MODULE - 1

1. Define Artificial Intelligence. Explain the Foundations of AI in Detail

Artificial Intelligence (AI) refers to the simulation of human intelligence in machines that are
programmed to think and learn. These systems are capable of performing tasks that typically require
human cognitive functions such as reasoning, learning, problem-solving, perception, and language
understanding. AI can be categorized as:

 Narrow AI: Specialized in one task (e.g., speech recognition, image classification).
 General AI: Possesses the ability to perform any intellectual task that a human can.
 Super AI: Hypothetical AI surpassing human intelligence across all fields.

Foundations of Artificial Intelligence

AI is built upon a multidisciplinary foundation involving the convergence of several core areas:

i. Mathematics

Mathematics formalized AI through logic, computation, and probability, addressing what can be
computed and how to reason with uncertainty.

 Linear Algebra: Forms the basis of many AI models, especially in neural networks.
 Probability and Statistics: Used in decision making, uncertainty modeling, and inference (e.g.,
Bayesian networks).
 Calculus: Crucial for optimization, especially in training machine learning models.
 Graph Theory: Useful in search algorithms and network analysis.

ii. Philosophy

Philosophy addresses fundamental questions about reasoning, the mind, knowledge, and action,
laying the groundwork for AI.

 Key Questions:
o Can formal rules be used to draw valid conclusions?
o How does the mind arise from a physical brain?
o Where does knowledge come from?
o How does knowledge lead to action?
 Topics include the Turing Test, mind-body problem, and logic-based reasoning.
 Philosophical questions underpin debates around AI consciousness and moral decision-
making.

iii. Computer Science

This provides the practical and theoretical backbone of AI.

 Algorithms and Data Structures: Fundamental for search, sorting, and efficient
computation.
 Programming Languages: Languages like Python, Java, and LISP are widely used.
  Software Engineering: For designing, testing, and maintaining AI systems.

iv. Economics

Economics contributed decision-making frameworks for maximizing outcomes, especially under

uncertainty and in multi-agent scenarios.

 Key Questions:
o How should we make decisions to maximize payoff?
o How should we do this when others may not go along?
o How should we do this when the payoff is far in the future?

v. Neuroscience

Neuroscience offers biological inspiration for artificial neural networks and cognitive architectures.

 Study of the brain helps model learning, memory, and perception.

 Computational neuroscience bridges biological systems and AI models.

vi. Psychology

AI aims to replicate human thinking and learning, making insights from these fields invaluable.

 Cognitive Architectures: Like ACT-R and SOAR are designed based on human cognition.
 Learning Theories: Inform how AI systems adapt over time.

vii. Linguistics

Language understanding is a key part of AI (e.g., in chatbots and translation systems).

 Syntax, Semantics, and Pragmatics: Help in building Natural Language Processing (NLP)
systems.
 Speech Recognition and Generation: Are directly related to human-computer
interaction.

viii. Control theory and cybernetics

This provides the physical infrastructure for AI systems, especially in autonomous vehicles and
manufacturing.

 How can artifacts operate under their own control?

 Robotics embodies AI in physical systems, combining perception, motion, and decision-
making.
2. Define Total Turing test, logical positivism, tractable problems, decision theory, Neurons

 Total Turing Test: An enhanced Turing Test where a computer must convince a human
interrogator of its intelligence through text, video signals (testing perception), and physical
object manipulation via a hatch. It requires capabilities in natural language processing,
knowledge representation, reasoning, learning, computer vision, and robotics.
or
The Total Turing Test is an extension of the standard Turing Test. In addition to natural
language conversation, it includes perceptual and motor capabilities. To pass the Total Turing
Test, a machine must be able to:
 Perceive (e.g., see and hear) as humans do.
 Physically interact with the world.
This means the test encompasses computer vision and robotics in addition to linguistic
reasoning.

 Logical Positivism: A philosophical doctrine (Vienna Circle, Carnap) asserting that all
knowledge can be expressed as logical theories linked to sensory observation sentences. It
combines rationalism and empiricism, rejecting metaphysics by requiring statements to be
verifiable or falsifiable.
or
Logical positivism is a philosophical theory that asserts:
 Only statements verifiable through direct observation or logical proof are meaningful.
 Metaphysical, religious, and ethical statements are considered non-cognitive and
meaningless unless empirically testable.
It emphasizes empiricism and formal logic, often associated with the Vienna Circle in the
early 20th century.

