Module-1 : Introduction to Artificial Intelligence

1. What is AI?
Artificial Intelligence (AI) is the science and engineering of creating intelligent machines,
particularly computer programs. In simple terms, it's the intelligence shown by machines that
helps them mimic human actions. From a technical standpoint, AI is a set of algorithms that
can produce results without being explicitly programmed for that specific outcome, allowing
machines to think and act rationally.

2. How does AI work?

AI systems function by processing vast amounts of labeled training data, analyzing it for
patterns and correlations. These discovered patterns are then used to predict future states.

The functioning of an AI system is centered around three primary cognitive skills:

• Learning Processes: This is the foundational stage where the AI acquires data and
creates rules, known as algorithms, for turning that data into actionable information.
The system is fed large volumes of labeled training data, which it analyzes to discover
correlations and patterns. For example, a chatbot is trained on millions of text chats to
learn how to converse with humans naturally.

• Reasoning Processes: This principle involves choosing the right algorithm to achieve a
desired outcome. The success of an AI program depends on its ability to apply logic
and select the most appropriate set of rules for a given task. For instance, after learning
patterns from data, the reasoning process decides which pattern is most relevant for
a new, unseen input.

• Self-Correction Processes: AI systems are designed to continually improve their

performance. After making a prediction, the system compares its output with the
correct outcome. If there's an error, it adjusts its internal algorithms to refine its model,
ensuring that future predictions are more accurate. This iterative feedback loop is
fundamental to AI's ability to "learn."

Case Study: How Uber Uses AI to Predict Demand

Uber provides a real-world example of these principles in action.

• Learning: Uber's AI system processes massive amounts of historical and real-time data,
including ride requests, traffic conditions, time of day, and local events. It learns the
patterns of when and where people are most likely to need a ride.

• Reasoning: Based on the learned patterns, the system's reasoning algorithms predict
future demand. For example, it might reason that demand in a business district will
surge around 5 PM on weekdays.
 • Self-Correction & Action: The system uses these predictions to proactively get drivers
on the road and direct them to high-demand areas before the surge actually happens.
If its predictions are off, it self-corrects by adjusting its model based on the new data,
constantly improving its accuracy for the next prediction cycle.

Conclusion In essence, AI works by ingesting large datasets, using learning and reasoning
processes to identify subtle patterns within that data, and applying self-correction
mechanisms to continually refine its predictive models. This allows AI to perform complex,
human-like tasks with increasing accuracy and efficiency.

3.1 Advantages of AI
[Link] well on tasks that uses detailed data.
[Link] less time to perform tasks that needs to process huge volumes of data.
[Link] consistent and accurate results.
[Link] be used 24 X 7.
[Link] tasks by better utilizing resources.
[Link] complex processes.
[Link] downtime by predicting maintenance needs.
[Link] companies to produce new products having better quality and speed.

3.2 Disadvantages of AI
[Link] more cost.

[Link] expertise required to develop and use AI applications.

3. Lack of trained professionals.

4. Incomplete or inaccurate data may result in disastrous results.

5. Lacks the capability to generalize tasks.

4. History of AI:
The history of Artificial Intelligence is a journey of ambitious ideas, significant breakthroughs,
and periods of challenge known as "AI Winters". From its conceptual beginnings in the 1950s
to its current boom, AI has evolved dramatically, driven by advances in computing power and
data availability.
Key Stages and Milestones

The Gestation Period (1943-1955):

• 1943: Warren McCullough and Walter Pitts proposed the first mathematical model for
building a neural network.

• 1950: Alan Turing published his seminal paper, "Computing Machinery and
Intelligence," where he introduced the Turing Test, a method for determining if a
machine is intelligent.

The Birth of AI (1956):

• The term "Artificial Intelligence" was officially coined by John McCarthy at the
Dartmouth Conference. This event is considered the birth of AI as a formal field of
research.

Early Enthusiasm and Great Expectations (1952-1969):

• 1958: John McCarthy developed the AI programming language Lisp.

• 1966: Joseph Weizenbaum created ELIZA, an early natural language processing

program that simulated a psychotherapist and laid the foundation for modern
chatbots.

The First "AI Winter" (1974-1980):

• A period of reduced funding and interest in AI research followed the initial hype.
Progress slowed as the computational limits of the time were reached and early
promises went unfulfilled.

The Boom of Expert Systems (1980-1987):

• 1980: The first successful commercial expert system, R1 (also known as XCON), was
developed by DEC to help configure computer systems. Expert systems, which use a
knowledge base to solve problems like a human expert, became a major focus of AI
research.

The Second "AI Winter" (1987-1993):

• The expert system market collapsed, and funding once again dried up, leading to
another period of stagnation.

The Rise of Intelligent Agents and Machine Learning (1990s-2010s):

• 1997: A major milestone was achieved when IBM's Deep Blue chess computer
defeated the reigning world champion, Garry Kasparov, demonstrating the power of
AI in complex strategic games.
 • 2005: A self-driving car from Stanford, named STANLEY, won the DARPA Grand
Challenge, reigniting interest in autonomous vehicles.

The Modern Era of Big Data and Deep Learning (2011-Present):

• 2011: Apple released Siri, integrating an AI-powered virtual assistant into a

mainstream consumer device for the first time.

• 2014: Amazon launched Alexa, its virtual assistant, further popularizing AI in the
home.

• 2016: Google DeepMind's AlphaGo defeated Lee Sedol, the world champion Go player.
This was a significant achievement because Go is considered far more complex than
chess, and the victory showcased the power of deep learning and neural networks.

Conclusion The history of AI shows a cyclical pattern of excitement, investment, and

periods of disillusionment. However, the recent convergence of massive data availability
and powerful computing resources has propelled AI into a new era of rapid progress,
making it one of the most transformative technologies of our time.

5. Types of AI
Artificial Intelligence systems can be classified into different types based on two primary
criteria: their capabilities (what they can do) and their functionalities (how they are designed
to work). This classification helps in understanding the current state of AI and its future
potential.

5.1 Classification Based on Capabilities

This classification distinguishes AI based on its ability to replicate human intelligence.

1. Weak AI (or Narrow AI): This type of AI is designed and trained to perform a specific,
narrow task. It operates within a limited, pre-defined context and does not possess
general intelligence. The vast majority of AI applications that we use today fall into this
category.

o Examples:

▪ Siri and Alexa: These virtual assistants are programmed to respond to

a wide range of voice commands but cannot perform tasks outside their
defined scope.

▪ Google Search: A highly sophisticated weak AI that excels at indexing

web pages and finding relevant information for user queries.

