Data:

Raw facts or figures without context. Example: 100, 75, 90 (just numbers, no meaning yet).

Information:
Data that has been organized or given context to make it meaningful. Example: "John scored 100, Sarah
scored 75, and Mike scored 90 in math."

Knowledge:
Insights or understanding gained by analyzing and applying information. Example: "John is the best
performer in math, while Sarah might need more practice."

Data = "What?"
Information = "What does it mean?"
Knowledge = "What can I do with it?"

Knowledge Representation
Study of ways how knowledge is actually represented. A good knowledge representation naturally represents
problem domain. It is not only about storing data in database. But it also enable intelligent machine to learn
from this knowledge and behave intelligently.

Knowledge and Representation are 2 distinct entities

What to Represent:
Following are the kind of knowledge which needs to be represented in AI systems:
o Object : Refers to things or items in the world that the AI system needs to understand. Includes facts
about their properties, characteristics, and relationships. Example "Guitars have strings" or
"Trumpets are brass instruments."
o Events: Actions or occurrences that happen in the world, which the AI system needs to recognize
and respond to. Example: "A concert is happening" or "A guitar string is being plucked."
o Performance: Knowledge about how to do something or perform a task. Example: "How to play a
chord on a guitar" or "How to recognize a tune."
o Meta-knowledge: Knowledge about the knowledge itself—understanding what the system knows or
doesn’t know. Example: "I know how a guitar sounds but don’t know how a trumpet sounds."
o Facts: Basic truths or real-world information that the AI system must represent and work with.
Example: "Water boils at 100°C" or "The capital of France is Paris."
o Knowledge-Base: The central "brain" of an AI system that stores all its knowledge. It is a collection
of sentences or representations that contain the facts, rules, and logic the AI system needs to
function. If an AI system for music has a KB, it may include:
Sentence 1: "Guitars have six strings."
Sentence 2: "Playing a chord involves pressing multiple strings."
Approaches to Knowledge Representation in AI:
1. Logical Representation
This uses logic (like math rules) to represent knowledge in a clear and precise way. AI uses this to reason
and make decisions based on facts and rules.
 Example: "If it’s raining, then carry an umbrella."
 Used in areas needing strict rules, like proving theorems or rule-based systems.

2. Semantic Networks
Imagine a big mind map! Concepts are shown as nodes, and their relationships are shown as connections
(edges).
 Example: A node for "dog" connected to "animal" (relationship: "is a").
 Helps AI understand relationships and categories, like organizing family trees or synonyms in
language.

3. Frames
Think of frames like templates or forms with slots to fill in details about something.
 Example: A "car" frame might have slots for "make," "model," "color," and "owner."
 AI uses frames to understand objects or situations in a structured way, like predicting what happens
in common scenarios (e.g., ordering at a restaurant).

4. Production Rules
These are simple "if-then" rules that guide actions.
 Example: "If the light is red, then stop."
 Commonly used in systems like virtual assistants or expert systems to solve problems or make
decisions step-by-step.

Knowledge Representation can be at 2 levels :

1. Knowledge Level : at which facts are represented
2. Symbol Level : Representation of objects in terms of symbols and can be manipulated by programs.

Normal English is insufficient, too hard currently for a computer program to draw inferences in natural
languages.Therefore, we must be able to map "facts to symbols" and "symbols to facts" using forward and
backward representation mapping.
Requirements of Knowledge Representation :
1. Representational Accuracy:
o The system should be able to store all the knowledge it needs for the task.
o Example: If the system is about animals, it should be able to represent details like types of
animals, their habitats, and diets.
2. Inferential Adequacy:
o The system should be able to use the knowledge it has to create new knowledge.
o Example: If the system knows "Dogs are mammals" and "Mammals have fur," it should
figure out that "Dogs have fur."
3. Inferential Efficiency:
o The system should find the best and quickest way to reason or make decisions.
o Example: Instead of checking all possible animals, the system should focus on "mammals" if
it knows the question is about "furry animals."
4. Acquisitional Efficiency:
o The system should easily learn and add new knowledge automatically.
o Example: If new information about "elephants" is given, the system should update its
knowledge base without much effort.

Approaches to Knowledge Representation in AI:

1. Simple Relational Knowledge:
o This is like a table in a database where facts are stored in rows and columns.
o How it works: Each column represents a type of information, and each row represents a
specific object or fact.
o Example: A table of players with their weight and age:
Copy code
Player | Weight | Age
Player1 | 65 | 23
Player2 | 58 | 18
Player3 | 75 | 24
o Limitation: It doesn’t allow for much reasoning or making new conclusions.

2. Inheritable Knowledge:
o Think of it as a family tree or hierarchy of categories.
o How it works: Information is stored in levels, where lower levels (children) inherit properties
from higher levels (parents).
 o Example:
 Animal (parent): Can move.
 Bird (child): Can fly (inherits "Can move").
o This helps organize and share common traits among groups.

