Unit -1

AI (Ar ficial intelligence)

Concept of intelligence:
The concept of intelligence in ar ficial intelligence (AI) is to simulate human intelligence through the use of data,
algorithms, and computa onal power. AI technologies allow machines to perform tasks that typically require human
intelligence, such as: learning, reasoning, problem-solving, percep on, language understanding, decision making,
crea vity, and autonomy.
AI technologies include machine learning, deep learning, and natural language processing (NLP).

Defini on of AI:
 Ar ficial intelligence (AI) is technology that enables computers and machines to simulate human learning,
Understanding, problem solving, decision making, crea vity and autonomy.
 Learning: AI can learn from experience and improve performance when exposed to data sets.
 Problem-solving: AI can process large amounts of data at once to find pa erns and solve complex problems.
 Decision-making: AI can make recommenda ons and enable faster, more accurate predic ons.
AI can perform a variety of advanced func ons, including: Seeing, Understanding and transla ng spoken and wri en
language, analyzing data, interac ng with the environment, and Exercising crea vity.

How Does AI Work?

Ar ficial intelligence (AI) enables machines to learn from data and recognize pa erns in it, to perform tasks more
efficiently and effec vely. AI works in five steps:
 Input: Data is collected from various sources. This data is then sorted into categories.
 Processing: Refers to the stage where raw data is analyzed, manipulated, and transformed into meaningful
informa on by an AI system.
 Outcomes: The AI can then use those pa erns to predict outcomes.
 Adjustments: If the data sets are considered a “fail,” AI learns from that mistake, and the process is repeated again
under different condi ons.
 Assessments: In this way, AI is constantly learning and improving.

Type of AI:
There are three types of AI.
1.Narrow AI(NAI): Narrow AI is a type of AI that is perform a dedicated task with intelligence. The most common
and currently available AI is Narrow AI in the world of Ar ficial Intelligence. Narrow AI cannot perform beyond
(out of) its field or boundaries, because it is trained only for a specific task.
Some Examples of Narrow AI are playing chess, purchasing sugges ons on e-commerce site, self-driving cars,
speech recogni on, and image recogni on.
2. General AI: General AI is a type of intelligence which can perform any intellectual task like a human.
General AI aims to create a system that is smarter and can think on its own like Human. Currently, there is no such
system exist which could come under general AI and can perform any task as perfect as a human.

3.Super AI: Super AI is a level of system intelligence at which machines can exceeds human intelligence, and
perform any task with cogni ve proper es be er than humans. This is the result of general AI.
Some of the key characteris cs of strong AI include the ability to think, reason, solve puzzles, make decisions, plan,
learn, and communicate on its own.

AI Intelligent Agents:
AI Intelligent Agents: An agent is a computer program or system that is designed to understand its environment,
make decisions and take ac ons to achieve a specific goal. it is not directly controlled by a human operator.
Agent is anything that understand its environment through sensors and actuator or effectors.

 Sensor: Sensor is a device which detects the changes in the environment and sends the informa on to
other electronic devices. An agent observes its environment through sensors.
 Actuators: Actuators are the component of machines that converts energy into mo on. The actuators are
only responsible for moving and controlling a system. An actuator can be an electric motor, gears, rails, etc.
 Effectors: Effectors are the devices which affect the environment. Effectors can be legs, wheels, arms,
fingers, wings, fins, and display screen.

An agent can be:

 Human-Agent: A human agent has eyes, ears, and other organs which work for sensors and hand, legs,
vocal tract work for actuators.
 Robo c Agent: A robo c agent can have cameras, infrared range finder, NLP for sensors and various
motors for actuators.
 So ware Agent: So ware agent can have keystrokes, file contents as sensory input and act on those inputs
and display output on the screen.
 structure of AI Agents:
To understand the structure of Intelligent Agents, we should be familiar with Architecture and Agent programs.
Architecture is the machinery that the agent executes on. It is a device with sensors and actuators, for example, a
robo c car, a camera, and a PC.
An agent program is an implementa on of an agent func on.
An agent func on is a map from the percept sequence to an ac on.

Agent = Architecture + Agent Program

Type of Agents:
There are five types of the Agents in AI.
1. Simple Reflex Agents
2. Model -Based Reflex Agents.
3. Goal-Based Agents.
4. U lity-Based Agents
5. Learning Agents
6. Mul -System Agents
7. Hierarchical Agents
1. Simple Reflex Agents:
Simple reflex Agents ignore the rest of concept history and act only on the basis of current concept.
The concept history is the history of all that an Agent has realized to date.
The agent func on is based on the condi on-ac on rule. A condi on-ac on rule is a rule that maps a state.
If the condi on is true, then the ac on is taken, else not. This agent func on is successful only when the
environment is completely observable

Problems with Simple reflex agents are:

 Very limited intelligence.
 No knowledge of non-perceptual(conceptual) parts of the state.
 Usually too big to generate and store.
 If there occurs any change in the environment, then the collec on of rules needs to be updated.

2. Model-Based Reflex Agents:

It works by finding a rule whose condi on matches the current situa on.
A model-based agent can handle par ally observable environments by the use of a model about the world.
The agent has to keep track of the internal state which is adjusted by each concept and that depends on the
concept history. Example: A robot that navigates based on a map.

3. Goal-Based Agents:
These kinds of agents take decisions based on how far they are currently from their goal.
The purpose of each of their ac ons is to reduce their distance from the target.
This allows the agent a way to choose among mul ple possibili es, selec ng the one which reaches a goal state.
The goal-based agent’s behavior can easily be changed. Example: A self-driving car naviga ng to a des na on.

4. U lity-Based Agents:
 When there are mul ple possible alterna ves (Op ons), then to decide which one is best, u lity-based agents
are used. They choose ac ons based on a preference (u lity) for each state. The agent uses a u lity func on to
evaluate and rank different states or outcomes. The func on assigns numerical values to outcomes, represen ng
their desirability or u lity.

5. Learning Agent:
A learning agent in AI is the type of agent that can learn from its past experiences or it has learning capabili es.
It starts to act with basic knowledge and then is able to act and adapt automa cally through learning. A learning
agent has mainly four conceptual components:
1. Learning element: It is responsible for making improvements by learning from the environment.
2. Cri c: The learning element takes feedback from cri cs which describes how well the agent is doing with respect
to a fixed performance standard.
3. Performance element: It is responsible for selec ng external ac on.
4. Problem Generator: This component is responsible for sugges ng ac ons that will lead to new and informa ve
5. experiences.

6. Mul -Agent System:

These agents interact with other agents to achieve a common goal. They may have to coordinate their ac ons and
communicate with each other to achieve their objec ve.
These agents may be autonomous or semi-autonomous and are capable of understood their environment, making
decisions, and taking ac on to achieve the common objec ve.
MAS can be used in a variety of applica ons, including transporta on systems, robo cs, and social networks.
A Mul -Agent System is a system composed of mul ple interac ng agents, which can be either:
 Homogeneous: All agents have similar capabili es or goals.
 Heterogeneous: Agents differ in their abili es, roles, or objec ves.
Mul -agent systems are a powerful tool in ar ficial intelligence that can help solve complex problems and improve
efficiency in a variety of applica ons.

7. Hierarchical Agents:
Hierarchical agents are agents that are organized into a hierarchy, with high-level agents overseeing the behavior
of lower-level agents. The high-level agents provide goals and constraints, while the low-level agents carry out
specific tasks. This structure allows for more efficient and organized decision-making in complex environments.
Hierarchical agents can be implemented in a variety of applica ons, including robo cs, manufacturing, and
transporta on systems.

Characteris c of AI:
Ar ficial intelligence (AI) has many characteris cs, including:
 Learning: AI can learn from data and improve its performance over me. For example, AI models can improve
their ability to iden fy objects in images by analyzing new data.
 Problem-solving: AI can solve complex problems, even those that humans can't.
 Reasoning: AI can think and make decisions.
 Percep on: AI can sense its environment.
 Adaptability: AI can adapt to new situa ons.
 Automa on: AI can automate repe ve tasks.
 Data handling: AI can process large amounts of data.
 Natural language processing (NLP): AI can understand and communicate with humans in natural language.
 Self-correc on: AI can improve its accuracy over me.
 Efficiency: AI can perform tasks faster, more accurately, and more consistently than humans.

Machine learning and Deep learning:

Deep learning is a subset of machine learning, and both are branches of ar ficial intelligence (AI):

Machine learning
A broader term that includes various techniques, including deep learning. Machine learning algorithms can
process large amounts of data, iden fy pa erns, and predict outcomes.

Deep learning
Uses ar ficial neural networks to process and analyze informa on. Deep learning algorithms are inspired by the
human brain and are used for complex tasks like image classifica on and object detec on

State Space Search:

State space search is a problem-solving technique used in Ar ficial Intelligence (AI) to find the solu on path from
the ini al state to the goal state by exploring the various states.
State Space Representa on is a crucial step in the state space search process as it determines the efficiency of
the search algorithm. it involves transi on from ini al state to a goal state through a series of valid moves.