▪ Self-driving cars: These use AI to navigate roads but cannot, for

example, compose music or diagnose a disease.
 2. Strong AI (or Artificial General Intelligence - AGI): This is a theoretical form of AI that
aims to fully replicate the human brain's cognitive abilities. An AGI system would be
able to understand, learn, and apply its intelligence to solve any problem, much like a
human being. It could perform complex tasks for which it was not specifically trained.

o Example: Currently, AGI exists only in science fiction. A famous example is HAL
9000 from the movie 2001: A Space Odyssey, a machine that could emulate
human intelligence, think strategically, and even exhibit emotions.

5.2 Classification Based on Functionalities

This classification is based on how an AI system functions and its level of "consciousness."

1. Reactive Machines: These are the most basic AI systems. They do not have memory
and cannot use past experiences to inform current decisions. They perceive the world
and react to it directly based on a set of pre-programmed rules.

o Example: IBM's Deep Blue, the chess computer that defeated Garry Kasparov,
is a prime example. It analyzed the current state of the chessboard and chose
the optimal move without any memory of previous games or moves.

2. Limited Memory: These AI systems can retain data for a short period and use it to
make decisions. They can learn from recent past experiences but cannot build a
permanent library of knowledge. Most modern AI applications have limited memory
capabilities.

o Example: Autonomous vehicles use this type of AI. They store and analyze
recent data about the speed of nearby cars, the distance between vehicles, and
traffic signals to navigate safely. However, this data is transient and not
permanently stored.

6. Machine Intelligence
Machine Intelligence, often used interchangeably with Artificial Intelligence, refers to the
ability of a machine or system to exhibit intelligent behavior. A system is considered artificially
intelligent when it can demonstrate one or more core components of intelligence that are
typically associated with the human mind.

Components of Intelligence

Intelligence is not a single entity but a composite of several distinct capabilities. The key
components are:

1. Reasoning: This is the process of making decisions, predictions, and drawing

conclusions from available information. It involves logical thinking. There are two main
types of reasoning:
 o Deductive Reasoning: This is a "top-down" approach that starts with a general
rule or premise and moves to a specific, certain conclusion. If the premises are
true, the conclusion must also be true.

▪ Example: Premise 1: All humans are mortal. Premise 2: Socrates is a

human. Conclusion: Therefore, Socrates is mortal.

o Inductive Reasoning: This is a "bottom-up" approach that uses specific

observations to form a general hypothesis or conclusion. The conclusion is
probable but not guaranteed to be true.

▪ Example: Observation: Every swan I have ever seen is white.

Conclusion: Therefore, all swans are white. (This is a probable
hypothesis, but it can be falsified by observing a single black swan).

2. Learning: Learning is the process of gaining knowledge or skills through study, practice,
teaching, or experience. For an AI, learning involves improving its performance on a
task by analyzing new data. Different forms include:

o Auditory Learning: Learning by listening (e.g., a speech recognition system

learning from audio recordings).

o Observational Learning: Learning by watching and imitating others (e.g., a

robot learning to perform a task by observing a human).

o Motor Learning:

Learning through physical movement of muscles. Example: Learning how to

pick up objects correctly.

o Observational Learning:

Learning by watching and imitating others. Example: Children learn by copying

their parents’ actions.

o Perceptual Learning:

Learning by recognizing things that have been seen before. Example:

Identifying objects and situations based on prior experiences.

o Relational Learning:

Learning by recognizing patterns in relationships between things. Example:

Adjusting the amount of spices in a dish after remembering how much was
used last time.

o Spatial Learning:
 Learning through visual stimuli like images, maps, and colors. Example:
Creating a mental map of a route before actually driving.

o Stimulus-Response Learning:

Learning by reacting to a specific stimulus. Example: Shouting when touching a

hot pan, because it causes pain.

3. Problem Solving: This component involves identifying a problem, analyzing it, and
selecting the best possible solution or path to reach a desired goal. Search algorithms,
covered in Module 2, are a primary method AI uses for problem-solving.

4. Perception: Perception is the process of acquiring, interpreting, selecting, and

organizing sensory information to understand the environment.

o Humans use sensory organs (eyes, ears, etc.).

o AI systems use sensors like cameras, microphones, and radar to acquire data
from their environment. Computer Vision and Speech Recognition are key AI
fields focused on perception.

5. Linguistic Intelligence: This is the ability to use, comprehend, speak, and write
language effectively. It is the foundation of interpersonal communication. In AI, this is
the domain of Natural Language Processing (NLP), which enables machines like
chatbots and virtual assistants to understand and respond to human language.

Conclusion Machine Intelligence is a multi-faceted concept built upon these core

components. For an AI system to be truly intelligent, it must be able to not only process data
but also learn from it, reason about it, perceive its environment, solve problems, and
communicate its findings, much like a human does.

7. Differences Between Human and Machine Intelligence.

The goal of Artificial Intelligence is to simulate human intelligence, the way machines "think"
is fundamentally different from how humans do. Understanding these differences is key to
appreciating both the capabilities and limitations of current AI systems. The primary
distinctions lie in perception, learning, memory, and the nature of intelligence itself.

Key Differences

1. Nature of Intelligence: General vs. Narrow

o Human Intelligence: Humans possess Artificial General Intelligence (AGI). We

have the ability to think abstractly, reason, learn from very few examples, and
apply knowledge from one domain to solve problems in a completely different
one.
 o Machine Intelligence: Current AI systems exhibit Narrow Intelligence. They are
designed for a specific task and excel at it, often surpassing human ability.
However, an AI that masters chess cannot use that knowledge to drive a car or
compose music.

2. Perception and Processing

o Human Intelligence: Humans perceive the world through patterns and

context. Our brains are hardwired to recognize faces, understand scenes, and
interpret ambiguous information based on prior experience and intuition.

o Machine Intelligence: Machines perceive by analyzing data based on a set of

rules and algorithms. For example, a computer "sees" an image as a matrix of
pixel values and uses mathematical models like CNNs to identify features and
objects.

3. Learning and Data Requirement

o Human Intelligence: Humans can learn new concepts and skills from a very
small number of examples, or even just one. We can generalize effectively from
limited information.

o Machine Intelligence: Most advanced AI models, particularly in deep learning,

require massive amounts of labeled data to learn effectively. A facial
recognition AI needs to be trained on thousands or millions of images to
achieve high accuracy, whereas a human child can recognize a cat after seeing
just a few examples.

4. Memory and Recall

o Human Intelligence: Human memory is associative and context-dependent.