 Tractable Problems: Problems solvable in polynomial time, contrasted with intractable

problems where solution time grows exponentially. Tractability is crucial for dividing
intelligent behavior into computationally feasible subproblems.
or
A problem is called tractable if it can be solved in polynomial time—that is, the time required
to solve it grows at most polynomially with the size of the input.
 These problems are considered efficiently solvable.
 Tractability is central in computational complexity theory and distinguishes "feasible"
from "infeasible" problems.

 Decision Theory: A formal framework integrating probability theory (to handle uncertainty)
and utility theory (to quantify preferences) to make decisions that maximize expected payoff.
It’s effective in large economies with minimal agent interactions.
or
Decision theory is the study of principles and models used for making rational choices under
uncertainty.
o It combines probability theory (to model uncertainty) and utility theory (to
model preferences).
 o It is used in fields like economics, artificial intelligence, and statistics to guide
optimal decision-making.

 Neurons : Units in artificial neural networks, modeled after biological neurons. Each neuron
processes inputs via a weighted sum, applies an activation function (e.g., sigmoid, ReLU), and
produces an output, enabling complex pattern recognition in AI systems.
or
A neuron is a basic unit of the nervous system responsible for transmitting information
through electrical and chemical signals.
 Neurons consist of dendrites, a cell body (soma), and an axon.
 In AI and machine learning, "artificial neurons" are simplified mathematical models used
in neural networks to simulate some properties of biological neurons.

3. Explain the interaction between agents and their environments in the context of AI.

In the context of Artificial Intelligence (AI), the interaction between agents and their environments is
fundamental to understanding how intelligent systems operate.

An agent is any entity that can perceive its environment through sensors and act upon that
environment using actuators. The environment is everything external to the agent with which it can
interact or which affects its performance.

How the Interaction Works:

1. Perception: The agent receives inputs from the environment through its sensors. These
inputs are called percepts.
2. Processing/Decision-Making: Based on these percepts and its internal state (which may
include memory, goals, and models), the agent decides what action to take using an agent
function or agent program.
3. Action: The agent then takes an action through its actuators to affect the environment.
4. Feedback Loop: The action changes the state of the environment, which the agent perceives
again, continuing the cycle.

This feedback loop allows the agent to adapt to dynamic conditions and learn from experiences if it's
equipped with learning capabilities.

Types of Environments:

 Fully vs. Partially Observable: Whether the agent can observe the entire state of the
environment.
 Deterministic vs. Stochastic: Whether the outcome of actions is certain or involves
randomness.
 Episodic vs. Sequential: Whether each interaction is independent or depends on
previous actions.
 Static vs. Dynamic: Whether the environment changes while the agent is deliberating.
  Discrete vs. Continuous: Whether the state, percepts, and actions are countable or
range over a continuum.

This interaction model is key in designing and evaluating AI systems for tasks such as robotics, game
playing, autonomous driving, and virtual assistants.

4. Explain Milestones of AI with reference to

I. Neural networks failure to generalize
II. Advent of DENDRAL
III. Emergence of Intelligent Agents

(i) Neural Networks' Failure to Generalize

In the early stages of AI, particularly during the 1960s and 1970s, neural networks were criticized for
their inability to generalize beyond their training data. The classic Perceptron, developed by Frank
Rosenblatt, was limited to linearly separable problems. The book "Perceptrons" by Marvin Minsky
and Seymour Papert (1969) highlighted these limitations, leading to a decline in neural network
research—often referred to as the first AI winter. It wasn't until the 1980s, with the introduction of
multi-layer networks and the backpropagation algorithm, that neural networks regained interest due
to improved generalization capabilities.

(ii) Advent of DENDRAL

DENDRAL, developed in the mid-1960s at Stanford by Edward Feigenbaum, Bruce Buchanan, and
Joshua Lederberg, was one of the first expert systems. It was designed to assist chemists in
identifying molecular structures using mass spectrometry data. DENDRAL marked a major milestone
as it demonstrated that domain-specific knowledge could be encoded into a computer system to
mimic expert-level problem solving. This success helped shift the focus of AI from general problem
solvers to knowledge-based systems.

