the way from A to B, so this comes at the knowledge level.

2. Logical level
At this level, we understand that how the knowledge representation of
knowledge is stored. At this level, sentences are encoded into different logics.
At the logical level, an encoding of knowledge into logical sentences occurs.
At the logical level we can expect to the automated taxi agent to reach to the
destination B.

3. Implementation level
This is the physical representation of logic and knowledge. At the
implementation level agent perform actions as per logical and knowledge
level. At this level, an automated taxi agent actually implement his knowledge
and logic so that he can reach to the destination.
Knowledge-based agents have explicit representation of knowledge that can
be reasoned. They maintain internal state of knowledge, reason over it, update
it and perform actions accordingly. These agents act intelligently according to
requirements.
Knowledge based agents give the current situation in the form of sentences.
They have complete knowledge of current situation of mini-world and its
surroundings. These agents manipulate knowledge to infer new things at
“Knowledge level”.

knowledge-based system has following features:

Knowledge base (KB): It is the key component of a knowledge-based agent.
These deal with real facts of world. It is a mixture of sentences which are
explained in knowledge representation language.
Inference Engine(IE): It is knowledge-based system engine used to infer new
knowledge in the system.
Actions performed by an agent
Inference System is used when we want to update some information
(sentences) in Knowledge-Based System and to know the already present
information. This mechanism is done by TELL and ASK operations. They
include inference i.e. producing new sentences from old. Inference must accept
needs when one asks a question to KB and answer should follow from what
has been Told to KB. Agent also has a KB, which initially has some
background Knowledge. Whenever, agent program is called, it performs some
actions.

Actions done by KB Agent:

1. It TELLS what it recognized from the environment and what it needs to
know to the knowledge base.
2. It ASKS what actions to do? and gets answers from the knowledge base.
3. It TELLS which action is selected , then agent will execute that action.
Algorithm :
function KB_AGENT (percept) returns an action
KB : knowledge base
t : time ( counter initially 0)
TELL(KB, MAKE_PERCEPT_SENTENCE (percept,t) )
action = ASK(KB, MAKE_ACTION_QUERY (t) )
TELL(KB, MAKE_ACTION_SENTENCE (action,t) )
t=t+1
return action
If a percept is given, agent adds it to KB, then it will ask KB for the best
action and then tells KB that it has in fact taken that action.

The Wumpus World

Knowledge representation, reasoning and planning. Knowledge-Based
agent links general knowledge with current percepts to infer hidden
characters of current state before selecting actions. Its necessity is vital in
partially observable environments.

Problem Statement:
The Wumpus world is a cave with 16 rooms (4×4). Each room is connected
to others through walkways (no rooms are connected diagonally). The
knowledge-based agent starts from Room[1, 1]. The cave has – some pits,
a treasure and a beast named Wumpus. The Wumpus can not move but
 eats the one who enters its room. If the agent enters the pit, it gets stuck
there. The goal of the agent is to take the treasure and come out of the cave.
The agent is rewarded, when the goal conditions are met. The agent is
penalized, when it falls into a pit or being eaten by the Wumpus.
Some elements support the agent to explore the cave, like -The wumpus’s
adjacent rooms are stenchy. -The agent is given one arrow which it can use
to kill the wumpus when facing it (Wumpus screams when it is killed). –
The adjacent rooms of the room with pits are filled with breeze. -The
treasure room is always glittery.

PEAS represents Performance Measures, Environment, Actuators, and

Sensors. The PEAS description helps in grouping the agents.