We store and recall information through patterns and relationships. Recalling
one memory often triggers related memories.

o Machine Intelligence: Machine memory is based on data storage and retrieval.

It uses searching algorithms to locate specific pieces of information from a
database or knowledge base. It is precise and literal but lacks the rich,
associative context of human memory.

5. Handling Ambiguity and Common Sense

o Human Intelligence: Humans have a deep well of common sense and life
experience, which allows us to effortlessly navigate ambiguous situations and
infer missing information.
 o Machine Intelligence: AI systems lack common sense and struggle with
ambiguity. They interpret instructions literally and can fail in unexpected ways
when faced with a situation not covered in their training data.

Conclusion In summary, human intelligence is general, adaptable, and context-aware, capable

of learning from minimal data. Machine intelligence, in its current state, is narrow, data-
intensive, and rule-based. While machines can outperform humans in specific, well-defined
tasks through sheer computational power, they have yet to replicate the flexible, intuitive, and
holistic nature of human thought.

8. Advantages and disadvantages of Artificial Intelligence in today's world.

Artificial Intelligence is a transformative technology that offers immense benefits across
various sectors, from business and healthcare to everyday life. However, its rapid adoption
also brings significant challenges and disadvantages that must be carefully considered and
managed.

Advantages of Artificial Intelligence

1. Automation of Repetitive and Complex Tasks: AI excels at automating mundane,

rules-based, and high-risk tasks that were previously done by humans. This frees up
human workers to focus on more creative, strategic, and high-value activities.

o Example: In manufacturing, AI-powered robots perform tasks like welding and

assembly 24/7 with higher precision and consistency than humans.

2. Enhanced Speed and Accuracy: AI systems can process and analyze massive amounts
of data far faster and more accurately than is humanly possible.

o Example: In healthcare, AI algorithms can analyze thousands of medical images

(like X-rays or MRIs) in minutes to detect signs of diseases like cancer, often
with greater accuracy than human radiologists.

3. Data-Driven Insights and Decision-Making: By identifying patterns in large datasets,

AI can provide businesses with critical insights into their operations, customers, and
markets, enabling smarter decision-making.

o Example: Retail companies use AI to analyze customer purchasing history to

create personalized product recommendations, which can significantly boost
sales.

4. 24/7 Availability and Consistency: Unlike humans, AI systems can operate

continuously without fatigue, emotion, or bias (assuming the data is unbiased). This
ensures consistent and reliable performance.

o Example: AI-powered chatbots provide round-the-clock customer service,

answering queries and resolving issues at any time of day.
Disadvantages of Artificial Intelligence

1. High Cost of Implementation: Developing, training, and maintaining AI systems is

expensive. It requires significant investment in specialized hardware (like GPUs),
software, and skilled technical talent.

2. Lack of Expertise and Trained Professionals: There is a significant skills gap in the AI
field. The demand for qualified AI experts, data scientists, and machine learning
engineers far outstrips the current supply, making it difficult for many organizations to
implement AI effectively.

3. Risk of Bias and Inaccurate Results: AI systems are only as good as the data they are
trained on. If the training data is incomplete, inaccurate, or contains historical biases,
the AI will learn and amplify those biases, leading to unfair and discriminatory
outcomes.

o Example: An AI hiring tool trained on historical data from a male-dominated

industry might learn to unfairly discriminate against female applicants.

4. Lack of Generalization and Common Sense: Most current AI is "Narrow AI," meaning
it is highly specialized for one task. It cannot easily transfer its knowledge to a different
domain. Furthermore, AI lacks the common sense reasoning and intuition that humans
use to navigate novel situations.

5. Ethical and Societal Concerns: The rise of AI brings concerns about job displacement
due to automation, the potential for autonomous weapons, and the erosion of privacy
through mass data collection and surveillance.

Conclusion While the advantages of AI—such as increased efficiency, accuracy, and data-
driven insights—are revolutionizing industries, it is crucial to address its disadvantages.
Mitigating risks related to cost, bias, and ethical implications is essential for harnessing the full
potential of AI responsibly and for the benefit of society.

9. Agent and Environment in Machine Intelligence.

In the field of Artificial Intelligence, the concepts of the agent and the environment are
fundamental building blocks for designing and understanding intelligent systems. An AI system
is conceptualized as an agent that operates within an environment, perceiving it and acting
upon it to achieve a goal.

The Agent An agent is anything that can be viewed as perceiving its environment through
sensors and acting upon that environment through actuators. In essence, an agent is a
decision-maker. The agent's behavior is governed by an agent function, which maps any given
sequence of perceptions (percepts) to an action.

Types of Agents:
 • Human Agent: A person is a biological agent. Our sensors are our sense organs (eyes,
ears, skin), and our actuators are our body parts (hands, legs, mouth).

• Robotic Agent: A physical robot is a common example. Its sensors include cameras and
infrared range finders, while its actuators are motors and mechanical limbs that allow
it to move and manipulate objects.

• Software Agent (Softbot): A program that operates in a digital environment. Its

sensors are the inputs it receives (e.g., keystrokes, file contents, network packets), and
its actuators are the actions it performs (e.g., displaying text on a screen, writing to a
file, sending network packets).

The Environment The environment is the world or context in which the agent operates. It is
where the agent exists, perceives, and acts. The environment provides the agent with percepts
and is affected by the agent's actions. Environments can be characterized along several
dimensions:

• Observable vs. Partially Observable: If an agent's sensors can detect all aspects of the
environment that are relevant to its choice of action, the environment is observable.
A chess game is fully observable. If some information is hidden, it is partially
observable, like a self-driving car that cannot see around a corner.

• Static vs. Dynamic: A static environment does not change while the agent is deciding
on its next action. A crossword puzzle is static. A dynamic environment can change on
its own, like a busy traffic intersection for a self-driving car.

• Discrete vs. Continuous: A discrete environment has a finite number of distinct states
and actions, like a chess game. A continuous environment has an infinite range of
states and actions, like the speed and steering angle of a car.

• Single-Agent vs. Multi-Agent: An environment with only one agent is a single-agent

environment (e.g., solving a puzzle). A multi-agent environment contains other agents
who may be cooperative or competitive, such as in a game of soccer or a team of
delivery drones.

The Agent-Environment Interaction The interaction is a continuous loop:

1. The agent receives a percept from the environment via its sensors.

2. The agent's internal program or function processes this percept.

3. The agent chooses an action to perform.

4. The agent executes the action using its actuators, which in turn changes the state of
the environment.

5. The loop repeats.

Conclusion The agent-environment model is a powerful abstraction for designing AI systems.
By clearly defining the agent's capabilities (sensors, actuators) and the properties of the
environment it will operate in, we can systematically design rational agents that can act
intelligently and effectively to achieve their goals.

