Artificial Intelligence (MCS101)

Module 2
Knowledge Representation and Reasoning
Topics: Types of knowledge representation, Propositional logic and first-order logic, Semantic networks
and frames, Ontologies and their applications, Deductive and inductive reasoning, Rule-based systems
and non-monotonic reasoning, Probabilistic reasoning and Bayesian networks.

1.1 Knowledge representation and reasoning

Knowledge Representation and Reasoning (KRR) is a key area in Artificial Intelligence (AI) that focuses
on how knowledge is stored, structured, and used for reasoning. It enables machines to simulate human-
like understanding and decision-making.
Key Aspects of KRR
 Knowledge Representation (KR): Storing and structuring information in a way that AI can
use it.
 Reasoning (Inference Mechanisms): Drawing conclusions based on available knowledge.
 Decision-Making: AI uses reasoning to solve problems and answer queries.
Why is KRR Important?
 Enables AI to understand and process real-world information.
 Helps in problem-solving and decision-making.
 Allows AI to communicate with humans in a meaningful way.

Components of Knowledge Representation

There are several key components that define knowledge representation:

1. Facts – Basic truths or assertions about the world.

o Example: "The sky is blue."
2. Objects – Entities or things that exist in the domain.
o Example: "Cats, Trees, Cars."
3. Concepts (Classes/Categories) – Groupings of similar objects.
o Example: "All mammals are warm-blooded animals."
4. Relationships – Connections between objects.
o Example: "A car has four wheels."
5. Rules – Conditions that define logical operations.
o Example: "If it rains, the ground will be wet."
6. Events – Actions or occurrences that happen in a system.
o Example: "A user logs into a website."

Types of Knowledge Representation

There are four primary ways to represent knowledge in AI:
(a) Logical Representation
  Uses formal logic (e.g., propositional and first-order logic).
 Allows AI to reason and derive conclusions.
 Example:
o Propositional Logic: "If it is raining (P), then the ground is wet (Q)."
 P→Q
o First-Order Logic (FOL): "All birds can fly."

 ∀x (Bird(x) → CanFly(x))
(b) Procedural Knowledge Representation
 Knowledge is represented as procedures or sequences of actions.
 Used in expert systems and rule-based AI.
 Example: If diagnosing a disease, a system may follow:
o Step 1: Check symptoms.
o Step 2: Match with known diseases.
o Step 3: Suggest treatment.
(c) Semantic Networks
 Represents knowledge in a graphical structure with nodes and links.
 Example: A semantic network for animals:
Dog → is-a → Mammal
Mammal → is-a → Animal
Dog → has → Fur
(d) Frames
 Similar to object-oriented programming.
 Represents knowledge in data structures called frames.
 Example: A frame for a car:
Car:
- Type: Sedan
- Engine: V6
- Fuel: Petrol
Reasoning in AI
Reasoning is the process of drawing logical conclusions from knowledge.
(a) Types of Reasoning
1. Deductive Reasoning:
o Moves from general to specific.
 o Example: "All humans are mortal. Socrates is a human. Therefore, Socrates is
mortal."
2. Inductive Reasoning:
o Moves from specific to general.
o Example: "Every swan I’ve seen is white. Therefore, all swans might be white."
3. Abductive Reasoning:
o Infers the most likely explanation.
o Example: "If the ground is wet, it probably rained."
4. Non-Monotonic Reasoning:
o Beliefs can change when new knowledge is added.
o Example: "Birds can fly, but penguins are birds that cannot fly."

Applications of Knowledge Representation and Reasoning

KRR is widely used in AI applications, such as:

 Expert Systems: Medical diagnosis (e.g., MYCIN, DENDRAL).

 Natural Language Processing (NLP): Understanding human speech (e.g., Siri,
ChatGPT).
 Robotics: Autonomous decision-making (e.g., self-driving cars).
 Semantic Web: Organizing and retrieving structured knowledge (e.g., Google
Knowledge Graph).

1.2 Propositional logic and first-order logic

Knowledge Representation in AI heavily relies on logic-based formalisms. Two fundamental types of
logic used in AI are Propositional Logic (PL) and First-Order Logic (FOL).