(iii) Emergence of Intelligent Agents

The concept of intelligent agents emerged prominently in the 1990s and represents a significant
milestone in AI evolution. Unlike rule-based or expert systems, intelligent agents are designed to
perceive their environment, reason about it, and act autonomously to achieve goals. These agents
are foundational in modern AI applications such as personal assistants (e.g., Siri, Alexa), autonomous
robots, and intelligent web services. The agent-based model unified several AI subfields, including
machine learning, planning, perception, and robotics.
5. In detail elaborate with neat diagram, the state space for the vacuum world. Links denote
actions: L = Left,R= Right,S=Suck.

The Vacuum World is a classic

problem in Artificial Intelligence
that illustrates a simple agent-
environment interaction scenario. It
consists of two rooms (labeled A
and B), each of which may be clean
or dirty. The vacuum cleaner can be
in either room and can perform
three actions: Left (L), Right (R), or
Suck (S).

Components of the Vacuum World Problem

The vacuum world is defined as a problem with the following components:

 States: The possible configurations of the environment, determined by:

o The location of the vacuum cleaner (in square A or square B).
o The dirt status of each square (A and B can each be Clean or Dirty).
 Initial State: Any of the possible states (e.g., vacuum in A, both squares dirty).
 Actions: The agent can perform:
o L (Left): Move the vacuum to square A.
o R (Right): Move the vacuum to square B.
o S (Suck): Remove dirt from the current square, making it clean.
 Transition Model: Specifies the outcome of each action in each state:
o L: If the vacuum is in B, it moves to A; if already in A, no change.
o R: If the vacuum is in A, it moves to B; if already in B, no change.
o S: If the current square is dirty, it becomes clean; if clean, no change.
 Goal Test: The environment is in a state where both squares are clean (i.e., [A, Clean, Clean]
or [B, Clean, Clean]).
 Path Cost: Each action has a cost of 1.

Number of States:

 Vacuum location: 2 possibilities (A or B).

 Dirt status: Each square (A, B) can be Clean or Dirty, so 2² = 4 combinations ([Clean, Clean],
[Clean, Dirty], [Dirty, Clean], [Dirty, Dirty]).
 Total states = 2 × 4 = 8.
 The states can be denoted as [Vacuum Location, A Status, B Status]. The eight states are:

 [A, Clean, Clean]

 [A, Clean, Dirty]
 [A, Dirty, Clean]
 [A, Dirty, Dirty]
 [B, Clean, Clean]
 [B, Clean, Dirty]
 [B, Dirty, Clean]
 [B, Dirty, Dirty]

Goal States: The states where both squares are clean:

 [A, Clean, Clean]

 [B, Clean, Clean]

Description of the State Space Diagram

The state space is represented as a directed graph (inspired by Figure 3.3):

 Nodes: Eight nodes, each labeled with a state:

o [A, Clean, Clean], [A, Clean, Dirty], [A, Dirty, Clean], [A, Dirty, Dirty]
o [B, Clean, Clean], [B, Clean, Dirty], [B, Dirty, Clean], [B, Dirty, Dirty]
o Nodes can be arranged in two columns:
 Left column: States with vacuum in A ([A, *, *]).
 Right column: States with vacuum in B ([B, *, *]).
 Within each column, states can be ordered by dirt status (e.g., [Clean, Clean], [Clean,
Dirty], [Dirty, Clean], [Dirty, Dirty]).

 Edges: Directed arrows between nodes, labeled with actions (L, R, S), representing
transitions:

 Horizontal Arrows (L and R actions):

 From [A, Clean, Clean] → [B, Clean, Clean] (R).
 From [A, Clean, Dirty] → [B, Clean, Dirty] (R).
 From [A, Dirty, Clean] → [B, Dirty, Clean] (R).
 From [A, Dirty, Dirty] → [B, Dirty, Dirty] (R).
 From [B, Clean, Clean] → [A, Clean, Clean] (L).
 From [B, Clean, Dirty] → [A, Clean, Dirty] (L).
 From [B, Dirty, Clean] → [A, Dirty, Clean] (L).
 From [B, Dirty, Dirty] → [A, Dirty, Dirty] (L).
 