Rationality: An agent is considered rational if it consistently performs the "right thing."

The "right thing" is determined by whether the sequence of states the environment goes
through, because of the agent's actions, is desirable (i.e., leads from an initial state to a goal
state).

Rationality is the ability to make responsible and sensible decisions.

• A rational agent always performs the right action to maximize its performance.
Problem Solved by Agent (PEAS):

• Performance measure, Environment, Actuators, and Sensors (PEAS) are used

to define a problem that an agent will solve.

A rational agent makes decisions that maximize its performance based on:

a. Performance measure (how successful the agent is).

b. Percept sequence (the inputs it has received).

c. Prior knowledge (what the agent already knows about the environment).

d. Possible actions (what the agent can do).

[Link] of agents
An AI agent is any entity that perceives its environment through sensors and acts upon that
environment through actuators. For example, a robotic vacuum's sensors are its cameras and
bump detectors, and its actuators are its wheels and suction motor. Agents are classified
based on how they process information to make decisions.

1. Simple Reflex Agents

These are the most basic types of agents. They choose actions based only on the current
percept, ignoring the rest of the percept history.

• How they work: They operate on a simple "condition-action" rule (an IF-THEN rule).
If a certain condition in the environment is met, they execute a corresponding action.
They do not have any memory of past events.
 • Limitations:

o They have very limited intelligence.

o They are only effective in fully observable environments where the current
percept provides all the necessary information to make the right decision.

o In partially observable environments, they can easily get stuck in infinite loops.

• Example: A simple automated vacuum cleaner that changes direction only when its
bump sensor hits an obstacle. It doesn't remember where it has been; it just reacts to
the current input from its sensor.

2. Model-Based Reflex Agents

These agents are an upgrade to simple reflex agents because they maintain an internal state
or "model" of the world to handle partially observable environments.

• How they work: The agent uses its internal state to keep track of parts of the world it
can't see at the moment. This internal model is updated based on two kinds of
knowledge:

1. How the world evolves on its own.

2. How the agent's own actions affect the world.

• Advantage: By maintaining a model of the environment, these agents can make better
decisions when the current percept alone is insufficient.
 • Example: A self-driving car needs to change lanes. It remembers a car was in its blind
spot a moment ago, even if it can't see it currently. This internal model of the other
car's likely position allows it to make a safer decision than a simple reflex agent could.

3. Goal-Based Agents

Unlike reflex agents that just react to the environment, goal-based agents are capable of
making decisions based on information about a desirable situation, or a goal.

• How they work: When making a decision, these agents consider the future. They ask,
"What will happen if I do action A versus action B?" and choose the action that will
lead them to a goal state. This requires search and planning.

• Advantage: They are much more flexible and can adapt their behavior if the goal
changes. The knowledge that supports a decision is explicitly modeled, which allows
for modifications.

• Example: A GPS navigation system in a car. Its goal is the destination address. It uses a
map and search algorithms to plan a sequence of turns (actions) to reach that goal. If
it encounters a roadblock, it can re-plan a new route to the same goal.

4. Utility-Based Agents

These agents are a more advanced form of goal-based agents. They don't just distinguish
between goal states and non-goal states; they try to find the best possible outcome by
maximizing their "utility"—a measure of preference or happiness.

• How they work: A utility function maps a state to a real number that describes the
associated degree of happiness. When there are conflicting goals (e.g., fast vs. safe) or
multiple ways to reach a goal, a utility-based agent chooses the action that leads to
the state with the highest expected utility.
 • Advantage: This allows for more nuanced decision-making in complex situations by
enabling the agent to make trade-offs between different objectives.

• Example: A ride-sharing AI deciding on a route. While multiple routes might reach the
destination (the goal), a utility-based agent would choose the one that provides the
best balance of speed, cost, safety, and rider comfort, thereby maximizing the overall
utility of the trip.

5. Learning Agents

Learning agents are the most sophisticated type. They are designed to improve their
performance over time by learning from their experiences. A learning agent can start with
basic knowledge and then adapt and improve automatically.

A learning agent has four main components:

 1. Performance Element: This is the core agent responsible for selecting and performing
actions.

2. Critic: This element provides feedback to the agent by comparing its performance
against a fixed standard. It tells the agent how well it is doing.

3. Learning Element: This component uses the feedback from the critic to make
improvements to the performance element's decision-making process.

4. Problem Generator: This component is responsible for suggesting new, exploratory

actions that can lead to new and informative experiences, helping the agent to learn
more effectively.

• Example: An AI designed to play the game of Go.

o The performance element makes the moves on the board.

o After a game, the critic provides feedback (e.g., "this sequence of moves led to
a win").

o The learning element analyzes this feedback and adjusts the agent's strategy
to favor winning moves in the future.

o The problem generator might suggest trying an unconventional opening to

explore new strategies.

[Link] of Environments
In Artificial Intelligence, the environment is the world or context in which an agent operates.
Understanding the properties of this environment is crucial because it dictates the type of
agent required to solve a problem effectively.

Environments are classified along several key dimensions, as detailed below.

Observable vs. Partially Observable

This dimension describes how much of the environment the agent can perceive at any given
moment.

• Fully Observable: An environment is fully observable if the agent's sensors can detect
the complete state of the environment relevant to its choice of action at any point in
time. In such an environment, the agent doesn't need to maintain an internal state or
memory to keep track of the world's history.

o Example: A classic chess game is fully observable. The agent can see the
position of every piece on the board, so it has all the information needed to
make its next move.
 • Partially Observable: An environment is partially observable if the agent's sensors
provide incomplete information about the state. This can be due to noise, inaccurate
sensors, or parts of the environment being hidden. In this case, the agent must
maintain an internal state or memory to track the unobserved aspects of the world.

o Example: A self-driving car operates in a partially observable environment

because its cameras cannot see around corners or through other vehicles. A
poker game is also partially observable as players cannot see each other's
cards.

Static vs. Dynamic

This dimension describes whether the environment can change while the agent is
deliberating.

• Static: A static environment does not change while the agent is performing its actions
or deciding what to do next. These are easier to work with because the agent doesn't
need to worry about the world changing during its thought process.

o Example: A crossword puzzle or a game of chess is static. The board waits for
the player to make a move.

• Dynamic: A dynamic environment can change on its own while the agent is acting. This
requires the agent to continuously monitor the world and react quickly.

o Example: A self-driving car in traffic operates in a dynamic environment. Other

cars move, pedestrians appear, and traffic lights change independently of the
agent's actions.