1. Propositional Logic (PL)

Propositional Logic (also known as Boolean Logic) is the simplest form of logic that deals
with statements (propositions) that are either true or false.

1.1 Syntax of Propositional Logic

 Atomic Propositions: Basic statements that are either True (T) or False (F).
o Example:
 P:P:P: "It is raining."
 Q:Q:Q: "The road is wet."
 Logical Connectives:

o Conjunction (P ∧ Q): "It is raining and the road is wet."

o Negation (¬P): "It is not raining."

o Disjunction (P ∨ Q): "It is raining or the road is wet."

o Implication (P → Q): "If it is raining, then the road is wet."
o Biconditional (P ↔ Q): "The road is wet if and only if it is raining."

1.2 Semantics of Propositional Logic

Each proposition is assigned a truth value:

o P∧Q=FalseP ∧ Q = FalseP∧Q=False
 If P=TrueP = TrueP=True and Q=FalseQ = FalseQ=False, then:

o P∨Q=TrueP ∨ Q = TrueP∨Q=True
o P→Q=FalseP → Q = FalseP→Q=False (since True → False is False)

1.3 Limitations of Propositional Logic

 No Objects or Relationships: PL cannot express statements like "John owns a car."

 Limited Expressiveness: It cannot express general knowledge (e.g., "All humans are
mortal").

2. First-Order Logic (FOL) (Predicate Logic)

First-Order Logic (FOL) is more expressive than PL because it introduces:

1. Objects (Entities)
2. Predicates (Properties of objects)
3. Quantifiers (Generalization of statements)

2.1 Syntax of First-Order Logic

 Constants (Objects): Represent specific entities.

o Example: John, Dog, Car
 Variables: Represent general objects.
o Example: x, y, z
 Predicates: Represent relations between objects.
o Example: Loves(John, Mary) means "John loves Mary."
 Functions: Represent mappings between objects.
o Example: Father(John) = David (John's father is David).

2.2 Quantifiers in FOL

 Universal Quantifier (∀) → "For all"

 ∀x(Human(x)→Mortal(x))\forall x (Human(x) →
o Example: "All humans are mortal."

Mortal(x))∀x(Human(x)→Mortal(x))
 Existential Quantifier (∃) → "There exists"

 ∃x(Student(x)∧Likes(x,AI))\exists x (Student(x) ∧ Likes(x,

o Example: "Some students like AI."

AI))∃x(Student(x)∧Likes(x,AI))

2.3 Example Statements in FOL

1. Propositional Logic:
o "If it rains, the ground is wet."
 Rain→WetGroundRain → WetGroundRain→WetGround
2. First-Order Logic:

 ∀c(Rains(c)→WetGround(c))\forall c (Rains(c) →
o "If it rains in a city, the ground in that city is wet."

WetGround(c))∀c(Rains(c)→WetGround(c))
 2.4 Advantages of First-Order Logic

 More expressive than PL.

 Can describe objects, properties, and relationships.
 Allows generalization using quantifiers.

2.5 Limitations of First-Order Logic

 Computational Complexity: Reasoning in FOL is more difficult than in PL.

 Requires More Processing Power: Checking logical consistency takes longer.

Comparison of Propositional Logic and First-Order Logic

Feature Propositional Logic (PL) First-Order Logic (FOL)

Limited (only true/false More expressive (handles objects, relations, and

Expressiveness
statements) quantifiers)

Connectives (∧, ∨, →) ∃)
Propositions (P, Q, R), Logical Objects, Predicates, Functions, Quantifiers (∀,
Symbols Used

"If it rains, the road is wet." "If it rains in any city, roads in that city are wet."
Example
(Rain → WetRoad) (∀c Rains(c) → WetRoad(c))