Vertical Arrows (S actions, within same vacuum location):

 From [A, Dirty, Clean] → [A, Clean, Clean] (S).
 From [A, Dirty, Dirty] → [A, Clean, Dirty] (S).
 From [B, Clean, Dirty] → [B, Clean, Clean] (S).
 From [B, Dirty, Dirty] → [B, Dirty, Clean] (S).
 Self-Loops (actions with no effect):
 L in all [A, *, *] states (stays in same state).
 R in all [B, *, *] states.
 S in states where the current square is clean (e.g., S in [A, Clean, Clean], [A, Clean,
Dirty], [B, Clean, Clean], [B, Dirty, Clean]).

6. Define rationality in the context of intelligent agents. Discuss how rationality guides agents
towards achieving their goals.

In artificial intelligence, rationality refers to an agent’s ability to make decisions that maximize its
expected performance measure, given its knowledge and perceptual inputs. A rational agent acts to
achieve the best expected outcome based on what it knows, not necessarily a perfect or correct one.
Rationality is evaluated relative to:

1. Performance measure (what constitutes success),

2. Percept sequence (history of observations),
3. Knowledge the agent has about the environment,
4. Available actions it can perform.

How Rationality Guides Agents Toward Their Goals

Rationality serves as a guiding principle that helps intelligent agents:

 Choose optimal actions: At every decision point, the agent selects the action that
maximizes expected utility.
  Adapt to incomplete or uncertain information: Rational agents operate under bounded
rationality, adjusting their behavior as they acquire more information.
 Focus on goals: The agent’s actions are always aimed at satisfying its predefined goals or
maximizing its performance measure.
 Avoid arbitrary behavior: Decisions are not random but based on reasoning or inference
from the agent’s current state and objectives.
For example, a self-driving car as a rational agent will decide its actions (like turning,
braking, or accelerating) based on maximizing safety and reaching the destination
efficiently, using sensor input and prior knowledge of the roads.

7. Differentiate:

i. Fully observable vs partially observable

ii. Single agent vs Multiagent
iii. Deterministic vs Stochastic
iv. Static vs Dynamic

Criteria Fully Observable Partially Observable

Definition Agent can access the complete Agent has incomplete/noisy access
environment state. to the state.

Sensor Dependency No reliance on memory; all data is Agent may need memory or
available. inference for decisions.

Predictability High, since all information is Lower; agent must make

known. assumptions or estimates.

Example Chess (with visible board). Poker (hidden cards), autonomous

driving.

Complexity for Agent Simpler to design. Requires state estimation, inference

Design mechanisms.

Criteria Single Agent Multiagent

Definition One agent acts in the Multiple agents operate, potentially

environment. interacting.

Interaction Type No interaction; only the May involve competition (games) or

environment matters. cooperation.

Performance Depends only on the agent’s Depends on actions of all agents.

Dependency actions.

Example Path-planning robot in empty Chess, auctions, traffic navigation.

 room.

Design Complexity Easier to model and evaluate. Requires modeling of other agents’
behavior.

Criteria Deterministic Stochastic

Definition Next state is exactly Outcome has elements of

predictable. randomness.

Need for Probability Not required. Essential for handling uncertainty.

Models

Predictability High and consistent. Variable; dependent on probabilistic

outcomes.

Example Calculator operations, maze Robot walking on ice, weather

solving. simulation.

Real-world Relevance Less realistic. More realistic for real-world

modeling.

Criteria Deterministic Stochastic

Definition Next state is exactly predictable. Outcome has elements of

randomness.

Need for Probability Not required. Essential for handling uncertainty.

Models

Predictability High and consistent. Variable; dependent on probabilistic

outcomes.

Example Calculator operations, maze Robot walking on ice, weather

solving. simulation.

Real-world Less realistic. More realistic for real-world

Relevance modeling.
 Criteria Static Dynamic
Definition Environment doesn't change Environment can change during or
during decision-making. before decisions.
Agent Timing No urgency in decision-making. Must act quickly; may need real-time
Requirement response.
Examples Puzzle solving, image Autonomous driving, stock trading
classification. bots.
Sensor Demands Less frequent sensing is Requires continuous/perpetual
acceptable. sensing.