Discrete vs. Continuous

This dimension relates to the nature of the states and actions in the environment.

• Discrete: A discrete environment has a finite number of distinct, clearly defined states
and actions.

o Example: A chess game has a finite (though very large) number of possible
board configurations and a limited set of legal moves from each position.

• Continuous: A continuous environment is one where states and actions have a

continuous, infinite range of values.

o Example: A self-driving car's environment is continuous. The car's speed,

steering angle, and the exact positions of other vehicles can take on an infinite
range of values within a given interval.

Deterministic vs. Non-deterministic (Stochastic)

This dimension describes the predictability of the outcome of an agent's actions.

 • Deterministic: In a deterministic environment, the next state is completely determined
by the current state and the agent's action. There is no uncertainty or randomness
involved.

o Example: A simple puzzle game where a specific move always produces the
exact same result.

• Non-deterministic (or Stochastic): In a non-deterministic environment, the outcome

of an action is uncertain. The same action in the same state may lead to different
results. The uncertainty is often quantified in terms of probabilities.

o Example: The game of Ludo is non-deterministic because the next state

depends on the random roll of a dice. A robot navigating on a wet floor might
also be in a non-deterministic environment, as trying to move forward could
result in slipping.

Single-agent vs. Multi-agent

This dimension describes how many agents are operating in the environment.

• Single-agent: An environment where only one agent is active.

o Example: A robot cleaning an empty room or a person solving a crossword

puzzle.

• Multi-agent: An environment containing two or more agents. These agents can be

either competitive or cooperative.

o Competitive Example: Chess is a competitive multi-agent environment where

each agent tries to maximize its own performance at the expense of the other.

o Cooperative Example: A team of two delivery robots working together to

transport a large package would be in a cooperative multi-agent environment.
They must coordinate their actions to succeed.

Episodic vs. Sequential

This dimension refers to how actions and their consequences are linked over time.

• Episodic: In an episodic environment, the agent's experience is divided into atomic

"episodes." Each episode consists of the agent perceiving and then performing a single
action. The action taken in one episode does not affect any subsequent episodes.
These environments are simpler because the agent doesn't need to think ahead.

o Example: An AI that classifies images. It analyzes one image, makes a

classification, and then moves to the next. The decision made for the first
image has no impact on the second one.
 • Sequential: In a sequential environment, the current decision can affect all future
decisions. The agent needs to think ahead and may need to remember past actions to
determine the next best action.

o Example: Chess and driving are sequential environments. A single bad move
early in a chess game can lead to an inevitable loss. Similarly, a turn taken while
driving determines the subsequent road and the decisions that follow.

[Link] Representation
Knowledge Representation (KR) is a subfield of Artificial Intelligence that focuses on
representing information about the world in a form that a computer system can utilize to solve
complex tasks. It is not merely about storing data in a database; it is about encoding
knowledge in a way that allows a machine to learn, reason, and behave intelligently, similar
to a human.

What is Knowledge Representation?

KR is the process of creating a model of the world inside an AI agent. This internal model
allows the agent to reason about its environment and make informed decisions. A good KR
system must be able to represent different types of knowledge, including:

• Objects: Facts about all relevant objects (e.g., in a self-driving car, other vehicles and
roads are objects).

• Events: Information about actions that occur (e.g., applying the brakes is an event).

• Performance: Knowledge about how to perform actions (e.g., how to turn the steering
wheel smoothly).

• Meta-Knowledge: Knowledge about what the agent knows (e.g., knowing the limits of
its own knowledge).

• Facts: Truths about the real world (e.g., "Red traffic light means stop").

Why Knowledge Representation is Crucial

KR is a cornerstone of AI for several critical reasons:

1. Enables Reasoning: The primary purpose of KR is to facilitate reasoning. Without a

structured way to represent knowledge, an AI system cannot perform logical
deductions or make inferences. It would just be a collection of data with no way to
interpret or act upon it.

o Example: A medical diagnosis system needs knowledge represented in a

structured way (e.g., "Symptom X and Symptom Y imply Disease Z") to be able
to reason about a patient's condition.
 2. Allows for Learning and Generalization: Effective KR allows an agent to learn from its
experiences and generalize that knowledge to new, unseen situations. It helps the AI
to move beyond rote memorization and understand the underlying principles of a
domain.

3. Handles Complexity and Ambiguity: The real world is complex and filled with
ambiguity. KR provides formalisms, such as logic and probability, to represent this
uncertainty and complexity in a way that a machine can handle.

o Example: A chatbot uses KR to understand that the word "bank" can mean a
financial institution or the side of a river, and it uses context to determine the
correct meaning.

4. Facilitates Communication: For an AI to communicate with humans or other agents, it

needs an internal representation of the concepts it is discussing. KR provides this
internal model, which can then be translated into natural language.

5. Forms the Basis of Intelligent Behavior: Ultimately, an agent's behavior is a direct

result of its knowledge and its ability to reason with that knowledge. A rich and
accurate internal representation of the world is what separates a simple reflexive
machine from an intelligent, goal-oriented agent.

Conclusion Knowledge Representation is arguably the most crucial component for building
intelligent systems because it provides the foundation upon which all higher-level cognitive
functions—such as reasoning, learning, and problem-solving—are built. It bridges the gap
between raw data and intelligent action, allowing an AI to understand its world and act
purposefully within it.

[Link] of Knowledge
1. Declarative Knowledge: This is knowledge of facts. It is explicit, conscious knowledge
that can be "declared" in a statement. It represents the "what" of a domain.

o Description: It includes information about objects, their attributes, and the

relationships between them. It is often represented as simple facts or
assertions.

o Example:

▪ "The Earth is a planet."

▪ "A car has four wheels."

▪ "Paris is the capital of France."

2. Procedural Knowledge: This is the knowledge of "how" to do something. It is often
implicit and is demonstrated through action rather than verbal explanation. It includes
rules, strategies, and procedures for performing tasks.

o Description: It represents the steps or processes required to achieve a goal. In

AI, this is often encoded as production rules (IF-THEN statements).

o Example:

▪ The sequence of steps to bake a cake.

▪ The rules for driving a car (e.g., "IF the traffic light is red, THEN apply
the brakes").

▪ An algorithm for sorting a list of numbers.

3. Meta-Knowledge: This is "knowledge about knowledge". It allows a system to reason

about its own knowledge, capabilities, and limitations.

o Description: It includes information about the properties of other knowledge,

such as its reliability, importance, or origin. It helps in controlling the reasoning
process.

o Example:

▪ Knowing that knowledge from a scientific expert is more reliable than

knowledge from a random blog post.