Cannot express relationships

Limitations Harder to compute reasoning due to complexity
between objects

1.3 Semantic networks and frames

1. Semantic Networks
Definition:
A Semantic Network (SN) is a graph-based representation of knowledge where:
 Nodes represent objects, concepts, or events.
 Edges (links) represent relationships between them.
Example of a Semantic Network:
Consider a representation of knowledge about animals: markdown
Animal
│
┌───┴───┐
│ │
Bird Mammal
│ │
┌───┴───┐ │
│ │ │
Eagle Parrot Dog
 Animal → Bird (A bird is an animal)
 Bird → Eagle (An eagle is a type of bird)
 Bird → Parrot (A parrot is a type of bird)
 Mammal → Dog (A dog is a type of mammal)
Types of Links in Semantic Networks
1. "IS-A" (Inheritance Relationship):
o Parrot IS-A Bird (A parrot is a type of bird).
2. "HAS-A" (Attribute Relationship):
o Bird HAS-A Wings (A bird has wings).
3. "CAN" (Capability Relationship):
o Bird CAN Fly (A bird can fly).
Advantages of Semantic Networks:
✔ Graph-based and easy to understand.
✔ Supports inheritance (if "Birds can fly," then "Parrots can fly" automatically).
✔ Useful in natural language processing (NLP) and knowledge-based systems.
Limitations of Semantic Networks:
✘ Lack of formal rules—can be ambiguous.
✘ Scalability issues—difficult to maintain for large knowledge bases.

2. Frames
Definition:
A Frame is a structured data representation that groups information into a template (similar to a database
record).
Each frame consists of:
 Slots (Attributes/Properties)
 Fillers (Values assigned to slots)
Example of a Frame for a Car:
Frame: Car
----------------------
Manufacturer: Tesla
Model: Model S
Color: Red
Engine: Electric
Speed: 200 km/h
Owner: John
Here,
  "Car" is the frame name.
 "Manufacturer", "Model", "Color", "Engine" are slots.
 "Tesla", "Model S", "Red" are fillers (values assigned to slots).
Inheritance in Frames:
Frames allow inheritance like semantic networks:
Frame: Vehicle
----------------------
Type: Transport
Fuel: Petrol/Diesel/Electric
Frame: Car (inherits from Vehicle)
----------------------
Type: Transport
Fuel: Petrol/Diesel/Electric
Manufacturer: Tesla
 A "Car" is a "Vehicle", so it inherits Type: Transport.
Advantages of Frames:
✔ Modular & Organized representation of knowledge.
✔ Allows default values (if not mentioned, use the parent class value).
✔ Used in expert systems and NLP applications.
Limitations of Frames:
✘ Not well-suited for reasoning (lacks logic).
✘ Does not support complex relationships as well as semantic networks.

Comparison: Semantic Networks vs. Frames

Feature Semantic Networks Frames
Structure Graph-based Slot-based (table format)
Representation Nodes and Links Objects and Attributes
Main Use Relationship mapping Data structuring
Inheritance Yes (via IS-A) Yes (via parent-child frames)
Example "Bird IS-A Animal" Frame: Bird → Has wings: Yes
Limitations Hard to manage in large networks Lacks reasoning capabilities

Applications of Semantic Networks & Frames in AI

Where are they used?
✔ Semantic Networks: Used in NLP, Knowledge Graphs, and Chatbots (Google Knowledge Graph).
✔ Frames: Used in Expert Systems, Robotics, and Object-Oriented AI.
Real-Life Example:
 Semantic Network:
 o Used in Google Search Knowledge Graph to link related concepts.
 Frames:
o Used in Medical Diagnosis Systems (e.g., a patient’s frame with symptoms,
diagnosis, and treatments).