Component of state space search:

1. State: A specific configura on or condi on of the problem at any given me.
2. State space: It is a set of all possible states that can be reached by applying valid ac ons.
3. Ini al State: The star ng point of the search process.
4. Goal State: The desired solu on or final configura on of the problem.
5. Operators/Ac ons: It is a set of rules that is how to proceed form one state to another state.
6. Path: It is a sequence state connected by a transac on that leads from the ini al state to the goal state.
7. Solu on: This is the final path that gave results and sa sfied all results.

Step in state space search:

1.define the problem: it includes ini al state, goal state and valid ac ons
2.represent the state space: in this we use graph or tree to represent all the state
3. Search for solu on: in this we find the shortest path using DFS, BFS, a* hastrick seer

Search and control strategies:

Most of the me, these agents perform some kind of search algorithm in the background in order to achieve
their tasks.
A search problem consists of:
o A State Space. Set of all possible states where you can be.

o A Start State. The state from where the search begins.

 o A Goal State. A func on that looks at the current state returns whether or not it is the goal state.
The Solu on to a search problem is a sequence of ac ons, called the plan that transforms the start state to the
goal state.

Search and control strategies:

Search and control strategies in AI refer to the methods used to navigate through problem spaces to find solu ons
or make decisions.

1. Search Strategies:
These are algorithms that explore the possible states or configura ons to find a solu on. Some common search
strategies include:
Type of search algorithm:
There are two types of search algorithms in ar ficial intelligence.

1. Uninformed Search:
This type does not have any addi onal informa on about the goal state.
Uninformed search is also called Blind search.
These algorithms can only generate the successors and differen ate between the goal state and non-goal state.
Examples include:
1. Depth First Search
2. Breadth First Search
3. Uniform Cost Search
1. Breadth-First Search (BFS):
Breadth-first search (BFS) is an algorithm for traversing or searching tree or graph data structures.
This strategy explores all the nodes at the present depth level before moving on to the nodes at the next depth
level. It guarantees the shortest path in an unweighted graph.

2. Depth-First Search (DFS):

Depth-first search (DFS) is an algorithm for traversing or searching tree or graph data structures.
This approach starts at root node and explores as far down a branch as possible before backtracking. It uses a
stack (or recursion) and can be more memory-efficient than BFS.

3. Uniform Cost Search:

UCS is different from BFS and DFS because here the costs come into play. In other words, traversing via
different edges might not have the same cost It's op mal for weighted graphs.

2. Informed Search: This type uses heuris cs to guide the search process toward the goal more efficiently.
Examples include:
- A* Search: Combines features of BFS and heuris cs to find the least-cost path to the goal.
- Greedy Best-First Search: Chooses the path that seems to be the most promising based on a heuris c.

2. Control Strategies:
These strategies determine the order in which nodes are expanded in the search space and how the search is
conducted. They can be:
 1. Sta c Control:
The strategy is fixed before the search begins. For example, always using BFS or DFS.
2. Dynamic Control:
The strategy can change based on the current state of the search process. For example, switching to a different
search method if the current one is not yielding results.

Produc on System:
A produc on system in AI is a type of computer program that uses a set of rules (produc ons) to make decisions
and solve problems. A produc on system is a framework for building AI applica ons that can reason and make
decisions based on a set of rules and a knowledge base. This approach is widely used in expert systems and other
AI applica ons.

It consists of three main components:

1. Knowledge Base:
This is where the informa on about the problem domain is stored.
It includes facts and data that the system can use to make inferences.

2. Set of Produc on Rules:

These are condi onal statements, o en in the form of "if-then" rules. For example, "If the temperature is above
100 degrees, then turn on the cooling system."
The system checks these rules against the current state of the knowledge base.

3. Inference Engine:
This is the component that applies the produc on rules to the knowledge base to derive new informa on or make
decisions. It processes the rules and determines which ones are applicable based on the current facts.

Problem: - Water- jug problem:

The water jug problem is a classic problem in AI and computer science that involves two jugs with different
capaci es and the goal of measuring out a specific amount of water using these jugs.

Problem Defini on:

 Two jugs with fixed capacities:

Jug A has a capacity of a liters.

Jug B has a capacity of b liters.
 A target amount of water, T, that must be measured exactly.
 Unlimited water supply (e.g., a tap) and the ability to empty the jugs.
Allowed Opera ons:

1. Fill a jug completely.

2. Empty a jug completely.
3. Pour water from one jug to the other until:
 The first jug is empty, or
The second jug is full.

Example
Problem: Measure 4 liters of water using a 3-liter jug (Jug A) and a 5-liter jug (Jug B).
Steps to Solve:
Ini al State: (0,0) (0, 0) (0,0)
Fill Jug B: (0,5) (0, 5) (0,5)
Pour from Jug B to Jug A: (3,2) (3, 2) (3,2)
Empty Jug A: (0,2) (0, 2) (0,2)
Pour from Jug B to Jug A: (2,0) (2, 0) (2,0)
Fill Jug B: (2,5) (2, 5) (2,5)
Pour from Jug B to Jug A: (3,4) (3, 4) (3,4)
Goal Achieved: Jug B contains exactly 4 liters of water.

Missionary and cannibal problem:

The Missionary and Cannibal problem is another classic problem in AI and problem-solving. In this scenario, you have
three missionaries and three cannibals who need to cross a river using a boat that can carry at most two people at a
me. The challenge is to ensure that at no point do the cannibals outnumber the missionaries on either side of the
river, as this would lead to the missionaries being eaten.

Rules of the problem:

1. The boat can carry a maximum of two people.
2. If at any point the number of cannibals exceeds the number of missionaries on either side of the river, the
cannibals will eat the missionaries.
3. Everyone starts on one side of the river, and the goal is to get everyone across safely.
Problem Representa on
1. State Representa on:
Each state is represented as (M, C, B) (M, C, B) (M, C, B), where:
o M: Number of missionaries on the le bank.
o C: Number of cannibals on the le bank.
o B: Posi on of the boat (000 for the le bank, 111 for the right bank).

Example: (3,3,0) (3, 3, 0) (3,3,0) means 3 missionaries, 3 cannibals, and the boat are on the le bank.
2. Ini al State: (3,3,0) (3, 3, 0) (3,3,0)
3. Goal State: (0,0,1) (0, 0, 1) (0,0,1)
Rules and Constraints
1. The boat can hold one or two people.
2. The boat cannot cross the river without passengers.
3. At no me can cannibals outnumber missionaries on either side of the river unless there are no missionaries on
that side.
Solu on Approach
State-Space Search
1. Generate States: Iden fy all valid transi ons from the current state by moving 1 or 2 people in the boat.
2. Check Validity: Ensure the new state sa sfies all constraints (e.g., missionaries are not outnumbered).
3. Search Strategy:
o Use Breadth-First Search (BFS) to find the shortest sequence of moves.
o Use Depth-First Search (DFS) for any valid sequence.

Example Solu on
Steps:
1. Start at (3,3,0) (3, 3, 0) (3,3,0):
o Move 2 cannibals to the right: (3,1,1)(3, 1, 1)(3,1,1)
2. (3,1,1) (3, 1, 1) (3,1,1):
o Move 1 cannibal back: (3,2,0) (3, 2, 0) (3,2,0)
3. (3,2,0) (3, 2, 0) (3,2,0):
o Move 2 cannibals to the right: (3,0,1) (3, 0, 1)(3,0,1)
4. (3,0,1) (3, 0, 1) (3,0,1):
o Move 1 cannibal back: (3,1,0) (3, 1, 0) (3,1,0)
5. (3,1,0)(3, 1, 0)(3,1,0):
o Move 2 missionaries to the right: (1,1,1)(1, 1, 1)(1,1,1)
6. (1,1,1)(1, 1, 1)(1,1,1):
o Move 1 missionary and 1 cannibal back: (2,2,0)(2, 2, 0)(2,2,0)
7. (2,2,0)(2, 2, 0)(2,2,0):
o Move 2 missionaries to the right: (0,2,1)(0, 2, 1)(0,2,1)
8. (0,2,1)(0, 2, 1)(0,2,1):
o Move 1 cannibal back: (0,3,0)(0, 3, 0)(0,3,0)
9. (0,3,0)(0, 3, 0)(0,3,0):
o Move 2 cannibals to the right: (0,1,1)(0, 1, 1)(0,1,1)
10. (0,1,1)(0, 1, 1)(0,1,1):
o Move 1 cannibal back: (0,2,0)(0, 2, 0)(0,2,0)
11. (0,2,0)(0, 2, 0)(0,2,0):
o Move 2 cannibals to the right: (0,0,1)(0, 0, 1)(0,0,1)
Goal Achieved: All missionaries and cannibals are on the right bank.