8. Give PEAS specification for Automated Taxi Driver

The PEAS framework is used to characterize the task environment of an intelligent agent by
specifying its Performance measure, Environment, Actuators, and Sensors

Component Description (Explain each in Detail)

P (Performance) - Safe driving
- Fast travel time
- Minimize fuel consumption
- Obey traffic laws
- Maximize passenger comfort and satisfaction
E (Environment) - Roads
- Other vehicles
- Traffic signals
- Pedestrians
- Passengers
- Weather conditions
A (Actuators) - Steering
- Accelerator
- Brake
- Signal indicators
- Horn- Display/interface for passengers
S (Sensors) - Cameras
- Lidar/Radar
- GPS- Speedometer
- Accelerometer
- Microphones (for detecting sirens or passenger requests)

9. Compare and contrast different agent architectures.

1. Simple Reflex Agents

 Structure: Operate based on condition-action rules (i.e., "if condition, then action").
 Strengths: Very fast and easy to implement.
 Weaknesses: Lack memory or ability to learn from the environment; cannot handle
complex or partially observable environments.
2. Model-Based Reflex Agents
  Structure: Maintain an internal model of the world to handle partially observable
environments.
 Strengths: More flexible than simple reflex agents; can infer hidden parts of the
environment.
 Weaknesses: Still reactive and limited in planning capabilities.
3. Goal-Based Agents
 Structure: Use goal information to choose among possible actions.
 Strengths: Can plan ahead and select actions that achieve goals, enabling more complex
behavior.
 Weaknesses: Planning can be computationally expensive; requires clearly defined goals.
4. Utility-Based Agents
 Structure: Extend goal-based agents by evaluating how “good” different states are using
a utility function.
 Strengths: Capable of making trade-offs between conflicting goals and optimizing
performance.
 Weaknesses: Designing utility functions can be complex.
5. Learning Agents
 Structure: Include components for learning from experiences and improving over time.
 Strengths: Adaptability to changing environments and ability to improve performance.
 Weaknesses: Require training data and may be complex to implement effectively.

10. Give PEAS specification of Tomato classification system

Component Description (Explain each in Detail)

P (Performance) - High classification accuracy
- Fast sorting rate
- Minimal damage to tomatoes
- Correct separation by ripeness or defects
E (Environment) - Tomatoes on a conveyor belt
- Varying lighting conditions
- Multiple ripeness levels
- Presence of defective tomatoes
A (Actuators) - Robotic arms
- Mechanical sorters
- Conveyor mechanisms for directing output
S (Sensors) - High-resolution cameras
- Color/vision sensors
- Lighting systems
- Possibly weight or firmness sensors

MODULE - 2
1. Explain five components and well-defined problem. Consider an 8-puzzle problem as an
example.

In the context of artificial intelligence, a well-defined problem is a problem that can be precisely
formulated to enable an agent to solve it systematically through search or other techniques.

The well-defined problem consists of five key components:

 Initial State,
 Actions,
 Transition Model,
 Goal Test, and
 Path Cost.

These components provide a complete description of the problem, allowing an agent to navigate the
state space to find a solution. Below, each component is explained in detail, followed by its
application to the 8-puzzle problem,

1) Initial State:
o Definition: The starting configuration of the environment when the agent begins solving
the problem. It describes the state of the world at the outset, providing the starting
point for the agent’s search.
o Role: The initial state anchors the problem, giving the agent a clear point from which to
apply actions and explore possible states.
2) Actions:
o Definition: The set of possible operations or moves the agent can perform in a given
state. The actions available may depend on the current state, and they define how the
agent can manipulate the environment.
o Role: Actions specify the agent’s capabilities, enabling it to transition from one state to
another in pursuit of the goal.
3) Transition Model:
o Definition: A description of what each action does, specifying the resulting state when
an action is applied to a given state. Formally, for a state s s s and action a a a, the
transition model defines the successor state s′=Result(s,a) s' = \text{Result}(s, a) s
′=Result(s,a).
o Role: The transition model governs the dynamics of the environment, allowing the agent
to predict the outcomes of its actions and build a state space.
4) Goal Test:
o Definition: A condition that determines whether a given state is a goal state, indicating
that the problem has been solved. It checks whether the agent has achieved the desired
outcome.
o Role: The goal test provides the stopping criterion, guiding the agent to recognize when
it has reached a solution.