1.4 Ontologies and their applications

What is an Ontology?
Ontology is a formal representation of knowledge that defines a set of concepts, relationships, and
rules within a domain. It provides a shared understanding of information that can be used by AI
systems, databases, and applications.
Key Components of Ontology:
1. Concepts (Classes) → The main categories or types of things (e.g., Person, Car, Animal).
2. Instances (Objects) → Specific examples of concepts (e.g., John is a Person, Tesla Model S
is a Car).
3. Properties (Attributes) → Characteristics of concepts (e.g., a Car has a color, an Animal
has a species).
4. Relationships (Relations) → Define how concepts relate to each other (e.g., A Car is
owned by a Person).
5. Rules & Constraints → Logical statements that enforce consistency (e.g., A Person must
have a Date of Birth).
Example of a Simple Ontology for Transportation:
Concepts:
- Vehicle
- Car (subclass of Vehicle)
- Bike (subclass of Vehicle)
- Person

Relationships:
- Person OWNS Vehicle
- Car HAS Engine
- Bike HAS Two Wheels
 Here, Car and Bike inherit properties from Vehicle.
 A Person can own a Vehicle.
 A Car has an Engine, while a Bike has Two Wheels.
Types of Ontologies
1. Domain Ontologies:
  Focus on a specific field (e.g., Medicine, Finance, Cybersecurity).
 Example: SNOMED CT (Medical Ontology) for diseases, symptoms, and treatments.
2. General Ontologies:
 Broad knowledge across multiple domains.
 Example: WordNet (a lexical database of words and meanings).
3. Upper-Level Ontologies:
 Provide high-level categories for structuring knowledge.
 Example: SUMO (Suggested Upper Merged Ontology) for AI and reasoning.
Applications of Ontologies in AI
1. Semantic Web & Knowledge Graphs
 Ontologies help create a structured web of data.
 Example: Google Knowledge Graph links related entities (e.g., searching for “Elon Musk”
shows Tesla, SpaceX, and related entities).
2. Natural Language Processing (NLP)
 Used in chatbots, search engines, and text understanding.
 Example: IBM Watson uses ontologies to interpret medical texts.
3. Expert Systems & AI Assistants
 Ontologies store expert knowledge for AI-driven decision-making.
 Example: Medical AI uses ontologies for diagnosing diseases based on symptoms.
4. Robotics & Autonomous Systems
 Robots use ontologies to interpret environments and make decisions.
 Example: Self-driving cars understand “Road, Traffic Signal, Pedestrian” using ontologies.
5. E-Commerce & Recommendation Systems
 Ontologies improve product recommendations by understanding user preferences.
 Example: Amazon & Netflix recommendation engines use ontologies to categorize
products/movies.
4. Advantages of Ontologies in AI
Standardized Knowledge Representation → Common vocabulary for AI systems.
Reasoning & Inference → AI can make logical deductions based on ontology rules.
Interoperability → Helps different AI systems communicate & share knowledge.
Improves Machine Learning & NLP → Structured data enhances AI models.

Challenges of Ontologies:
Complex to Design → Requires domain expertise.
Scalability Issues → Large ontologies can be difficult to maintain.
Knowledge Updates → Must be updated as domains evolve.
Ontologies in Action: Real-World Use Cases
Healthcare: Disease diagnosis using medical ontologies (SNOMED, UMLS).
Finance: Risk assessment in banking using financial ontologies.
Cybersecurity: Threat intelligence and attack detection using ontologies.
Education: AI-driven personalized learning paths based on student data.
Autonomous Vehicles: Understanding road environments and traffic rules.
1.5 Deductive and inductive reasoning

Reasoning is a fundamental part of artificial intelligence that allows machines to draw

conclusions based on available information. There are two major types of reasoning:

1. Deductive Reasoning → Moves from general principles to specific conclusions.

2. Inductive Reasoning → Moves from specific observations to general conclusions.

Both types of reasoning are used in AI systems, expert systems, and machine learning models.

1. Deductive Reasoning

Definition:
Deductive reasoning is a top-down approach where conclusions are drawn from a set of general
rules or premises. If the premises are true, the conclusion must also be true.

Example:

1. Premise 1: All humans are mortal.

2. Premise 2: Socrates is a human.
3. Conclusion: Therefore, Socrates is mortal.

Since both premises are true, the conclusion must be true.

Deductive Reasoning in AI:

AI systems use deductive reasoning to apply rules and logic to make decisions.

Applications of Deductive Reasoning in AI:

Expert Systems: AI doctors diagnose diseases using predefined medical rules.