Monkey and Banana Problem:

The Monkey and Banana Problem is a classic AI problem that demonstrates problem-solving, planning, and state-
space representa on. It involves a monkey a emp ng to retrieve bananas hung from the ceiling, which are out of
reach.
Problem Descrip on
1. A monkey is in a room where:
o A bunch of bananas is hanging from the ceiling, beyond the monkey's direct reach.
o A box is available in the room that the monkey can climb to reach the bananas.
o The monkey can perform several ac ons, such as moving, pushing the box, climbing the box, or grabbing
the bananas.
2. Goal: The monkey needs to grab the bananas.
Problem Representa on
1. Ini al State:
o Monkey is on the floor.
o The box is at a different loca on.
o Bananas are hanging from the ceiling.
2. Goal State:
o Monkey is on the box, directly under the bananas, and holding the bananas.
3. Possible Ac ons:
o Move: The monkey can move to a different loca on.
o Push: The monkey can push the box to a different loca on.
o Climb: The monkey can climb onto the box.
o Grab: The monkey can grab the bananas if it is on the box and directly under the bananas.

Solu on Approach
The solu on involves planning a sequence of ac ons to achieve the goal.
1. Ini al State:
 o Monkey is at Loca on A.
o Box is at Loca on B.
o Bananas are hanging from the ceiling at Loca on C.
2. Plan:
1. Move the monkey to Loca on B (where the box is).
2. Push the box to Loca on C (under the bananas).
3. Climb onto the box.
4. Grab the bananas.

Example Solu on
States:
 S1: Monkey at A, Box at B, Bananas at C.
 S2: Monkey at B, Box at B, Bananas at C (a er moving to the box).
 S3: Monkey at C, Box at C, Bananas at C (a er pushing the box under the bananas).
 S4: Monkey on the box, Bananas at C (a er climbing the box).
 S5: Monkey holding the bananas (goal achieved).

Applica on of AI:
Ar ficial intelligence (AI) has many applica ons, including:
 Healthcare: AI can help doctors diagnose diseases and develop new treatments. AI can analyze pa ent data to
iden fy pa erns and rela onships that can help doctors develop treatment plans.
 Automo ve: AI can improve safety and efficiency in the automo ve industry. For example, AI systems can enable
self-driving cars to navigate, detect obstacles, and make driving decisions.
 Retail: AI can help retailers personalize the shopping experience for customers. AI can analyze customer behavior,
preferences, and purchase history to offer tailored product sugges ons.
 So ware development: AI can automate many processes in so ware development, DevOps, and IT. For example,
AI-powered monitoring tools can help flag poten al anomalies in real me.
 Virtual assistants: AI can be used in virtual assistants like Siri and Alexa. Google Assistant is an example of a virtual
assistant that uses natural language processing to support both voice and text commands.
 Security systems: AI can be used in security systems for image and facial recogni on.
 Chatbots: AI can be used in chatbots for customer service.
 Recommenda on systems: AI can be used in recommenda on systems used in e-commerce pla orms.
 Fraud detec on: AI can be used in financial ins tu ons for fraud detec on.
 Unit – 2
(Searching)
Searching for Solu ons in Ar ficial Intelligence (AI) refers to exploring a set of possible ac ons, states, or paths to
iden fy a way to achieve a specified goal. AI systems solve problems by naviga ng through a state space (a set of all
possible states) using search algorithms.
1. Machine Learning:
This involves training algorithms on large datasets to recognize pa erns and make predic ons. Common techniques
include supervised learning, unsupervised learning, and reinforcement learning.
2. Natural Language Processing (NLP):
This area focuses on the interac on between computers and human language. Solu ons include chatbots, language
transla on, sen ment analysis, and text summariza on.
3. Computer Vision:
AI solu ons in this field allow machines to interpret and understand visual informa on from the world.
Applica ons include facial recogni on, object detec on, and image classifica on.

4. Robo cs:
AI is used to enhance the capabili es of robots, enabling them to perform tasks autonomously. This includes
everything from industrial robots in manufacturing to drones for delivery services.

5. Recommenda on Systems:
These systems analyze user behavior and preferences to suggest products, services, or content. They are widely
used in e-commerce and streaming pla orms.

6. Predic ve Analy cs:

AI can forecast future trends based on historical data. This is used in various industries, including finance for stock
predic ons and healthcare for pa ent outcomes.

Uniformed search strategies:

Uniformed search strategies, also known as blind search strategies, are algorithms used in ar ficial intelligence to
explore a search space without any addi onal informa on about the goal's loca on.
types of uniformed search strategies:
1. Breadth-First Search (BFS):
- Descrip on: This algorithm explores all the nodes at the present depth level before moving on to nodes at the
next depth level.
- Characteris cs: It guarantees finding the shortest path in terms of the number of edges, but it can consume a lot
of memory as it stores all the nodes at the current level.
- Use Case: Useful for problems where the solu on is expected to be close to the root of the search tree.
2. Depth-First Search (DFS):
- Descrip on: This algorithm explores as far down a branch as possible before backtracking. It goes deep into the
search tree and only backtracks when it hits a dead end.
- Characteris cs: It uses less memory than BFS but does not guarantee the shortest path. It can get stuck in infinite
loops if the search space is infinite.
- Use Case: Suitable for problems with large search spaces or where solu ons are expected to be deep in the tree.

3. Depth-Limited Search:
- Descrip on: This is a variant of DFS that imposes a limit on the depth of the search. If the limit is reached, the
search backtracks.
- Characteris cs: It prevents infinite loops and can be more efficient than standard DFS.
- Use Case: Useful when you have some knowledge about the maximum depth of the solu on.

4. Itera ve Deepening Search:

- Descrip on: This combines the benefits of BFS and DFS by repeatedly performing depth-limited searches with
increasing depth limits.
- Characteris cs: It finds the shortest path while using less memory than BFS, as it only stores nodes along the
current path.
- Use Case: Ideal for large search spaces where the depth of the solu on is unknown.

5. Uniform Cost Search:

- Descrip on: This algorithm expands the node with the lowest path cost. It is similar to BFS but takes into account
the cost of the path to reach each node.
- Characteris cs: It guarantees the op mal solu on if the path costs are non-nega ve.
- Use Case: Useful when the cost of moving between nodes varies.

Informed search Strategies:

Informed search strategies, also known as heuris c search strategies, u lize addi onal informa on (heuris cs) to
guide the search process more efficiently compared to uniformed search strategies. These heuris cs are essen ally
educated guesses about which paths are likely to lead to a goal state more quickly.

informed search strategies:

1. Heuris c Func on (h(n)):
 - Descrip on: A heuris c func on es mates the cost from a given node \( n \) to the goal. The accuracy of this
func on greatly influences the efficiency of the search.
- Characteris cs: A well-designed heuris c can significantly reduce the number of nodes explored.

2. A* Search Algorithm:
The A* algorithm is a pathfinding algorithm used in ar ficial intelligence (AI) to find the most efficient route
between two points. It's a graph traversal algorithm that uses a heuris c func on to es mate the cost of reaching a
goal from a given node.

 The A* algorithm is used in map traversal to find the shortest path.

 It's also used to help robots find their own course.
Why it's important:

 The A* algorithm is efficient and op mal.

 It's versa le and can solve real-world pathfinding problems.

3. Best-First Search:
- Descrip on: This algorithm selects nodes based solely on the heuris c func on \ (h (n) \), aiming to expand the
node that appears to be closest to the goal.
- Characteris cs: It can be faster than A* in some cases but does not guarantee an op mal solu on.
- Use Case: Useful when a quick solu on is needed, and op mality is not a concern.

4. Hill Climbing:
- Descrip on: This is a local search algorithm that con nuously moves towards the direc on of increasing value (or
decreasing cost) based on the heuris c.
- Characteris cs: It can get stuck in local maxima, plateaus, or ridges, which might prevent finding the global
op mum.
- Use Case: Useful in op miza on problems where the search space is large.

Constraint Sa sfac on:

Constraint sa sfac on is a process in ar ficial intelligence (AI) that involves finding a solu on to a problem by
assigning values to variables that sa sfy a set of constraints:
Here are some key aspects of constraint sa sfac on:

 Variables:
Variables in a CSP are the objects that must have values assigned to them in order to sa sfy a par cular set
of constraints. Boolean, integer, and categorical variables are just a few examples of the various types
of variables. Example: Student, Teachers, subject,
  Domains:
The range of poten al values that a variable can have been represented by domains. Depending on the
issue, a domain may be finite or limitless. Example: Time slot.

 Constraints
The guidelines that control how variables relate to one another are known as constraints. Constraints in a
CSP define the ranges of possible values for variables. Unary constraints, binary constraints, and higher-
order constraints are example of constraint.
Example: 1. Two slots of same not given to teacher/Student.