PEAS Description for the Wumpus World problem:

1. Performance measures:
 Agent gets the gold and return back safe = +1000 points
 Agent dies = -1000 points
 Each move of the agent = -1 point
 Agent uses the arrow = -10 points
2. Environment:
 A cave with 16(4×4) rooms
 Rooms adjacent (not diagonally) to the Wumpus are stinking
 Rooms adjacent (not diagonally) to the pit are breezy
 The room with the gold glitters
 Agent’s initial position – Room[1, 1] and facing right side
  Location of Wumpus, gold and 3 pits can be anywhere, except
in Room[1, 1].
3. Actuators:
Devices that allow the agent to perform the following actions in the
environment.
 Move forward
 Turn right
 Turn left
 Shoot
 Grab
 Release
4. Sensors:
Devices which helps the agent in sensing the following from the
environment.
 Breeze
 Stench
 Glitter
 Scream (When the Wumpus is killed)
 Bump (when the agent hits a wall)

Wumpus World Characterization:

 Partially Observable: knows only the local perceptions

 Deterministic: outcome is precisely specified
 Sequential: subsequent level of actions performed
 Static: Wumpus, pits are immobile
 Discrete: discrete environment
 Single-agent: The knowledge-based agent is the only agent whereas the
wumpus is considered as the environment’s feature.

Propositional logic
Propositional logic is a fundamental aspect of artificial intelligence (AI),
particularly in the areas of knowledge representation, reasoning, and automated
theorem proving. Here's a brief overview of how propositional logic is utilized
in AI
Basics of Propositional Logic

Propositional logic, also known as propositional calculus or Boolean logic,

deals with propositions that can either be true or false. It uses logical
connectives such as:

 AND (∧)
 OR (∨)
 NOT (¬)
 IMPLIES (→)
 IF AND ONLY IF (↔)

Applications in AI

1. Knowledge Representation:
o Declarative Knowledge: Propositional logic is used to represent
facts about the world. Each fact is represented as a proposition.
o Rules: Logical rules can be formulated to describe how different
propositions relate to each other.

2. Automated Reasoning:
o Inference Engines: Systems like rule-based systems use
propositional logic to infer new information from known facts and
rules.
o Theorem Proving: Automated theorem provers check the validity
of logical statements and derive new conclusions.

3. Search Algorithms:
o Propositional logic is used to define the states and transitions in
search problems. For example, in game playing, each possible
move can be represented as a proposition.

4. Expert Systems:
o These systems use a knowledge base of propositional logic
statements and an inference engine to make decisions or solve
problems in specific domains, such as medical diagnosis or
financial advice.

5. Constraint Satisfaction Problems (CSP):

o Propositional logic can express constraints that must be satisfied
for a solution to be valid. CSP solvers use logical inference to find
solutions that meet all constraints.

6. Planning:
 o In AI planning, actions are represented as logical operators that
transform states represented by propositions. Planners use logical
inference to determine sequences of actions that achieve a goal.

Tools and Techniques

 SAT Solvers: These are algorithms that determine if a given

propositional logic formula can be satisfied (i.e., if there is some
assignment of truth values to variables that makes the formula true).
Modern SAT solvers are highly efficient and are used in various AI
applications.
 Resolution: A rule of inference used for automated theorem proving in
propositional logic. It is fundamental to many reasoning systems.
 Truth Tables: Used to systematically explore the truth values of
propositions and their logical connectives.

Example

Consider an AI system designed to diagnose a simple problem with a car. The

knowledge base might include propositions like:

 PPP: The car has fuel.

 QQQ: The battery is charged.
 RRR: The car starts.

The system might include rules like:

 P∧Q→RP ∧ Q → RP∧Q→R (If the car has fuel and the battery is
charged, then the car starts)
 ¬R→(¬P∨¬Q)¬R → (¬P ∨ ¬Q)¬R→(¬P∨¬Q) (If the car does not start,
then either the car has no fuel or the battery is not charged)

Given observations (e.g., the car does not start), the system can use
propositional logic to infer possible causes (e.g., the car might have no fuel or
the battery might be discharged).

Challenges

 Scalability: Propositional logic can become computationally infeasible as

the number of variables grows.
 Expressiveness: Propositional logic is less expressive than predicate
logic (first-order logic), which can handle variables and quantifiers,
allowing for more complex representations.
Propositional logic remains a crucial foundation for many AI systems,
providing a clear and structured way to represent and reason about information.