Theorem Proving: AI proves mathematical and logical theorems.
Rule-Based Systems: AI chatbots follow strict logical rules for responses.
Automated Planning: AI follows pre-defined sequences of actions.

Limitations of Deductive Reasoning:

Requires Predefined Rules: AI must already know the rules.

No New Knowledge: It does not learn new patterns but only applies existing logic.
2. Inductive Reasoning

Definition:
Inductive reasoning is a bottom-up approach where patterns and observations are used to form
general rules or conclusions. The conclusions are probable but not always guaranteed to be true.

Example:

1. Observation 1: The sun has risen in the east every day so far.
2. Observation 2: The sun rose in the east today.
3. Conclusion: The sun will always rise in the east.

The conclusion is likely true, but not 100% certain (e.g., if Earth's rotation changes).

Inductive Reasoning in AI:

AI uses inductive reasoning to learn patterns from data and make predictions.

Applications of Inductive Reasoning in AI:

Machine Learning: AI learns patterns from data and makes predictions.

Fraud Detection: AI detects fraudulent transactions based on past patterns.
Medical Diagnosis: AI predicts diseases based on patient history.
Speech & Image Recognition: AI learns from examples to recognize sounds and images.

Limitations of Inductive Reasoning:

Not Always 100% Accurate: Predictions can be wrong if the data is incomplete.
Data Dependency: AI must be trained on a large dataset to make good predictions.

Differences Between Deductive and Inductive Reasoning

Feature Deductive Reasoning Inductive Reasoning

Approach Top-down Bottom-up

Starting
General rules Specific observations
Point

Certainty 100% certain if premises are true Probable but not always certain

Use in AI Rule-based systems, Logic programming Machine learning, Pattern recognition

"All birds have feathers → A parrot is a bird "I have seen 100 swans, all are white →
Example
→ A parrot has feathers" All swans are white"
1.5 Rule-based systems and non-monotonic reasoning

1. Rule-Based Systems

What is a Rule-Based System?

A rule-based system (RBS) is an AI system that uses if-then rules to make decisions or
solve problems. It consists of:

 Knowledge Base: Stores facts and rules.

 Inference Engine: Applies rules to derive conclusions.
 Working Memory: Stores intermediate results.

Example of a Rule-Based System:

A simple medical diagnosis system may have rules like:

1. Rule 1: IF a person has a fever AND cough, THEN they may have the flu.
2. Rule 2: IF a person has a headache AND a stiff neck, THEN they may have
meningitis.

How Rule-Based Systems Work in AI:

✅ Input Data: AI receives information (e.g., patient symptoms).

✅ Matching Rules: AI checks the rules in the knowledge base.
✅ Inference Engine: Applies logic to reach a conclusion.
✅ Output Decision: AI provides a diagnosis or solution.

Applications of Rule-Based Systems:

✔ Expert Systems: AI in medicine, finance, and legal fields.

✔ Chatbots: AI assistants that follow pre-defined rules.
✔ Fraud Detection: AI detects suspicious transactions.
✔ Automated Customer Support: AI answers FAQs based on rules.

Limitations of Rule-Based Systems:

✘ Rigid Rules: Cannot adapt to new data without manual updates.

✘ Scalability Issues: Large rule sets make systems complex.
✘ Lack of Learning: Cannot improve without human intervention.

2. Non-Monotonic Reasoning

What is Non-Monotonic Reasoning?

Unlike classical logic (which is monotonic and does not change), non-monotonic reasoning
allows AI to revise beliefs when new evidence is introduced.

🔹 Monotonic Logic: Once a fact is added, it cannot be removed.

🔹 Non-Monotonic Logic: AI retracts previous conclusions when new facts appear.

Example of Non-Monotonic Reasoning:

 1. Initial Knowledge: "Birds can fly." → AI concludes that Tweety (a bird) can fly.
2. New Information: "Tweety is a penguin." → AI revises its belief: Tweety cannot
fly.

Why is Non-Monotonic Reasoning Important in AI?

✅ Handles Uncertainty: AI can update knowledge dynamically.