Types of Constraint Sa sfac on Problems:

CSPs can be classified into different types based on their constraints and problem characteris cs:
1. Binary CSPs: In these problems, each constraint involves only two variables. For example, in a scheduling problem,
the constraint could specify that task A must be completed before task B.
2. Non-Binary CSPs: These problems have constraints that involve more than two variables. For instance, in a sea ng
arrangement problem, a constraint could state that three people cannot sit next to each other.
3. Hard and So Constraints: Hard constraints must be strictly sa sfied, while so constraints can be violated, but at
a certain cost. This dis nc on is o en used in real-world applica ons where not all constraints are equally
important.

CSP Algorithms:
The most commonly used CSP algorithms:

1. Backtracking algorithm:
The backtracking algorithm is a basic algorithm used to solve constraint sa sfac on problems (CSPs) in ar ficial
intelligence (AI).
Explana on

 The backtracking algorithm assigns values to variables one by one.

 At each step, the algorithm checks if the constraints are sa sfied.
 The algorithm only checks the constraints between the current variable and the past variables.

2. Forward-Checking Algorithm:
The Forward-Checking Algorithm is a technique used in constraint sa sfac on problems (CSPs), which are problems
where you need to find values for variables that sa sfy certain constraints. This algorithm is par cularly useful in
scenarios like scheduling, map coloring, and puzzle solving.
Here's how the Forward-Checking Algorithm works:
1. Variable Assignment: When you assign a value to a variable, the algorithm checks the constraints that involve that
variable and the unassigned variables.
2. Domain Reduc on: For each unassigned variable that is connected to the assigned variable through a constraint,
the algorithm removes values from its domain that are inconsistent with the assigned value.
3. Early Detec on of Failure: If at any point an unassigned variable has no remaining values in its domain a er the
domain reduc on step, the algorithm detects that the current assignment cannot lead to a solu on. It then
backtracks to try a different assignment for the previous variable.
4. Backtracking: The algorithm con nues this process of assigning values, reducing domains, and backtracking un l
either a solu on is found or all possibili es are exhausted.

Problem Reduc on:

Problem reduc on is a technique in ar ficial intelligence (AI) that breaks down complex problems into smaller, more
manageable subproblems. This approach is used to simplify problem-solving and make it easier for AI systems to find
solu ons.

Why it's important:

 Complexity management:
Problem reduc on helps AI systems handle complex problems that are difficult to solve directly

 Efficient search and planning:

Problem reduc on reduces the search space, making it easier to find solu ons
How it works:
1. Iden fy the main problem
2. Break it down into smaller subproblems
3. Solve the subproblems independently
4. Combine the solu ons to solve the original problem

Characteris c of problem Reduc on:

1. Breaking down problem into subproblem:
For AI if the problem is deficult into solve in one go than it breaks its into smaller parts.
Solving these subproblem individual than combine there solu on. Ex finding a best trvel rout in
one step.
Game Playing Adersial Search:
Game-playing adversarial search in AI refers to a search strategy used to make op mal decisions in environments
where mul ple agents (players) compete against each other. This is common in games like chess, checkers, c-tac-
toe, or Go, where one player's gain is another's loss. Such environments are modeled as adversarial games, where
agents seek to maximize their own payoff while minimizing that of their opponents.

Here’s how it works:

1. Adversarial Nature: In these games, one player's gain is another player's loss. So, the AI has to consider not just its
own moves but also predict the opponent's moves.
2. Search Trees: The AI constructs a search tree where each node represents a game state. The branches show
possible moves from that state. The AI explores these nodes to find the best possible outcome.
3. Minimax Algorithm: This is a popular method used in adversarial search. The AI assumes that the opponent will
always make the best possible move to minimize the AI's score. The AI tries to maximize its own score while
minimizing the opponent's score.
4. Alpha-Beta Pruning: To make the search process more efficient, the AI can use alpha-beta pruning. This technique
helps it ignore branches of the tree that won’t affect the final decision, speeding up the search without losing the
op mal move.
5. Heuris c Evalua on: If the search tree is too large to explore fully, the AI uses heuris cs to evaluate game
posi ons. This helps it make good decisions based on the current state of the game rather than exploring every
possible outcome.

Problem in Game-Playing:
Game playing in AI faces several challenges, including:
1. Complexity of Search Space: Many games have an enormous number of possible moves and game states, making
it difficult for the AI to evaluate all possibili es. For example, chess and Go.
2. Real-Time Decision Making: In many games, players must make decisions quickly. The AI needs to balance
between exploring the search space and making mely decisions, which can be challenging.
3. Uncertainty and Incomplete Informa on: In some games, players may not have complete informa on about the
opponent's strategy or moves, adding an extra layer of complexity to the decision-making process.
4. Opponent Modeling: Understanding and predic ng the opponent's behavior is crucial. Different opponents may
have different strategies, and the AI needs to adapt its approach accordingly.
5. Dynamic Environments: In some games, the environment can change rapidly, which requires the AI to
con nuously adapt its strategy in response to new informa on.
6. Heuris c Limita ons: The effec veness of heuris cs (rules of thumb used to evaluate game states) can vary
greatly, and finding the right heuris cs for a specific game can be challenging.
 Unit – 3
(Knowledge Representa on)

Defini on of Knowledge:
The knowledge refers to the informa on, understanding, and insights that an AI system has acquired and uses to
perform tasks, make decisions, or solve problems. It is o en structured in ways that allow the system to process,
infer, and act effec vely within its domain.
There are a few key concepts related to knowledge in AI:
1. Knowledge Representa on: This is how informa on is structured and stored within an AI system. It can include
various forms such as rules, ontologies, and seman c networks, which help the AI understand and reason about the
informa on.
2. Knowledge Base: This is a collec on of knowledge that an AI system uses to make inferences and provide answers.
It can be built from structured data (like databases) or unstructured data (like text).
3. Inference: This refers to the process by which AI systems use their knowledge to draw conclusions or make
predic ons based on the informa on they have.
4. Learning: AI systems can acquire new knowledge through learning processes, such as machine learning, where
they improve their performance on tasks by analyzing data and iden fying pa erns.

Types of Knowledge:
There are 5 types of Knowledge such as
 2. Declara ve Knowledge.
3. Structured Knowledge.
4. Procedural Knowledge.
5. Meta Knowledge.
6. Heuris c Knowledge.

1. Declara ve Knowledge:
Declara ve Knowledge also known as Descrip ve knowledge, is the type of knowledge which tells the basic
knowledge about something and it is more popular than Procedural Knowledge. It is o en described as "knowing
that". This type of knowledge is explicit and can be easily communicated or ar culated. It includes informa on
such as dates, defini ons, and concepts.

2. Procedural Knowledge:
Procedural Knowledge also known as Interpre ve knowledge, is the type of knowledge in which it clarifies how a
par cular thing can be accomplished. It can be directly applied to any task. It includes rules, strategies,
procedures, agendas, etc. Procedural knowledge depends on the task on which it can be applied.

3.

Difference between Procedural and Declara ve:

S.NO Procedural Knowledge Declara ve Knowledge
1. It is also known as Interpre ve It is also known as Descrip ve knowledge.
knowledge.

2. Procedural Knowledge means how While Declara ve Knowledge means basic knowledge about
a par cular thing can be something.
accomplished.

3. Procedural Knowledge is generally Declara ve Knowledge is more popular.

not used means it is not more
popular.

4. Procedural Knowledge can’t be Declara ve Knowledge can be easily communicated.

easily communicated.

5. Procedural Knowledge is generally Declara ve Knowledge is data oriented in nature.

process oriented in nature.

6. In Procedural Knowledge In Declara ve Knowledge debugging and valida on is easy.

debugging and valida on is not
easy.

7. Procedural Knowledge is less Declara ve Knowledge is more effec ve in compe ve

effec ve in compe ve programming.
programming.
 Approach to Knowledge:
 Rela onal: Uses tables or databases to store informa on about objects and their rela onships
 Inheritable: Uses hierarchies and inheritance to represent general and specific informa on about objects
 Inferen al: Uses formal logic statements to represent knowledge so that AI can infer new informa on
 Procedural: Uses small programs and codes to describe how to perform tasks
 Seman c networks: Uses labeled edges to connect nodes, represen ng knowledge as a graph
 Ontologies: Defines concepts, rela onships, and limita ons within a domain
 Reinforcement learning: Allows AI systems to learn from interac ons with their environment

Knowledge representa on using Preposi onal and predicate logic:

Proposi onal logic in Ar ficial intelligence:
Proposi onal logic (PL) is the simplest form of logic where all the statements are made by proposi ons. A
proposi on is a declara ve statement which is either true or false. It is a technique of knowledge representa on in
logical and mathema cal form. Proposi onal logic is also called Boolean logic as it works on 0 and 1. A proposi on
formula which is always false is called Contradic on.