First Order Logic and Semantics of First order

First Order Logic in Artificial Intelligence is a technique used for knowledge
representation. It is an extension of propositional logic and unlike
propositional logic, it is sufficiently expressive in representing any natural
language construct. First Order Logic in AI is also known as Predicate
Logic or First Order Predicate Logic. It is a robust technique to represent
objects as well as their relationships. Unlike propositional logic, First Order
Logic in Artificial Intelligence doesn't only include facts but also different other
entities as listed below.

1. Objects:
Objects can denote any real-world entity or any variable.
E.g., A, B, colours, theories, circles etc.
2. Relations:
Relations represent the links between different objects. Relations can be
unary(relations defined for a single term) and n-ary(relations defined for
n terms). E.g., blue, round (unary); friends, siblings (binary); etc.
3. Functions:
Functions map their input object to the output object using their
underlying relation. Eg: father_of(), mother_of() etc.

In the subsequent sections, we present the parts of First Order Logic in AI, i.e.,
the syntax and the semantics.

Parts of First Order Logic

First-order logic in Artificial Intelligence comprises two main components,
which are as follows.

1. Syntax:
Syntax represents the rules to write expressions in First Order Logic in
Artificial Intelligence.
2. Semantics:
Semantics refers to the techniques that we use to evaluate an expression
of First Order Logic in AI. These techniques use various known relations
and facts of the respective environment to deduce the boolean value of
the given First Order Logic expression.
We elaborate on the syntax of First Order Logic in AI in the following sections.

Syntax of First Order Logic

Syntax of First Order Logic in Artificial Intelligence represents the rules of
representing any natural language construct in terms of First Order Logic in AI.
This involves the rules to describe any object and the relationships between
different objects. These rules comprise rules for writing constants, variables,
predicates, quantifiers, etc. In the following sections, we present various
examples of the syntax of first-order logic in AI. These rules are further
illustrated in the following sections.

Basic Elements of First-Order Logic

The following table furnishes the basic elements of First Order Logic in
Artificial Intelligence.

Meaning
Element Example
1, 2, A, John, Mumbai, Values that can not be changed
Constant
cat, ....
Can take up any value and can also change
Variables x, y, z, a, b, ....
Defines a relationship between its input
Predicates Brother, Father, >, .... terms

Function sqrt, LeftLegOf, .... Computes a defined relation of input

Connectives ∧, ∨, ¬, ⇒, ⇔ Used to form complex sentences using atomic

Equality == Relational operator that checks equality

Quantifier ∀, ∃ Imposes a quantity on the respective variable

Atomic Sentences
Atomic sentences are the most basic expressions of First Order Logic in AI.
These sentences comprise a predicate followed by a set of terms inside a
parenthesis. Formally stating, the structure of an atomic sentence looks like the
following.

Predicate ( term 1, term 2, term 3,...) Predicate ( term 1, term 2, term 3,...)

One example of an atomic sentence in First Order Logic in Artificial

Intelligence can be Sisters ( Geeta , Seeta ) Sisters ( Geeta , Seeta ).

Another example involving a unary predicate can be Polygon ( Rectangle).

Quantifiers in First-Order Logic

Quantifiers in First Order Logic in AI, as the name suggests, are used to
quantify any entity in a given environment. Quantification refers to the
identification of the total number of an entity that is present in the environment
and satisfies a given expression in First Order Logic in Artificial Intelligence.
Quantifiers enable us to determine the range and scope of a variable in a logical
expression. Two types of quantifiers are stated as follows.

1. Universal Quantifier
2. Existential Quantifier

Universal Quantifier

Universal Quantifier in First Order Logic in AI is a symbol in a logical

expression that signifies that the given expression is true in its range for all
instances of the concerned entity. It is represented by the symbol ∀∀ (an
inverted A). If 𝑥x is a variable, then ∀𝑥∀x is read as "For all x" or "For every
x" or "For each x".