3. Inferential Knowledge:
o This uses logical statements to represent knowledge and derive new facts.
o How it works: It uses rules of logic to make conclusions.
o Example:
 Fact 1: "Marcus is a man."
 Fact 2: "All men are mortal."
 Conclusion: "Marcus is mortal."

4. Procedural Knowledge:
o This involves step-by-step instructions or rules for how to do something.
o How it works: Uses small programs or “if-then” rules to guide decisions.
o Example:
 Rule: "If it’s raining, then carry an umbrella."
o Used in programming languages like LISP or Prolog to solve specific tasks.

Issues with Knowledge Representation in AI

1. Important Attributes
 Definition: Attributes like "instance" and "isa" are essential because they help represent relationships
and support inheritance of properties.
 Example: If "dog" is an instance of "animal" (isa relationship), it can inherit attributes like
"breathes" or "eats."

2. Relationship Among Attributes

 Definition: Attributes themselves have relationships that help describe objects better.
 Key Points:
o Inverse: Ensures consistency when one attribute changes. For example, if "A is taller than
B," then "B is shorter than A."
o Hierarchy: Attributes can be general or specialized. For example, "height" is part of
"physical-size," which is part of "physical-attribute."
o Reasoning: We can infer missing values, like "height must be in units of length."
 o Single-Valued Attributes: Some attributes only take one value, e.g., a person can belong to
only one country at a time.

3. Choosing Granularity
 Definition: Granularity is the level of detail to represent knowledge.
 Key Points:
o Too broad: May not give enough detail for reasoning.
o Too detailed: Requires a lot of storage.
 Example:
o High-level fact: "John spotted Sue" helps answer, "Who spotted Sue?"
o Low-level: To answer "Did John see Sue?" you need more facts like "Spotted(x, y) → Saw(x,
y)."

4. Sets of Objects
 Definition: Sets are used to group objects and represent properties common to all or most members.
 Key Points:
o Efficient to attach properties to sets rather than individual objects.
o Example: "Sheep outnumber people in Australia" is a fact about the set of "sheep" and
"people."
 Methods:
o Logical representations use quantifiers like "all" or "most."
o Hierarchies use inheritance to pass down properties to individual members.

5. Finding the Right Structure

 Definition: Accessing the correct knowledge structure when needed is challenging.
 Key Problems:
o How to choose the best structure for a situation.
o How to revise the structure if it doesn’t fit.
o How to create new structures if none exist.
 Solution: No universal method exists; some KR techniques address specific issues better than others.
Syntax (How things are written)
 What it is: Syntax is about the rules that tell us how to arrange words or symbols in a specific way
so that they make sense in a knowledge system.
 Example: Just like grammar tells you how to arrange words in a sentence (like "She is playing"),
syntax in a knowledge system tells you how to put symbols together correctly.
o For instance, in logic: A AND B means both A and B are true, and this is the "correct" way to
write it.
 Why it's important: Without proper syntax, it would be impossible for a computer to understand
what you're trying to say. It makes sure everything is written properly.

2. Semantics (What things mean)

 What it is: Semantics is about the meaning behind the symbols and sentences you write. It's how the
symbols in your system relate to the real world or ideas.
 Example: If you write A → B, you can give it meaning like "If it rains (A), then the ground gets wet
(B)". Here, A and B stand for real things (rain and wet ground).
 Why it's important: Syntax tells you how to write something, but semantics tells you what it
means. It connects your knowledge to reality .

3. Inference Procedure (How to make new conclusions)

 What it is: Inference is the process of using rules to figure out new information from what you
already know. It's like solving a puzzle by putting the pieces together.
 Example: If you know:
o "If it rains, the ground gets wet" (A → B)
o "It is raining" (A)
o You can use the inference rule to conclude that "The ground is wet" (B).
 Why it's important: Inference helps the system think and come up with new facts or answers from
the knowledge it already has. It's how a computer can "reason" and solve problems.

What is a Semantic Network?

A semantic network is a form of knowledge representation that visually illustrates how concepts are related
to each other. In AI, it helps in structuring and organizing data in a way that machines can interpret, process,
and use it for decision-making. The nodes in a semantic network represent concepts, and the edges define
the relationships between these concepts, such as "is a," "part of," or "related to."
Limitations :
Lack of logic to improve the notion’s expressive power and robustness.
Lack of Reasoning
Binary relation difficult to represent
Quantified statements are difficult to represent

Partitioned Semantic Network

A Partitioned Semantic Network is a way to organize and represent knowledge in a structured and modular
way. It breaks down the information into smaller, independent sections (called partitions) to make it easier to
handle and process. Every node and arc of a n/w belongs to one or more spaces.
Features :
 Divided Knowledge: Information is grouped into smaller, independent partitions based on topics or
domains.
 Interconnected Partitions: Links between partitions allow access to related knowledge across
domains.
 Hierarchical Structure: General concepts at the top, specific details below for easy inference.
 Focused Search: Only relevant partitions are searched, improving efficiency.
 Scalability: New knowledge can be added as new partitions without disrupting existing ones.

Frames
Conceptual Dependency

Script
Production Rules