Types of preposi ons:

There are two types of Proposi ons:
1. Atomic Proposi ons
2. Compound proposi ons
1. Atomic Proposi on: Atomic proposi ons are the simple proposi ons. It consists of a single proposi on
symbol. These are the sentences which must be either true or false.
Example:
1. a) 2+2 is 4, it is an atomic proposi on as it is a true fact.
2. b) "The Sun is cold" is also a proposi on as it is a false fact.
2. Compound proposi on: Compound proposi ons are constructed by combining simpler or atomic proposi ons,
using parenthesis and logical connec ves.
Example:
1. a) "It is raining today, and street is wet."
2. b) "Ankit is a doctor, and his clinic is in Mumbai."

Predicate logic:
 Predicate logic in ar ficial intelligence (AI) is a formal system that uses variables and quan fiers to represent
complex statements and rela onships. It's also called first-order logic.
Predicate:
In predicate logic, a predicate is an expression that describes a property of objects or a rela onship between
objects. It can contain variables, and when values are subs tuted for the variables, it becomes a proposi on.
𝑃(𝑥) is a predicate that means "x > 5".
Quan fier
Quan fiers are words or phrases that indicate how many elements sa sfy a predicate in predicate logic. They are
used to modify variables and formalize English words like "all", "some", "any", and "every".
For example,
∀x R(x): "For all x, x is a cat." (Universal quan fica on)
∃y L(John, y): "There exists a y such that John loves y." (Existen al quan fica on)

Conversion of clause form:

In ar ficial intelligence (AI), conversion to clause form is a process that uses an algorithm to convert statements
into a disjunc ve normal form (CNF). This process is used to prove statements using the resolu on method.
Steps for conver ng to clause form
1. Remove implica ons
2. Change universal and existen al quan fiers
3. Separate clauses connected by ∧
4. Write the clauses as separate rules
5. Standardize variables
6. Put the resul ng expression in CNF
7. Write one clause per line

Why use conversion to clause form?

 Conversion to clause form is used to prove statements using the resolu on method.
 The resolu on method uses proof by refuta on to prove a statement.
 The resolu on method can be used in both proposi onal and first-order predicate logic.

Resolu on in preposi onal logic:

Resolu on in proposi onal logic is a fundamental rule of inference used in automated theorem proving and logic-
based systems.that produces a new clause from two clauses with complementary literals (p and ¬p).
The resolu on rule states:
 P ∨ A,¬P ∨ B
AVB
Where:
 P is a literal, A and B are disjunc ons of other literals.
 The new clause A∨B is called the resolvent.
Example: Resolu on in Proposi onal Logic
Prove: P ⟹ QP \implies QP⟹Q given (P∨R) and (¬R∨Q).
1. Convert to CNF:
o Rewrite P ⟹ Q as ¬P∨Q.
o The set of clauses is: {P∨R, ¬R∨Q, ¬P∨Q, ¬Q}
o Here ¬Q is the nega on of the goal Q.
2. Apply Resolu on:
o Resolve ¬Q with ¬R∨Q: (¬R∨Q) and (¬Q) ⟹ ¬R.
o Resolve ¬Rwith P∨R: (P∨R) and (¬R) ⟹ P.
o Resolve P with ¬P∨Q: (¬P∨Q) and (P) ⟹ Q.
3. Result:
o Q is derived, and since it is the nega on of the goal (¬Q), we reach a contradic on.
o This proves the original statement (P ⟹ Q) is valid.

Resolu on in predicate logic:

Resolu on in predicate logic extends the resolu on rule from proposi onal logic to handle quan fiers, variables,
and predicates. It is widely used in automated theorem proving and logical inference systems, such as Prolog.
Steps for Resolu on in Predicate Logic
1. Convert to Clause Form (CNF):
o Eliminate implica ons and bicondi onals.
o Move nega ons inward using De Morgan's laws.
o Standardize variables to ensure all variables are unique.
o Skolemize to remove existen al quan fiers by introducing Skolem constants or func ons.
o Drop universal quan fiers (implicit for predicate logic in clause form).
o Convert to conjunc ve normal form (CNF) and split into separate clauses.
2. Negate the Goal:
o Add the nega on of the statement you want to prove to the set of clauses.
3. Apply Unifica on:
o Use a unifica on algorithm to match literals with variables by finding a subs tu on that makes the literals
iden cal.
4. Apply Resolu on Rule:
o Resolve clauses containing complementary literals by combining them into a new clause with subs tuted
values.
5. Repeat Un l Contradic on or No Progress:
o Con nue resolving clauses un l an empty clause (⊥\bot⊥) is derived (proof by contradic on) or no further
resolu on is possible.

LISP in AI:
LISP (List Processing) is a programming language that is commonly used in ar ficial intelligence (AI). Every Lisp
procedure is a func on, and when called, it returns a data object as its value. It allows developers to create complex
AI models and algorithms. Lisp is the second-oldest high-level programming language in the world which is invented
by John McCarthy in the year 1958 at the Massachuse s Ins tute of Technology (MIT).

Features of LISP Programming Language:

1. It is a machine-independent language
2. It uses itera ve design methodology and is easily extensible
3. It allows us to create and update the programs and applica ons dynamically.
4. It provides high-level debugging.
5. It supports object-oriented programming.
6. It supports all kinds of data types like objects, structures, lists, vectors, adjustable arrays, set, trees, hash-tables,
and symbols.
7. It is an expression-based language
8. It can support different decision-making statements like if, when, case, and Cond
9. It will also support different itera ng statements like do, loop, loopfor, do mes and dolist.
10. It will support input and output func ons
11. By using lisp we can also create our own func ons.
Hello World program in LISP:

Syntax:
(write-line string)
Example:
; this is a comment
 (write-line "Hello Geeks")

PROLOG:
PROLOG is a programming language that uses logic to solve problems. It has important role in ar ficial intelligence.
Unlike many other programming languages. In prolog, logic is expressed as rela ons (called as Facts and Rules).
Facts and Rules: You write down facts (like "Cats are animals") and rules (like "If something is a cat, then it is an
animal"). PROLOG uses these to make conclusions.
Advantage of PROLOG:
1. It contains a database of data structures in language and human thinking.
2. Its execu on is based on the defini on of predicates.
3. It supports pa ern matching and backtracking.
4. Its rules are created using recursive thinking, which is comparable to an applica ve language.
5. It is declara ve, compact, ra onal, interpre ve, and modular by defini on.
6. It uses simple coding to store and operate data lists.

Unit – 4
(Natural Language Processing)
What is NLP:
NLP stands for Natural Language Processing, which is a part of Computer Science, Human language, and Ar ficial
Intelligence. It is the technology that is used by machines to understand, analyse, manipulate, and interpret human's
languages. It helps developers to organize knowledge for performing tasks such as transla on, automa c
summariza on, Named En ty Recogni on (NER), speech recogni on, rela onship extrac on, and topic
segmenta on.

Advantages of NLP:
 NLP helps users to ask ques ons about any subject and get a direct response within seconds.
 It does not provide unnecessary and unwanted informa on.
 NLP helps computers to communicate with humans in their languages.

Components of NLP:
There are following component of NLP:-
1. Natural Language Understanding (NLU):
Natural Language Understanding (NLU) helps the machine to understand and analyse human language by extrac ng
the metadata from content such as concepts, en es, keywords, emo on, rela ons, and seman c roles. The most
basic form of NLU is parsing, which takes text wri en in natural language and converts it into a structured format that
computers can understand.
2. Natural Language Genera on (NLG):
Natural Language Genera on (NLG) acts as a translator that converts the computerized data into natural language
representa on. It mainly involves Text planning, Sentence planning, and Text Realiza on.

Steps of NLP:
The main steps involved in Natural Language Processing (NLP):
1. Text Input: The process starts with inpu ng the text data that needs to be analyzed. This could be anything from a
sentence to a large corpus(Collec on) of text.
2. Text Preprocessing: This step involves cleaning and preparing the text for analysis. Common preprocessing tasks
include:
- Tokeniza on: Breaking down text into smaller units, like words or sentences.
- Lowercasing: Conver ng all text to lowercase to ensure uniformity.
- Removing Stop Words: Elimina ng common words (like "and," "the," etc.) that may not add significant meaning.
- Stemming/Lemma za on: Reducing words to their base or root form (e.g., "running" to "run").
3. Feature Extrac on: This involves conver ng the text into a numerical format that can be processed by machine
learning algorithms. Techniques include:
- Bag of Words: Represen ng text as a set of words and their frequencies.
- TF-IDF (Term Frequency-Inverse Document Frequency): A sta s cal measure that evaluates the importance of a
word in a document rela ve to a collec on of documents.
4. Model Building: In this step, machine learning or deep learning models are trained using the processed text data.
This could involve supervised learning (using labeled data) or unsupervised learning (finding pa erns in unlabeled
data).
5. Text Analysis: Depending on the goal, various NLP tasks can be performed, such as:
- Sen ment Analysis: Determining the sen ment (posi ve, nega ve, neutral) of the text.
- Named En ty Recogni on (NER): Iden fying and classifying key en es (like names, dates, loca ons) in the text.
- Text Classifica on: Categorizing text into predefined classes or topics.
6. Output Genera on: Finally, the results of the analysis are generated. This could be a summary, classifica on labels,
or even a response in a conversa onal AI system.
7. Evalua on: The performance of the NLP model is evaluated using metrics like accuracy, precision, recall, and F1
score to ensure it meets the desired standards.