▪ An AI system knowing which of its problem-solving strategies is most

effective for a particular type of problem.

4. Heuristic Knowledge: This is knowledge based on rules of thumb, educated guesses,

intuition, or experience. It provides a plausible approach to problem-solving but does
not guarantee an optimal or perfect solution.

o Description: Heuristics are used to guide a search or decision-making process,

making it more efficient by pruning less promising options.

o Example:

▪ In a game of chess, a heuristic rule is "it is generally good to control the

center of the board."

▪ A doctor using their experience to guess that a patient's rare

combination of symptoms might indicate a specific disease, even before
tests confirm it.

5. Structural Knowledge: This is knowledge about the organization and relationships of

concepts. It describes how different pieces of knowledge are interconnected.
 o Description: It often takes the form of a hierarchy or a network, showing
relationships like "is-a" (a dog is a mammal), "part-of" (a wheel is part of a car),
and other associations.

o Example: A semantic network or an ontology that organizes concepts. For

instance, knowing that a "poodle" is a type of "dog," which is a type of
"mammal," which is a type of "animal." This structure allows for inheritance of
properties (if mammals are warm-blooded, then dogs and poodles are also
warm-blooded).

[Link] is Logic?
Introduction In the field of Artificial Intelligence, logic is the fundamental component of
knowledge that enables an intelligent agent (whether human or machine) to reason and draw
conclusions by filtering required information from large volumes of sentences. It provides a
formal, unambiguous way to represent knowledge so that it can be processed by a computer.

Every logical system is built upon three main elements: Syntax, Semantics, and Logical
Inference.

1. Syntax

Syntax refers to the set of rules that defines how to construct well-formed or legal sentences
in a language. It is concerned with the structure and representation of the language, not its
meaning. Just as English grammar has rules about subject-verb agreement and sentence
structure, a logical language has rules for combining symbols into valid expressions.

• Function: It dictates the sequence of symbols and connectives that form a complete
and structurally correct statement.

• Example: Consider the following sentences:

o "Ram is riding a bike." — This sentence follows the grammatical rules of English
and is therefore syntactically correct.

o "Bike is riding Ram." — This sentence also follows a valid Noun-Verb-Noun

structure. So, according to the rules of syntax, it is also syntactically correct,
even though it doesn't make logical sense.

2. Semantics

Semantics is concerned with the meaning of the sentences. It provides the rules for
interpreting a sentence and relating it to the real world to determine if it is logically correct
and meaningful. A sentence can be perfectly structured (syntactically correct) but still be
nonsensical (semantically incorrect).
 • Function: Semantics checks the validity and truth of a statement in a given context. It
defines the sense of the sentence.

• Example: Let's revisit the previous examples:

o "Ram is riding a bike." — This is semantically correct because it describes a

possible, meaningful event in the real world.

o "Bike is riding Ram." — This is semantically incorrect. Although the grammar

is fine, the meaning is illogical because a bike is an inanimate object and cannot
perform the action of "riding" a person.

The distinction between syntax and semantics is crucial for an AI to not only process language
correctly but also to understand it.

3. Logical Inference

Inference is the process of deducing new conclusions from existing facts, evidence, or a
problem description. Logical inference is the mechanism that uses formal inference
algorithms to reason with the knowledge represented in the system. It allows an agent to
derive new facts that are not explicitly stated in its knowledge base.

• Function: It is the "thinking" part of the system. It uses established rules (like Modus
Ponens) to extend the agent's knowledge and find solutions to problems.

• Example:

o Fact 1: "All humans are mortal."

o Fact 2: "Socrates is a human."

o Inferred Conclusion: "Socrates is mortal." The AI agent doesn't need to be

explicitly told that Socrates is mortal. Its logical inference system can deduce
this conclusion by applying a logical rule to the existing facts.

[Link] of Knowledge Representation in AI

The Cycle of Knowledge Representation in AI describes the process through which an AI
system perceives its environment, learns from it, reasons about it, and finally acts upon it. This
continuous loop is what enables a machine to exhibit intelligent behavior.

An AI system has several interconnected components that work together to complete this
cycle.

The components of this cycle are:

1. Perception

This is the first and foundational step in the cycle. The perception component is responsible
for gathering data from the environment. It acts as the sensory system for the AI agent.
 • Function: It retrieves raw data using various types of sensors. This input can be in the
form of video from cameras, audio from microphones, text from documents, or any
other sensor-based input like temperature or pressure readings.

• Role: The perception component takes in the external world and converts it into a
format that the system can process.

2. Learning

Once data is captured by the perception component, it is passed to the learning component.
The goal of this stage is to enable the system to improve its performance over time through
experience, rather than being explicitly programmed for every possible scenario.

• Function: The learning component analyzes the perceived data to identify patterns,
acquire new knowledge, develop heuristics (rules of thumb), and refine its internal
models.

• Role: It focuses on self-improvement. For example, by analyzing thousands of images

of cats, the learning component helps the system get better at identifying cats in new,
unseen images.

3. Knowledge Representation and Reasoning (KRR)

This is the core "thinking" component of the AI system and is what most clearly demonstrates
human-like intelligence. It has two parts:

• Knowledge Representation: This involves taking the knowledge acquired during the
learning phase and encoding it into a formal, structured, machine-understandable
format (like logical rules or semantic networks). It defines what the agent needs to
know to behave intelligently.

• Reasoning: This involves using automated reasoning procedures (like an inference

engine) to manipulate the represented knowledge. The reasoning component can
deduce new facts from existing knowledge and make this knowledge available as
needed to solve problems.

• Role: The KRR component allows the machine to "think" about what it knows and
make logical deductions to inform its actions.

4. Planning and Execution

This is the final stage of the cycle, where the agent acts on the world. This component uses
the output from the Knowledge Representation and Reasoning stage to formulate and carry
out a plan.

• Function: It has two main responsibilities:

 1. Planning: The planning component selects a sequence of actions that will help
the agent achieve its desired goal. It considers the agent's current state, the
goal state, and the possible effects of its actions.

2. Execution: Once a plan is formulated, the execution component carries it out

by sending commands to the agent's actuators (e.g., motors in a robot, text
output in a chatbot).

• Role: This component translates thought into action, allowing the agent to affect its
environment.

The entire process is a continuous loop. After the agent executes an action, the state of the
environment may change. The perception component senses this new state, and the cycle
begins again, allowing the agent to continuously learn, adapt, and refine its behavior.