✅ Real-World Adaptation: AI adapts to new data (e.g., self-driving cars).
✅ Better Decision-Making: AI changes its conclusions based on evidence.

Applications of Non-Monotonic Reasoning in AI:

✔ Autonomous Systems: AI-driven cars update navigation based on traffic.

✔ Medical Diagnosis: AI revises diagnoses with new test results.
✔ Legal AI Systems: AI changes legal decisions when new laws apply.

Types of Non-Monotonic Reasoning:

1. Default Logic: AI assumes defaults unless contradicted.

2. Auto epistemic Logic: AI reasons about its own beliefs.
3. Circumscription: AI minimizes assumptions to refine conclusions.

Comparison of Rule-Based Systems and Non-Monotonic

Reasoning
Feature Rule-Based Systems Non-Monotonic Reasoning
Logic Type Fixed if-then rules Can revise beliefs
Adaptability Rigid, needs manual updates Updates automatically
Use Cases Expert systems, chatbots AI planning, medical diagnosis
Example "IF fever, THEN flu" "IF fever, THEN flu, UNLESS malaria detected"

1.6 Probabilistic Reasoning and Bayesian Networks

Probabilistic reasoning is a mathematical approach that helps AI systems deal with

uncertainty in decision-making. Unlike classical logic, where facts are either true or false,
probabilistic reasoning assigns degrees of belief to events.

🔹 Why is Probabilistic Reasoning Important?

 The real world is uncertain (e.g., weather forecasting, medical diagnosis).

 AI must make decisions with incomplete information.
 Helps AI systems learn from data and make predictions.

🔹 Example:
A medical AI must diagnose a patient. It doesn't know for sure if the patient has COVID-19,
but it can estimate the probability based on symptoms like fever, cough, and fatigue.

2. Probability Basics
Probabilistic reasoning uses probability theory, which measures the likelihood of an event
occurring.

Key Probability Concepts:

✔ Prior Probability (P(A)) → Initial probability before observing new data.

✔ Conditional Probability (P(A | B)) → Probability of A occurring given that B is true.
✔ Bayes' Theorem → Updates probability estimates based on new evidence.

📌 Formula for Bayes' Theorem:

P(A∣B)=P(B∣A)⋅P(A)P(B)P(A | B) = \frac{P(B | A) \cdot P(A)}{P(B)}

Where:

 P(A∣B)P(A | B) → Probability of A given B (Posterior Probability)

 P(B∣A)P(B | A) → Probability of B given A (Likelihood)
 P(A)P(A) → Prior probability of A (Prior Knowledge)
 P(B)P(B) → Probability of B (Evidence)

3. Bayesian Networks

A Bayesian Network (BN) is a graphical model that represents probabilistic relationships

between variables.

Structure of a Bayesian Network:

 Nodes = Random Variables (e.g., Disease, Symptoms)

 Edges = Dependencies between variables
 Each node has a Conditional Probability Table (CPT)

📌 Example: Diagnosing a Disease

Consider a Bayesian Network for Flu Diagnosis:

↘ Cough
Cold → Fever → Test Result

 Cold increases the chance of Fever.

 Fever increases the chance of Cough.
 A Test Result depends on Fever.

Each node has a Conditional Probability Table (CPT):

| Fever | Cold | P(Fever | Cold) | |-------|------|---------| | Yes | Yes | 0.8 | | Yes | No | 0.2 | | No |
Yes | 0.2 | | No | No | 0.8 |

This allows AI to compute the probability of having Cold given symptoms like Fever and
Cough.

4. Applications of Bayesian Networks

✔ Medical Diagnosis: AI predicts diseases based on symptoms.
✔ Fraud Detection: AI detects suspicious transactions.
✔ Speech Recognition: AI predicts words in speech-to-text systems.
✔ Self-Driving Cars: AI estimates risks of accidents.

5. Advantages & Limitations

Feature Bayesian Networks Classical Logic

Handles Uncertainty ✅ Yes ❌ No

Learns from Data ✅ Yes ❌ No

Requires Prior Knowledge ✅ Yes ❌ No

Computationally Expensive ❌ Yes ✅ No