Discourse Knowledge:
In Natural Language Processing (NLP), "discourse knowledge" refers to the ability of a system to understand the
meaning of a sentence or phrase by considering the broader context of the conversa on or text, including previous
sentences, implied informa on, and overall narra ve flow, allowing for a deeper interpreta on beyond just individual
words or syntax.
Key points about discourse knowledge:
 Beyond sentence level:
Unlike basic seman c analysis which focuses on individual sentences, discourse analysis looks at how sentences
connect and build upon each other within a larger piece of text.
 Context-dependent meaning:
Discourse knowledge helps iden fy how the meaning of a word or phrase can change depending on the surrounding
context.

Important aspaects:
 Anaphora and Cataphora: Iden fying pronouns or noun phrases that refer back to previously men oned en es
in the text (anaphora) or forward to upcoming en es (cataphora).
 Topic coherence: Understanding the main topic of a conversa on or document and how different sentences relate
to it.
 Implicature: Recognizing implied meaning based on social and cultural norms, even if not explicitly stated.

Applica ons of discourse knowledge in NLP:

 Machine transla on:
Understanding the context of a sentence to produce an accurate transla on that fits the overall narra ve.
 Text summariza on:
Iden fying the key points of a text by considering the overall discourse structure.
 Ques on answering systems:
Providing more accurate answers by taking into account the context of the ques on and the surrounding text.
 Dialogue systems:
Crea ng more natural and coherent conversa ons by maintaining context across mul ple turns.

Pragma c Knowledge:
In Natural Language Processing (NLP), "pragma c knowledge" refers to the ability of a system to understand the
meaning of a sentence based on its context, including the speaker's inten on, social situa on, and surrounding
discourse, going beyond just the literal words used, essen ally allowing the system to interpret the "implied"
meaning within a conversa on or text.
Key points about pragma c knowledge in NLP:
 Context-dependent meaning:
Pragma cs focuses on how the meaning of a sentence can change depending on the situa on and who is speaking,
which is crucial for accurate interpreta on in real-world scenarios.
 Implicatures:
A key aspect of pragma cs is understanding "conversa onal implicatures," where the speaker conveys a meaning
that is not explicitly stated but can be inferred based on context.
 Speech acts:
Analyzing the intended ac on behind a statement, like making a request, giving a command, or asking a ques on, is
another important aspect of pragma c understanding.
Example of pragma c knowledge in NLP:
 "It's cold in here":
 "Can you pass the salt?":
 Complex context modeling:
 World knowledge integra on:
How NLP systems can incorporate pragma c knowledge:
 Large language models (LLMs):
 Training models on large amounts of text data can help them learn contextual nuances and improve pragma c
understanding.
 Dialogue state tracking:
Maintaining informa on about the conversa on flow and current topic can aid in interpre ng implied meanings.
 Explicit knowledge bases:
Integra ng external knowledge sources about the world and social conven ons can enhance pragma c capabili es.

What is Grammer:
A grammar is a set of rules that define a language as a set of permissible(jus fied) word strings. It serves as a
blueprint for construc ng syntac cally correct sentences or meaningful sequences in a formal language.

Representa on of Grammar
Any Grammar can be represented by 4 tuples – <N, T, P, S>
 N – Finite Non-Empty Set of Non-Terminal Symbols.
 T – Finite Set of Terminal Symbols.
 P – Finite Non-Empty Set of Produc on Rules.
 S – Start Symbol (Symbol from where we start producing our sentences or strings).

Produc on rule:
A produc on or produc on rule in computer science is a rewrite rule specifying a symbol subs tu on that can be
recursively performed to generate new symbol sequences. It is of the form α-> β where α is a Non-Terminal
Symbol which can be replaced by β which is a string of Terminal Symbols or Non-Terminal Symbols.
Chomsky hierarchy of Grammer:
The Chomsky hierarchy is a framework in formal language theory that classifies languages (or grammars) into four
categories based on their genera ve power. It provides a systema c way to understand the rela onship between
languages and computa onal models.

The hierarchy consists of four levels:

1. Type 0 (Recursively Enumerable Languages):
 Grammar: Unrestricted grammar.
 Rules: Production rules have the form α→β, where α and β are strings of terminal and non-terminal symbols, and
α≠ϵ (empty string).
 Recognition: Can be recognized by a Turing machine.
 Example: Any language that a Turing machine can compute, e.g., the set of valid programs in a programming
language with infinite resources.

2. Type 1: Context-Sensitive Languages

 Grammar: Context-sensitive grammar.
 Rules: Production rules have the form αAβ→αγβ, where γ≠ϵ, and A is a non-terminal.
 Recognition: Can be recognized by a linear bounded automaton (LBA).
 Characteristics: Every string derivation respects the context of the symbols around it.
 Example: The language {a^n b^n c^n ∣ n ≥ 1}.

3. Type 2: Context-Free Languages

 Grammar: Context-free grammar.
 Rules: Production rules have the form A→, where A is a non-terminal and γ is a string of terminals and/or non-
terminals.
 Recognition: Can be recognized by a pushdown automaton.
 Characteristics: Suitable for languages with nested structures (e.g., parentheses, XML, programming
languages).
Example: The language {a^n b^n | n≥1} (balanced parentheses).

4. Type 3: Regular Languages

 Grammar: Regular grammar.
 Rules: Production rules have the form A→aB or A→a, where A and B are non-terminals, and a is a terminal.
 Recognition: Can be recognized by a finite state automaton (FSA).
 Characteristics: The simplest type of language, with no memory or nesting requirements.
 Example: The language { a^n ∣ n≥0} (strings of repeated symbols).

Transformational Grammer:
Transformational Grammar is a theory of grammar developed by Noam Chomsky in the 1950s as part of his
groundbreaking work in linguistics. It is a framework that seeks to explain how humans produce and understand
sentences by transforming abstract underlying structures into surface structures (the sentences we actually say or
write). It starts with a simple sentence structure, called the "deep structure." And Rules are applied to change this
deep structure into different forms, creating "surface structures."
Examples: For instance, the deep structure "The cat chased the mouse" can be transformed into "Did the cat chase
the mouse?" by applying specific rules.
 Deep Structure vs. Surface Structure:
 Deep Structure:
Represents the underlying, abstract syntactic structure of a sentence. It captures the fundamental
grammatical relationships, such as who is doing what to whom.

 Surface Structure:
Represents the actual form of the sentence as spoken or written. It is derived from the deep structure
through transformational rules.

Case Grammars (FILLMORE’s Grammar):

Case Grammar, proposed by linguist Charles Fillmore in 1968, is a theory of grammar that emphasizes the
relationship between the syntactic structure of a sentence and the semantic roles (or "cases") played by its
constituents. It focuses on the deep structure of sentences and the functions of different elements within them, rather
than purely their syntactic form.

1. Semantic Roles: It categorizes the roles that nouns play in relation to the verb in a sentence. These roles are
called "cases."

2. Some common cases include: -

Agent: The doer of the action (e.g., "The dog" in "The dog chased the ball").
Theme: The entity that is affected by the action (e.g., "the ball" in the same sentence).
Instrument: The means by which the action is performed (e.g., "a stick" in "He hit the ball with a stick").

Semantic Grammar:
Semantic grammar is a linguistic theory that studies the connection between syntax and meaning in a language. In
artificial intelligence (AI), semantic grammar helps machines understand the meaning of words and structures in a
language. It is commonly used in natural language processing (NLP) applications to facilitate understanding.

Context- Free-Grammar (CFG):

Context free grammar is a formal grammar, which is a formal grammar used to describe the syntax of languages,
including programming languages and natural languages. It is widely used in Natural Language Processing (NLP),
compilers, and parsers to analyze and generate sentences or structures based on a defined set of rules.
A Context-Free Grammar is defined as G= (V, T, P, S)
Where:
 G: describes the grammar
 T: describes a finite set of terminal symbols.
 V: describes a finite set of non-terminal symbols
 P: describes a set of production rules
 S: is the start symbol.
Production rule:
a. S → aSa
b. S → bSb
c. S → c
Parsing process:
Parser is a compiler that is used to break the data into smaller elements coming from lexical analysis phase.
A parser takes input in the form of sequence of tokens and produces output in the form of parse tree.
Parsing is of two types: top-down parsing and bottom-up parsing.