[Link]-Based Agent.
Introduction A Knowledge-Based Agent is an intelligent agent that acts by using its internal
knowledge and reasoning capabilities to make decisions. Instead of relying on hard-coded
reflexes, it uses a representation of the world to act intelligently, making it more flexible and
powerful than simpler agents. Such an agent is composed of two primary components: the
Knowledge Base and the Inference Engine.

Primary Components

1. Knowledge Base (KB): The Knowledge Base is the central repository of the agent's
knowledge. It is the most important component and contains information about the
environment and the world, stored in the form of sentences. These are not sentences
in plain English but are expressed in a formal knowledge representation language (like
propositional or first-order logic) that the system can understand and process.

The KB holds two main types of knowledge:

o Factual Knowledge: Information that is widely accepted and true within the
domain (e.g., "All mammals are warm-blooded").

o Heuristic Knowledge: Rule-of-thumb knowledge based on experience or

expertise, which is useful for making educated guesses (e.g., "If a patient has a
fever and a cough, they might have the flu").

2. Inference Engine (IE): The Inference Engine is the "brain" or the reasoning mechanism
of the agent. Its job is to apply logical rules to the facts stored in the Knowledge Base
to deduce new information or to decide on an action to take. The IE allows the agent
to reason about its knowledge and the information it perceives from the environment.
 o For example, if the KB contains the rule "If it is raining, then the ground is wet"
and it receives a percept that "it is raining," the Inference Engine can deduce
the new fact that "the ground is wet" without needing to directly observe it.

How the Components Interact: Key Operations

The agent interacts with its components through a few core operations:

• TELL: This operation is used to add new sentences or facts to the Knowledge Base.
When an agent perceives something new about its environment, it "TELLS" the KB so
that its internal knowledge is updated.

• ASK: This operation is used to query the Knowledge Base to determine which action
to take. The agent "ASKS" the KB for a course of action, and the Inference Engine uses
the stored knowledge to derive an answer.

• PERFORM: Once the agent receives an answer from the ASK operation, it performs the
selected action in the environment.

Example: A Simple Medical Diagnosis Agent

• Knowledge Base: Contains rules like "If a patient has a fever AND a sore throat, THEN
they might have strep throat." and facts like "Patient X has a fever."

• Inference Engine: The agent receives a new percept: "Patient X has a sore throat." It
TELLS this new fact to the KB. It then ASKS the KB what the diagnosis is. The Inference
Engine applies the rule to the facts ("Patient X has a fever" and "Patient X has a sore
throat") and infers that "Patient X might have strep throat."

• Action: The agent can then PERFORM an action, such as recommending a test for strep
throat.

Conclusion The combination of a Knowledge Base (to store what is known) and an Inference
Engine (to reason with that knowledge) makes a Knowledge-Based Agent highly adaptable. It
can handle complex situations, learn from new information by updating its KB, and make
rational decisions based on logical deduction, forming the foundation for more advanced AI
systems like expert systems.

[Link] Knowledge-Based Agent (KBA)

Generic Knowledge-Based Agent (KBA) is a high-level design for an intelligent agent that
makes decisions by reasoning with an internal representation of the world. This generic
structure serves as a blueprint, outlining the fundamental program loop that enables any such
agent to act intelligently.

The agent's primary function is to accept information it perceives from the environment as
input and return an action as output. To do this, it maintains two key things:
 • A Knowledge Base (KB), which stores sentences about the world.

• A counter, t, to keep track of time, which is initialized to zero.

The agent operates in a continuous three-step cycle each time it is called upon to act.

The Three-Step Process of a Generic KBA

Each time the agent's program is executed, it performs the following sequence of operations
to decide on an action.

1. TELL the KB what it Perceives

First, the agent takes its current percept from the environment (e.g., what it sees or hears).
This raw sensory input is then converted into a formal sentence using a function like MAKE-
PERCEPT-SENTENCE. The agent then TELLS this sentence to its Knowledge Base.

• Purpose: This step updates the agent's internal knowledge of the world with the most
recent information. It's how the agent stays informed about changes in its
environment.

• Analogy: Imagine you're in a room and you feel a draft. You "tell" your internal
knowledge base, "There is a draft."

2. ASK the KB what Action it Should Take

Next, the agent needs to decide on an action. It uses a function like MAKE-ACTION-QUERY to
formulate a question about the best action to perform at the current time. The agent then
ASKS its Knowledge Base this question.

• Purpose: This is the reasoning step. The agent's inference engine uses the facts and
rules stored in the KB to deduce the most rational action to achieve its goals.

• Analogy: After noting the draft, you "ask" yourself, "What should I do about this draft?"
Based on your knowledge (e.g., "Drafts can make me cold," "Closing a window stops a
draft"), your brain infers that closing the window is a good action.

3. TELL the KB about the Chosen Action

After the inference engine determines the best action, the agent records this decision before
executing it. It uses a function like MAKE-ACTION-SENTENCE to create a sentence declaring
the chosen action. It then TELLS this sentence to the Knowledge Base. Finally, the agent
returns the action so it can be performed in the environment.

• Purpose: This step is crucial for the agent to keep track of its own actions over time.
This helps it understand the effects of its actions on the environment and allows it to
learn from its behavior.
 • Analogy: You decide to close the window. You update your internal state: "I am now
closing the window." This prevents you from trying to close the same window again
immediately.

Example: A Wumpus World Agent Consider an agent in the Wumpus World game.

1. TELL (Perceive): The agent is in square [1,2] and perceives a stench. It TELLS its KB:
Stench_in_[1,2].

2. ASK: The agent ASKS its KB: "What is a safe move?" The inference engine uses the rule
"If there is a stench, a Wumpus is in an adjacent square" to infer that square [1,3] or
[2,2] might have a Wumpus. It also knows [1,1] and [2,1] are safe from previous moves.
The best action returned might be "Move to [2,2]".

3. TELL (Act): The agent TELLS its KB: Action_is_Move_to_[2,2]. It then returns the action
to be executed.

Conclusion This generic "Perceive-TELL, ASK, Act-TELL" cycle provides a powerful and flexible
framework for any knowledge-based agent. It allows the agent to maintain a consistent view
of the world, learn from new perceptions, and make rational decisions by logically reasoning
with its accumulated knowledge.

[Link] Levels of Knowledge-Based Agent

A knowledge-based agent can be described at three distinct levels of abstraction, which help
us understand its design and behavior from the most abstract goals down to the physical
implementation.

These levels are the Knowledge Level, the Logical Level, and the Implementation Level.

1. Knowledge Level

This is the most abstract level. It describes the agent in terms of what it knows and what its
goals are, without any detail about how this knowledge is actually represented or computed.