For example, let us take the sentence, "All cats like fish". Let us take a
variable 𝑥x which can take the value of "cat". Let us take a
predicate cat (𝑥) cat (x) which is true if 𝑥x is a cat. Similarly, let us take another
predicate likes (𝑥,𝑦) likes (x,y) which is true if 𝑥x likes 𝑦y. Therefore, using the
universal quantifier ∀∀, we can write

∀𝑥 cat (𝑥)⟹ likes (𝑥, fish ).∀x cat (x)⟹ likes (x, fish ).

This expression is read as "For all 𝑥x, if 𝑥x is a cat, then 𝑥x likes to fish".

Existential Quantifier

An existential Quantifier in First Order Logic in Artificial Intelligence is a

symbol in a logical expression that signifies that the given expression is true in
its range for at least one of the instances of the concerned entity. It is
represented by the symbol ∃∃ (an inverted E). If 𝑥x is a variable, then ∃𝑥∃x is
read as "There exists x" or "For some x" or "For at least one x".

For example, let us take the sentence, "Some students like ice cream". Let us
take a variable 𝑥x which can take the value of "student". Let us take a
predicate student (x), which is true if 𝑥x is a student. Similarly, let us take
another predicate likes (𝑥,𝑦) likes (x,y), which is true if 𝑥x likes 𝑦y. Therefore,
using the existential quantifier ∃∃, we can write

∃𝑥 student (𝑥)∧ likes (𝑥, ice-cream )∃x student (x)∧ likes (x, ice-cream )

This expression reads, "There exists some 𝑥x such that 𝑥x is a student and also
likes ice cream".

Points to Remember About Quantifiers in First-Order Logic

While using quantifiers in writing expressions for First Order Logic in Artificial
Intelligence, we need to keep the following points in mind.

1. The main connective for the universal quantifier ∀∀ is the implication

(⟹⟹).
2. The main connective for existential quantifier ∃∃ is and (∧).
 Forward and Backward Chaining
What is Forward Chaining?

Forward chaining, forward deduction or forward reasoning is a method

involving inference or logical rules (facts) for data extraction. It is a bottom-up
approach performed redundantly to reach the endpoint or goal (decision). It
begins with evaluating existing information, followed by the manipulation
based on the knowledge base. The existing information can be as facts,
derivations, and conditions.

Example

Let us say we have the following:

Fact 1: A dog is up for adoption through person A.

Fact 2: Person B is looking for a dog.

Inference rule: If a dog is up for adoption and someone is looking to adopt it, that person is
free to adopt it.
Here, the decision can be reached as person b can adopt the dog from person A. This is how
forward chaining works to make a decision.

Advantages and Disadvantages of Forward Chaining

The strategies have its own set of advantages and disadvantages:

Advantages

 Suitable to draw multiple conclusions simultaneously

 Higher flexibility than backward chaining
 Reliable for conclusion

Disadvantages

 Time-consuming due to data synchronization

 The fact explanation is unclear

What is Backward Chaining?

Backward chaining, backward deduction or backwards is also a reasoning

method that acts in the reverse direction to forward chaining. The top-down
approach involves using decisions or goals to reach the facts. The backward
chaining is the backtracking process of finding usage in diagnostics, debugging
and prescription.
Example

Let us take the same example.

Decision/Goal: Person B adopts a dog.

Fact 1: A dog is up for adoption from Person A.

Fact 2: Person B is looking for a dog.

Inference rule: If a person wants to adopt a dog, he can if there is any up for adoption.

Here, the inference engine will begin with the goal and look if the conditions are met. If both
conditions are met, the stated decision can be concluded.

Advantages and Disadvantages of Backward Chaining

While backward chaining has its benefits, it also has certain drawbacks, both enlisted below:

Advantages

 Swifter than forward chaining

 Easier process
 Efficiently drives correct solutions

Disadvantages

 Provides single answer

 Less flexibility

 Suitable only if the endpoint is known

 Difficult to execute