Top-down paring:
o The top-down parsing is known as recursive parsing or predictive parsing.
o It starts from the start symbol and ends down on the terminals. It uses left most derivation.
o In the top-down parsing, the parsing starts from the start symbol and transform it into the input symbol.
Top-down parser is classified into 2 types:
1. Recursive descent parser.
2. Non-recursive descent parser.
1. Recursive descent parser :
Recursive descent parser is also known as the Brute force parser or the backtracking parser. It basically
generates the parse tree by using brute force and backtracking techniques.

2. Non-recursive descent parser.

Non-recursive descent parser is also known as LL (1) parser or predictive parser or without backtracking
parser or dynamic parser. It uses a parsing table to generate the parse tree instead of backtracking.

Bottom-Up Parser
Bottom-up Parser is the parser that generates the parse tree for the given input string with the help of grammar
productions by compressing the terminals. It starts from terminals and ends upon the start symbol. It uses the
rightmost derivation in reverse order.

Bottom-up parser is classified into two types:

LR parser:
 This is a bottom-up parser that generates the parse tree for the given string by using unambiguous grammar. It
follows the reverse of the rightmost derivation.
LR parser is classified into four types:
 LR (0)
 SLR (1)
 LALR (1)
 CLR (1)
Operator precedence parser:
Operator Precedence Parser generates the parse tree from given grammar and string but it has the condition that
two consecutive non-terminals and epsilon will never appears on the right-hand side of any production. The
operator precedence parsing technique is applied to Operator grammars.

Transition Network:
Transition networks are graph-based structures used in artificial intelligence (AI) to break down complex tasks into
smaller steps. They are used in natural language processing (NLP) and compiler design.
Types of transition networks
 Recursive transition networks (RTNs)
Recursive Transition Networks (RTNs) are a type of finite state machine used to describe the syntax of
languages. A type of transition network that can handle nested structures in language. RTNs are used to
represent complex structures, such as recursive elements in language.
 Augmented transition networks (ATNs)
Augmented Transition Networks (ATNs) are a type of transition network used for parsing sentences
in natural language processing. A type of transition network that extends finite state machines with
recursive procedures and registers. ATNs can capture hierarchical structures in language, making them
capable of representing complex syntactic constructs. ATNs introduce augmented features which can store
and manipulate extra information as well as permitting recursive transitions into these networks.

Components of RTNs
1. States: Represent points in the parsing process.
2. Transitions: Connect states and are labeled with terminal symbols, non-terminal symbols, or epsilon (ε)
transitions.
3. Recursive Calls: Allow transitions to invoke other RTNs, enabling the representation of recursive grammar
rules.

Application of Natural language processing (NLP)

Natural Language Processing (NLP) has a variety of applications that are quite impactful. Here are some of the
main applications along with explanations:

1. Sentiment Analysis: This is used to determine the emotional tone behind a series of words. Businesses often use
sentiment analysis to understand customer opinions by analyzing reviews and social media mentions.
2. Chatbots and Virtual Assistants: NLP is the backbone of chatbots and virtual assistants like Siri or Google
Assistant. These systems can understand and respond to user queries in a conversational manner.
3. Machine Translation: This application translates text from one language to another. Services like Google
Translate utilize NLP to understand the context and nuances of the source language, ensuring that the translation is as
accurate as possible.
4. Text Summarization: NLP can condense large texts into shorter summaries while retaining the main ideas. This is
useful for quickly grasping the content of lengthy articles or documents without having to read everything in detail.
5. Speech Recognition: This technology converts spoken language into text. It’s used in voice-activated systems and
transcription services, allowing users to interact with devices through voice commands.
6. Information Retrieval: NLP improves search engines by allowing them to better understand user queries and
deliver more relevant results. It helps in interpreting the intent behind search terms, making the search process more
efficient.
7. Text Classification: This involves categorizing text into predefined categories. A common example is spam
detection in emails, where NLP algorithms learn to identify and filter out unwanted messages based on their content.

Eliza System:
The ELIZA system is one of the earliest examples of natural language processing in artificial intelligence. Developed
by Joseph Weizenbaum in the mid-1960s, ELIZA was designed to simulate a conversation with a human by using
pattern matching and substitution techniques.

Key Features of ELIZA:

1. Scripted Conversations: ELIZA operated based on predefined scripts, the most famous of which was the
"DOCTOR" script that mimicked a psychotherapist. It would respond to user inputs with questions or reflections,
often rephrasing what the user said to create the illusion of understanding.

2. Pattern Matching: ELIZA used simple pattern matching to identify keywords in the user's input and generate
appropriate responses.

3. Limited Understanding: While ELIZA could create the appearance of a meaningful conversation, it did not truly
understand the content of the dialogue. Its responses were based on surface-level patterns rather than
comprehension of the underlying meaning.

4. Influence on AI: ELIZA is significant in the history of AI because it demonstrated the potential for machines to
engage in human-like conversation, even with very limited capabilities. It sparked interest in the field of natural
language processing and set the stage for more advanced conversational agents.

How Does ELIZA Work?

1. Input Processing: ELIZA takes user input and searches for specific keywords or phrases.
2. Pattern Matching: When a match is found, the program applies simple rules to generate a response. These rules
are not based on deep understanding but on syntactic structures.
3. Response Generation: The output is a reformulation of the input, often transformed into a question. If no
keywords are detected, ELIZA resorts to generic responses such as “Tell me more” or “Why do you say that?

Lunar System:
The LUNAR System was an early artificial intelligence (AI) system developed in the 1970s by William A. Woods
at Bolt, Beranek, and Newman (BBN). It was designed to allow users to interact with a database of geological
information about the Moon using natural language queries. The system's purpose was to demonstrate the potential
of natural language processing (NLP) and query-answering systems in AI.
Here are some simple points:

Lunar system refers to the Moon and its relationship with Earth. Here are some simple points:

1. The Moon: It's the only natural satellite of Earth, about 3,474 km wide, and orbits Earth at around 384,400 km
away.

2. Phases: The Moon goes through different phases like new moon, full moon, and others, based on how much
sunlight it reflects.

3. Tides: The Moon's gravity affects the ocean, causing tides—this means the water level goes up and down
regularly.

4. Exploration: Humans have sent missions to the Moon, like the Apollo missions, to learn more about it.

5. Surface: The Moon has craters, flat dark areas called maria, and highlands that show its history.
 Unit – 5
Introduction to Export System

Definition of Expert system:

An expert system is a computer program that is designed to solve complex problems and to provide decision-
making ability like a human expert. It is a computer program that uses AI to mimic the judgment and behavior of a
human expert in a specific field. These systems use a knowledge base filled with domain-specific information and
rules to interpret and solve complex problems. It provides solutions to complex problems that typically require
human expertise.

Characteristics of Expert System:

o High Performance: The expert system provides high performance for solving any type of complex problem of a
specific domain with high efficiency and accuracy.
o Understandable: It responds in a way that can be easily understandable by the user. It can take input in human
language and provides the output in the same way.
o Reliable: It is much reliable for generating an efficient and accurate output.
o Highly responsive: ES provides the result for any complex query within a very short period of time.

Components of Expert System

An expert system mainly consists of three components:
o User Interface
o Inference Engine
o Knowledge Base

1. User Interface:
it is an interface that helps a non-expert user to communicate with the expert system to find a solution. With
the help of a user interface, the expert system interacts with the user, takes queries as an input in a readable format,
and passes it to the inference engine. After getting the response from the inference engine, it displays the output to
the user.

2. Inference Engine (Rules of Engine):

o The inference engine is known as the brain of the expert system as it is the main processing unit of the system. It
applies inference rules to the knowledge base to derive a conclusion or deduce new information. With the help of an
inference engine, the system extracts the knowledge from the knowledge base.
o There are two type of inference Engine:
o Deterministic Inference engine: The conclusions drawn from this type of inference engine are assumed to be true.
It is based on facts and rules.
o Probabilistic Inference engine: This type of inference engine contains uncertainty in conclusions, and based on
the probability.
Inference engine uses the below modes to derive the solutions:
o Forward Chaining: It starts from the known facts and rules, and applies the inference rules to add their conclusion
to the known facts.
o Backward Chaining: It is a backward reasoning method that starts from the goal and works backward to prove the
known facts

3. Knowledge base:
o The knowledgebase is a type of storage that stores knowledge acquired from the different experts of the particular
domain. It is considered as big storage of knowledge.
o It is similar to a database that contains information and rules of a particular domain or subject.

Components of Knowledge Base:

o Factual Knowledge: The knowledge which is based on facts and accepted by knowledge engineers comes under
factual knowledge.
o Heuristic Knowledge: This knowledge is based on practice, the ability to guess, evaluation, and experiences.
o Knowledge Representation: It is used to formalize the knowledge stored in the knowledge base using the If-else
rules.
o Knowledge Acquisitions: It is the process of extracting, organizing, and structuring the domain knowledge,
specifying the rules to acquire the knowledge from various experts, and store that knowledge into the knowledge
base.