• Focus: The focus is purely on the agent's knowledge and objectives. We specify the
information the agent has about its environment and what it is trying to achieve. This
level allows us to predict the agent's behavior by assuming it will act rationally to
achieve its goals based on its knowledge.

• Analogy: Describing a doctor at the knowledge level means saying, "She knows about
human anatomy, diseases, and treatments, and her goal is to cure the patient." We are
not concerned with the specific language she thinks in or the neural pathways in her
brain.
 • Example: For an AI agent tasked with navigating from point A to point B, the knowledge
level description would simply be: "The agent knows the map of the area and the
location of point B, and its goal is to reach point B."

2. Logical Level

This level is concerned with how the knowledge from the knowledge level is represented
inside the agent. It describes the encoding of knowledge into formal sentences.

• Focus: At this level, the abstract knowledge is translated into a specific knowledge
representation language, such as propositional logic or first-order logic. The
knowledge base is viewed as a collection of logical sentences, and the agent's
reasoning process is described in terms of logical inference.

• Analogy: For the doctor, this level would represent her knowledge as a set of logical
rules, such as: "IF a patient has a fever AND a cough, THEN they might have a
respiratory infection."

• Example: For the navigation agent, the knowledge is now encoded into logical
sentences like At(Agent, A) and Path(A, X, B). The agent uses these sentences and an
inference engine to deduce a sequence of actions.

3. Implementation Level

This is the most concrete level. It describes the physical implementation of the agent,
including the data structures and algorithms that run on the hardware.

• Focus: This level deals with how the logical sentences and inference mechanisms are
actually implemented in code and executed on a computer. It specifies the data
structures used to store the knowledge base (e.g., lists, hash tables) and the algorithms
that perform the logical inference.

• Analogy: For the doctor, this is the physical brain itself—the neurons, synapses, and
electrical signals that store her knowledge and carry out the thought processes. For an
AI, this is the actual code, the CPU, and the memory.

• Example: The navigation agent's knowledge base is stored in a specific data structure,
and its inference engine is an algorithm running on a processor. The agent performs
actions by sending electronic signals to its motors to physically move from point A to
B.

Tying It All Together: Automated Air Conditioner Example

• Knowledge Level: The agent knows it should adjust the room temperature based on
the weather.

• Logical Level: This knowledge is encoded as a rule: IF Weather(Hot) THEN

Set_Temp(Cool).
 • Implementation Level: A temperature sensor (hardware) sends a signal to the
processor (hardware), which runs a program (software) that executes the rule and
sends a command to the AC unit to start cooling.

[Link] to Designing a Knowledge-Based Agent

There are two primary approaches to building a knowledge-based agent: the Declarative
approach and the Procedural approach. In practice, a combination of these two, known as a
Hybrid approach, is often used.

1. Declarative Approach

The declarative approach focuses on telling the agent what it needs to know. The design
process starts by initializing the agent with an empty knowledge base and then systematically
adding sentences and rules about the world one by one.

• Core Principle: The focus is on building a comprehensive and explicit Knowledge Base
(KB). The agent's desired behavior is not programmed directly. Instead, it emerges
from the agent's ability to reason with the knowledge it has been given.

• Process:

1. Start with an empty KB.

2. Add facts and rules (sentences) about the environment.

3. Continue adding knowledge until the agent is knowledgeable enough to make

intelligent decisions using its inference engine.

• Analogy: This is similar to teaching a human. You don't program a child's brain directly.
Instead, you teach them facts ("The stove is hot"), rules ("Don't touch hot things"), and
general knowledge, and you trust that they will use this information to reason about
how to act in a new situation.

• Advantage: This approach is highly flexible. To change the agent's behavior, you simply
need to update the knowledge base by adding, removing, or modifying sentences. You
don't need to rewrite the core program.

2. Procedural Approach

The procedural approach focuses on programming the agent's behavior directly. The desired
behavior is explicitly encoded into the agent's program code.

• Core Principle: The focus is on the algorithms and control flow that dictate the agent's
actions. The knowledge is often implicit, embedded within the structure and logic of
the program itself.
 • Process: The programmer writes a set of procedures or functions that the agent will
execute. The agent's intelligence is a direct result of the programmer's code. This
approach is often implemented using programming languages like LISP or Prolog.

• Analogy: This is like writing a script for a robot to follow. The code gives explicit
instructions: "Move forward 5 steps, turn right, pick up the object." The robot is not
reasoning from a knowledge base; it is simply executing a pre-defined sequence of
commands.

• Disadvantage: This approach is less flexible. If you want to change the agent's
behavior, you must modify and recompile the program code, which can be a complex
and error-prone process.

The Hybrid Approach

In practice, a purely declarative or procedural approach is rare. Most modern AI systems use
a hybrid approach that combines the strengths of both.

• Core Principle: This method uses a declarative representation of knowledge for

flexibility and ease of maintenance, but then compiles this knowledge into more
efficient procedural code for better performance.

• Process:

1. Knowledge is first defined declaratively (e.g., as a set of rules and facts).

2. A compiler or interpreter then transforms this declarative knowledge into a

highly optimized, fast-executing procedural program.

• Benefit: This provides the best of both worlds: the flexibility of being able to easily
update the high-level knowledge base, combined with the speed and efficiency of a
hard-coded procedural system.

• Example: An expert system for medical diagnosis might have its rules defined
declaratively. When the system is run, these rules are compiled into a highly efficient
decision tree (a procedural structure) that can provide diagnoses very quickly.

[Link] of Knowledge representation systems

There are two different types of knowledge in a knowledge representation system,

• Simple Relational Knowledge

• Inferential Knowledge

Simple Relational Knowledge

• In this technique, facts about a set of objects are systematically stored using relations
(or tables) in database systems.
 • These relations depict relationship between different entities.

• The main shortcoming of this approach is that there is little opportunity for inference.

Inheritable Knowledge

• Inheritable knowledge is one that stores data using a hierarchy of classes.

• The representation starts using generalized classes and move down to specialized
classes in the hierarchy.

Example:

Zinya and Vidhi both are instances of under-graduate and post-graduate students,
respectively.

Inheritable knowledge shows the relation between instance a class, and is therefore known
as instance relation of IS-A relation.

In this diagram, objects and values are represented in boxed nodes.

Inferential Knowledge Represents Knowledge in the Form of Formal Logic and Can be Used to
Derive More Facts Accurately.

Statement 1: Diya is a student.

Statement 2: All students are bright.

Then, it can be represented as;

Student(Diya)

∀x = Student (x) ———-> Bright (x)