5) Path Cost:
 o Definition: A numerical cost associated with the sequence of actions (path) taken to
reach a state, reflecting the effort or resources required. Typically, each action has a
cost, and the path cost is the sum of the costs of the actions in the path.
o Role: The path cost enables the agent to evaluate and compare different solution paths,
often aiming to find the one with the lowest cost (optimal solution).

• States: A state description specifies the location of each of the eight tiles and the blank in one of
the nine squares.

• Initial state: Any state can be designated as the initial state. Note that any given goal can be
reached from exactly half of the possible initial states (Exercise 3.4).

• Actions: The simplest formulation defines the actions as movements of the blank space Left, Right,
Up, or Down. Different subsets of these are possible depending on where the blank is.

• Transition model: Given a state and action, this returns the resulting state; for example, if we apply
Left to the start state in Figure 3.4, the resulting state has the 5 and the blank switched.

• Goal test: This checks whether the state matches the goal configuration shown in Figure 3.4.
(Other goal configurations are possible.)

• Path cost: Each step costs 1, so the path cost is the number of steps in the path.

2. Explain the principles of breadth-first search as a problem-solving strategy with example.

Breadth-first search (BFS) explores the shallowest nodes first, level by level, using a FIFO queue. It
selects the shallowest unexpanded node, ensuring the shallowest path to each state. BFS applies the
goal test when generating nodes and avoids revisiting states already in the frontier or explored set.

BFS is complete (guarantees finding a solution if one exists) and optimal when all step costs are
equal. However, it suffers from poor time and space complexity: both are O(b^d), where b is the
branching factor and d the solution depth. It stores all nodes in memory, making space usage
especially problematic.
Principles of Breadth-First Search

1. Queue-Based Exploration :
o BFS uses a First-In-First-Out (FIFO) queue to manage the nodes (states) to be explored.
Nodes are added to the end of the queue (enqueued) and removed from the front
(dequeued), ensuring that nodes are processed in the order they are generated.
o This results in a level-by-level exploration, where all nodes at a given depth (distance
from the initial state) are explored before moving to nodes at the next depth.
2. Systematic Expansion of Nodes :
o BFS starts with the initial state as the root node of the search tree. It expands this
node by applying all possible actions defined in the problem, generating successor
nodes (child nodes) according to the transition model.
o Each successor represents a new state reached by applying an action. The process
continues by expanding nodes in the order they are dequeued, generating their
successors, and adding them to the queue.
3. Goal Test Application :
o For each node dequeued, BFS applies the goal test to check if the node’s state is a goal
state. If a goal state is found, the search terminates, and the solution path (sequence of
actions from the initial state to the goal) is returned.
o The goal test is applied when a node is dequeued (before expansion), ensuring
efficiency by avoiding unnecessary expansions of goal states.
4. Avoiding Redundant States :
o To prevent exploring the same state multiple times (e.g., in cyclic state spaces), BFS
maintains a closed list (or explored set) to track states that have already been visited. If
a generated successor state is in the closed list, it is discarded.
o Additionally, BFS checks if a successor is already in the queue (open list) to avoid
redundant paths, ensuring each state is explored via the shortest path first.
5. Shortest Path Guarantee :
o BFS guarantees finding the shortest path to the goal in terms of the number of actions,
assuming each action has a uniform cost (e.g., cost of 1 per action). This is because it
explores all nodes at depth d d d before any nodes at depth d+1 d+1 d+1, ensuring the
first goal state found is reached via the fewest actions.
o For problems with varying action costs, BFS can be modified (e.g., as uniform-cost
search) to find the least-cost path.
6. Completeness and Optimality :
o Completeness: BFS is complete, meaning it will find a solution if one exists, provided
the state space is finite or the branching factor (number of successors per state) is
finite (Page 82). In infinite state spaces, BFS may fail unless modified to avoid infinite
branches.
o Optimality: BFS is optimal (finds the shortest path) for problems with uniform action
costs, as it explores all shallower nodes before deeper ones (Page 82).
7. Time and Space Complexity :
o Time Complexity: BFS’s time complexity is O(bd) O(b^d) O(bd), where b b b is the
branching factor (average number of successors per state) and d d d is the depth of the
shallowest goal state. This reflects the number of nodes generated in the worst case.
o Space Complexity: BFS’s space complexity is also O(bd) O(b^d) O(bd), as it stores all
generated nodes in the queue and closed list. The queue size grows exponentially with
depth, making BFS memory-intensive for large state spaces.
o The textbook notes that BFS’s high space requirements are a significant limitation,
especially compared to depth-first search (Page 83).
8. Algorithm Description:
Example:

3. Construct Uniform-cost search on a graph algorithm with example

Uniform-Cost Search (UCS) is a search algorithm that always chooses the path with the lowest total
cost so far. It’s like breadth-first search (BFS), but instead of choosing the shallowest node (fewest
steps), it chooses the cheapest one (least cost, regardless of number of steps).

It works well even when actions have different costs.

Algorithm Steps:

1) Initialize:
 Add the start node to a priority queue (frontier) with cost 0.
 Create an explored set to keep track of visited nodes.

2) Loop:
 While the frontier is not empty:
 Remove the node with the lowest cost.
  If it's the goal, return the path and cost.
 Add the node to the explored set.
 For each neighbor of this node:
 Calculate its total path cost.
 If it's not in the frontier or explored set, add it to the frontier.
 If it's already in the frontier but this new path is cheaper, update it with the lower
cost.

Example (Sibiu to Bucharest):

1) Start from Sibiu.

2) UCS adds Rimnicu Vilcea (cost 80) and Fagaras (cost 99).
3) It chooses Rimnicu Vilcea next (cheaper), then adds Pitesti (total cost 177).
4) Then it chooses Fagaras, which adds Bucharest (cost 310).
5) Bucharest is found, but UCS doesn’t stop yet.
6) It checks the path through Pitesti to Bucharest (new cost 278), which is cheaper.
7) The old, more expensive path to Bucharest is removed.
8) The new, cheaper Bucharest is expanded, and the solution is returned.

Example Graph:

A
/ \
B C
/\ \
D E F
/
G

We represent each path like this:

[Node, Total Cost, Path]
Initial state:
 Frontier: [(A, 0, [A])]
1. Expand A:
 Neighbors: B (cost 2), C (cost 1)
 Frontier: [(C, 1, [A, C]), (B, 2, [A, B])]
2. Expand C:
 Neighbor: F (C → F = 4), total = 1 + 4 = 5
 Frontier: [(B, 2, [A, B]), (F, 5, [A, C, F])]
3. Expand B:
 Neighbors: D (2 + 5 = 7), E (2 + 3 = 5)
 Frontier: [(F, 5, [A, C, F]), (E, 5, [A, B, E]), (D, 7, [A, B, D])]
4. Expand F:
 Neighbor: G (F → G = 2), total = 5 + 2 = 7
 Frontier: [(E, 5, [A, B, E]), (D, 7, [A, B, D]), (G, 7, [A, C, F, G])]
5. Expand E:
 No new neighbors
 Frontier: [(D, 7, [A, B, D]), (G, 7, [A, C, F, G])]
6. Expand D:
 No new neighbors
  Frontier: [(G, 7, [A, C, F, G])]
7. Expand G → Goal Reached!
 Cheapest path: A → C → F → G
 Total cost: 7

Final Result:
 Path: A → C → F → G
 Cost: 7
 UCS ensures this is the optimal (cheapest) path.

4. Apply the key principles of iterative deepening depth-first search as an uninformed search
strategy

Iterative Deepening Depth-First Search (IDDFS) is an uninformed search strategy that combines the
advantages of both Depth-First Search (DFS) and Breadth-First Search (BFS) to find the shallowest
goal node without knowing the depth of the solution in advance.

Principle Description
Search Strategy Performs DFS up to a depth limit; increases limit iteratively (0, 1, 2...)
Completeness Yes—guaranteed to find a solution if one exists
Optimality Yes—returns the shallowest solution (like BFS) when cost per step is uniform
Time Complexity O(bᵈ), where b is the branching factor and d is the depth of the shallowest goal
Space Complexity O(bd), same as DFS—much better than BFS

IDDFS is an effective uninformed search strategy

that provides a good balance between time and
memory. It searches layer by layer like BFS, but
uses less memory like DFS. It is complete,
optimal (if costs are uniform), and practical for
large problems where the goal depth is not
known in advance.