Development of Expert System:

The development of an expert system involves several key steps and components. Here’s a comprehensive overview:

1. Define the Problem Domain: Identify the specific area where the expert system will be applied. Understanding
the problem domain is crucial for gathering relevant knowledge.

2. Knowledge Acquisition: This step involves collecting information from human experts in the field. Techniques
for knowledge acquisition can include interviews, questionnaires, and observation. The goal is to gather enough data
to create a robust knowledge base.
 3. Knowledge Representation: Once the knowledge is acquired, it needs to be represented in a format that the
expert system can use. Common methods for knowledge representation include:

- Rules: Using IF-THEN statements.

- Frames: Data structures that hold knowledge in a way similar to objects in programming.

- Semantic Networks: Graph structures for representing knowledge in terms of entities and their relationships.

4. Designing the Inference Engine: The inference engine is the core component that applies logical rules to the
knowledge base. It can use forward chaining or backward chaining methods to derive conclusions based on the
provided facts.

5. User Interface Development: A user-friendly interface is essential for users to interact with the expert system.
This could include input forms, graphical displays, and output reports.

6. Testing and Validation: The system must be rigorously tested to ensure it provides accurate and reliable results.
This involves comparing the system’s output with that of human experts and making adjustments as necessary.

7. Implementation: Once validated, the expert system can be deployed for use. This may involve integrating it with
existing systems or making it available as a standalone application.

8. Maintenance and Updates: An expert system requires ongoing maintenance to ensure it remains accurate and
relevant. This includes updating the knowledge base as new information becomes available and refining the
inference engine as needed.

Applications of Expert System:

o In designing and manufacturing domain

It can be broadly used for designing and manufacturing physical devices such as camera lenses and automobiles.
o In the knowledge domain
These systems are primarily used for publishing the relevant knowledge to the users. The two popular ES used for
this domain is an advisor and a tax advisor.
o In the finance domain
In the finance industries, it is used to detect any type of possible fraud, suspicious activity, and advise bankers that
if they should provide loans for business or not.
o In the diagnosis and troubleshooting of devices
In medical diagnosis, the ES system is used, and it was the first area where these systems were used.
o Planning and Scheduling
The expert systems can also be used for planning and scheduling some particular tasks for achieving the goal of that
task.

Example of Expert System:

Expert systems that are in use include the following examples:

 CaDet (Cancer Decision Support Tool) is used to identify cancer in its earliest stages.
 DENDRAL helps chemists identify unknown organic molecules.
 DXplain is a clinical support system that diagnoses various diseases.
 MYCIN identifies bacteria such as bacteremia and meningitis, and recommends antibiotics and dosages.
 Structure of Expert System:
The structure of an expert system in AI typically consists of several key components, each playing a crucial role in
its functionality. Here’s a breakdown of the main components:

1. Knowledge Base: This is the core of the expert system, containing all the domain-specific knowledge. It
includes facts and rules. The knowledge can be represented in various forms, such as:
- Production Rules: IF-THEN statements that describe the relationships between different concepts.
- Frames: Data structures that hold knowledge in a structured format, similar to objects in programming.
- Semantic Networks: Graph structures that represent knowledge in terms of entities and their relationships.

2. Inference Engine: This component applies logical rules to the knowledge base to derive new information or
make decisions. It processes the input data and uses reasoning techniques, such as:
- Forward Chaining: Starting with the available facts and applying rules to infer conclusions.
- Backward Chaining: Starting with a goal and working backward to see if the available facts support it.

3. User Interface: The user interface allows users to interact with the expert system. The interface can include
forms, menus, and graphical displays.

4. Explanation Facility: This component provides users with explanations of the reasoning process. It helps users
understand how the system arrived at a particular conclusion or recommendation.

5. Knowledge Acquisition Module: This is a subsystem that helps in updating and refining the knowledge base.

6. Knowledge Base Management System (KBMS): This component manages the knowledge base, allowing for
storage, retrieval, and updating of knowledge.

Types of Expert Systems in AI:

There are five types of Expert System.

1. Rule-Based Expert Systems

 Description: Use a set of “if-then” rules to process data and make decisions. These rules are typically written by
human experts and capture domain-specific knowledge.
 Example: MYCIN, an early system for diagnosing bacterial infections.

2. Frame-Based Expert Systems

 Description: Represent knowledge using frames, which are data structures similar to objects in programming. Each
frame contains attributes and values related to a particular concept.
 Example: Systems used for knowledge representation in areas like natural language processing.

3. Fuzzy Logic Systems

 Description: Handle uncertain or imprecise information using fuzzy logic, which allows for partial truths rather
than binary true/false values.
 Example: Fuzzy control systems for managing household appliances like washing machines and air conditioners.

4. Neural Network-Based Expert Systems

 Description: Use artificial neural networks to learn from data and make predictions or decisions based on learned
patterns. They are often used for tasks involving pattern recognition and classification.
 Example: Deep learning models for image and speech recognition.
 5. Neuro-Fuzzy Expert Systems
 Description: Neuro-Fuzzy Expert Systems are a combination of neural networks and fuzzy logic systems. They
leverage the strengths of both methodologies to handle complex problems that involve uncertainty and imprecision.
This hybrid approach helps in dealing with complex problems where both pattern recognition and uncertain
reasoning are required.
 Example: Automated control systems that adjust based on uncertain environmental conditions or financial
forecasting models that handle both quantitative data and fuzzy inputs.

Difference between Human expert and Expert system:

 Benefits of Expert System:

Expert systems in AI offer several benefits that make them valuable tools in various fields. Here are some of the key
advantages:

1. Consistency: Expert systems provide consistent responses and decisions based on the same set of rules and
knowledge.

2. Availability: They can operate 24/7 without fatigue, allowing users to access expert knowledge anytime.

3. Scalability: Expert systems can be scaled to handle large amounts of data and complex problems. They can be
updated with new knowledge and rules without significant downtime.

4. Cost-Effective: By automating decision-making processes. They can also help organizations save time and
resources by streamlining processes

5. Training and Education: They can serve as educational tools, providing training to new employees or students
by simulating expert decision-making processes and offering explanations for their reasoning.

7. Complex Problem Solving: Expert systems can analyze complex data and provide solutions that may be difficult
for humans to derive, enhancing decision-making capabilities in intricate scenarios.

Shells of Expert System:

An expert system shell in AI refers to the framework or software platform that provides the infrastructure for building
and running expert systems. It serves as a foundation for developing customized expert systems by providing tools
and functionalities to represent knowledge, make inferences, and deliver intelligent advice. It makes development
very easy. This contains all the expert system generic logic system logic required to build an expert system.

Limitation of Expert System:

Expert systems in artificial intelligence (AI) have some limitations, including:

 Limited creativity

Expert systems are unable to generate innovative solutions that go beyond their programmed rules.

 Lack of common sense:

Expert systems are unable to understand context beyond their programmed knowledge.

 Dependency on data quality

The accuracy of an expert system's advice depends on the quality of the data in its knowledge base.

 Maintenance

Expert systems require regular updates to their knowledge base.

 Difficulty with ambiguous problems

Menu-driven expert systems may not be able to handle ambiguous problems well.

 Incorrect responses

If the knowledge base contains errors or incorrect information, the expert system may provide incorrect responses.

MYCIN:

MYCIN was an early backward chaining expert system that used artificial intelligence to identify bacteria causing
severe infections, In 1972 work began on MYCIN at Stanford University in California. MYCIN worked with a
knowledge base of about 600 rules and a simple inference engine. It will ask the doctor running the program a long
list of easy yes/no or text-based questions

How did MYCIN work?

 Knowledge base

MYCIN's knowledge base was made up of production rules that represented the clinical decision criteria of infectious
disease experts.

 Inference engine

MYCIN used a simple inference engine to ask doctors questions and rank possible bacteria.

 Explanation system

MYCIN could answer questions in simple English to explain its reasoning and recommendations.

What could MYCIN do?

 Identify bacteria that cause severe infections like bacteremia and meningitis
 Diagnose blood clotting diseases
 Recommend antibiotics and adjust the dosage for a patient's weight
 Explain the reasoning behind its diagnosis and recommendations
 Allow experts to teach the system new therapeutic decision rules

DENDRAL:

DENDRAL was an early expert system developed in the mid-1960s at Stanford University by Edward Feigenbaum,
Bruce G. Buchanan, Joshua Lederberg, and Carl Djerassi. It was designed to assist chemists in identifying the
molecular structure of organic compounds based on mass spectrometry and other analytical data.
The main goal of DENDRAL was to automate the process of molecular structure identification. It aimed to
generate hypotheses about unknown chemical compounds from given empirical data.

How DENDRAL Worked

1. Input: The system received mass spectrometry data of an unknown organic compound.
2. Generation of Hypotheses: It generated possible molecular structures using predefined chemical rules.
3. Evaluation: Applied heuristic rules to analyze and eliminate incorrect structures.
4. Output: Suggested the most probable molecular structure of the compound.