(a) Intelligence

Intelligence is the ability to think, learn, and solve problems. It helps us understand new things,
make good decisions, and adapt to different situations.

Example:
A student who figures out a math problem by trying different methods and learning from
mistakes is using intelligence.

(b) Artificial Intelligence (AI)

AI is when computers or machines are made to act like humans by learning from data,
recognizing patterns, and making decisions. These systems can perform tasks that usually require
human thinking.

Example:
A smartphone that suggests words while typing or recognizes a person’s face in photos is using
AI.

(c) Agent

An agent is something that interacts with its surroundings and takes actions based on what it
senses. Agents can be living beings, like humans and animals, or artificial, like robots and
software.

Example:
A home-cleaning robot that moves around, detects dirt and cleans it up is an agent.

(d) Rationality

Rationality means making smart and logical decisions by considering all available information.
A rational decision is one that leads to the best possible result in a given situation.

Example:
If it’s going to rain, a rational person would take an umbrella before going outside.
(e) Logical Reasoning

Logical reasoning is the ability to think step by step to reach a conclusion. It follows clear rules
to figure out the right answer.

Example:
If all apples are fruits, and a banana is also a fruit, then we can logically say that both apples and
bananas belong to the same category—fruits.

QUESTION 2

Artificial Intelligence:

AI is about making computers act smart and solve problems like humans.

Example: A voice assistant like Siri or Alexa that understands and responds to your questions.

Machine Learning:

ML is a type of AI where computers learn from data and improve over time without needing to
be directly programmed for every task.

Example: Netflix recommends movies based on what you have watched before.

Deep Learning:

DL is a special type of ML that uses neural networks to recognize patterns in a large amount of
data.

Example: Self-driving cars use DL to recognize traffic signs, people, and other cars.

Question 3

Generative AI is a type of artificial intelligence that creates new content, such as text, images,
music, and videos, by learning patterns from existing data. Unlike traditional AI, which mainly
analyzes or classifies information, Generative AI produces original content similar to what
humans create. For example, ChatGPT can write human-like text. These AI systems use deep
learning and neural networks to recognize patterns and generate meaningful outputs.

Large Language Models are a special type of Generative AI that focus only on understanding
and generating text. They are trained on large amounts of text data and use advanced machine
learning techniques to process language, predict words, and create natural-sounding text. Some
common examples of LLMs include ChatGPT, Google Bard, and Claude, which are used for
chatbots, content writing, and summarizing information. These models have greatly improved
how computers understand and communicate using human language.

Generative AI and LLMs are closely connected. LLMs are a part of Generative AI, but while
Generative AI can create various types of content, LLMs focus only on text. All LLMs belong to
Generative AI, but not all Generative AI models are LLMs. Generative AI also includes tools
that create images, music, and videos, whereas LLMs are designed for working with written
language.

Question 4

(a) Rationality of the Simple Vacuum-Cleaner Agent Function

The simple vacuum-cleaner agent function in Figure 2.3 follows a straightforward rule-based
approach:
 Percepts: The agent perceives its current location and whether it is clean or dirty.
 Actions:
o If the current square is dirty, the agent performs the Suck action.

o If the current square is clean, the agent moves (either Left or Right).

Evaluating Rationality
A rational agent is one that maximizes its performance measure. According to Section 2.2 of
AIMA, the performance measure in this scenario is:
 One point is awarded for each clean square at every time step.
Under the given assumptions:
1. The environment consists of only two locations, A and B.
2. The agent always knows its current location and whether it is dirty.
3. Clean squares remain clean unless dirt reappears (which is not assumed in this case).
4. The agent can only move Left, Right, or Suck.
Since the agent correctly cleans when needed and moves when appropriate, it maximizes
cleanliness and is therefore rational. However, if all squares are already clean, the agent will
continue moving back and forth wasting steps, which could be suboptimal under different
performance measures.

(b) Rational Agent Function When Movements Cost One Point

If each movement (Left or Right) incurs a -1 point penalty, then the current agent is not
optimal, because unnecessary movements will decrease the score.
Modified Rational Agent Function
To optimize performance while considering movement cost, the agent should:
 Suck if the current square is dirty.
 Do nothing if the current square is clean and it is certain that all squares are already clean.
 Move to the other square only if it knows that square is dirty.
Does this require an internal state?
Yes, The agent needs to remember whether it has already cleaned the other square.
 Without memory, the agent would blindly move, wasting points.
 With memory (internal state), it can track which squares have been cleaned and remain
idle when necessary, avoiding penalties.
Thus, an internal state is required to track the cleanliness status of both squares.

(c) Cases with Dynamic Dirt and Unknown Geography

Case 1: Clean Squares Can Become Dirty
If dirt can reappear, then the agent cannot simply stop after cleaning all squares. Instead, it
should:
 Periodically revisit squares to check for new dirt.
 Possibly learn patterns (e.g., dirt appears more frequently in a specific location).
A learning-based agent would be useful here, as it could optimize cleaning schedules instead of
randomly checking.
Case 2: Unknown Geography
If the agent does not know the size or layout of the environment:
  It must explore to discover new locations.
 It should build an internal map of the environment.
This makes learning essential, as a purely reactive agent (without memory) would not be able to
navigate efficiently or remember where dirt is likely to appear.

Question 5

1. Rationality
Rationality refers to an agent’s ability to make decisions that maximize its performance based on
the available information and predefined performance measures. A rational agent always takes
the best possible action to achieve its goal, given what it knows.
Example:
 A vacuum-cleaner agent that sucks when it detects dirt and moves only when necessary is
rational because it maximizes cleanliness while minimizing unnecessary movement.
 A chess-playing AI that chooses the best move based on its evaluation function is rational
in that context.

2. Autonomy
Autonomy refers to the agent’s ability to operate without human intervention by relying on its
own observations and learning from the environment. A fully autonomous agent does not need
external instructions and can adapt over time.
Example:
 A self-driving car that detects obstacles, follows traffic rules, and navigates to its
destination without human input is highly autonomous.
 A robotic vacuum cleaner that adjusts its cleaning pattern based on the room layout and
avoids obstacles has some level of autonomy.
3. Reflex Agent
A reflex agent acts only based on current percepts (immediate sensor data) without considering
history or future consequences. It follows predefined rules (if-then conditions).
Example:
 A thermostat that turns on the heater if the temperature drops below 20°C is a reflex
agent, it reacts instantly based on the current temperature.
  A simple vacuum cleaner that sucks when it sees dirt and moves randomly otherwise is
also a reflex agent.
Limitation: Reflex agents struggle in complex environments where memory or planning is
required.
4. Model-Based Agent
A model-based agent maintains an internal model of the environment to keep track of past states
and predict future outcomes. This helps it handle partially observable environments.
Example:
 A self-driving car keeps track of nearby vehicles, road conditions, and traffic signals to
make informed decisions.
 A robot exploring a maze remembers previously visited locations to avoid backtracking.
Advantage: More efficient decision-making in complex environments.

5. Goal-Based Agent
A goal-based agent makes decisions by considering future consequences and choosing actions
that bring it closer to a specific goal. Instead of just reacting, it evaluates different possibilities.
Example:
 A GPS navigation system finds the best route to a destination instead of just reacting to
traffic.
 A warehouse robot selects the shortest path to pick up an item and deliver it efficiently.
Advantage: Can handle more complex tasks that require planning.

6. Utility-Based Agent
A utility-based agent extends goal-based agents by assigning numerical values (utilities) to
different outcomes, choosing the action that maximizes its expected benefit.
Example:
 A stock trading AI selects investments based on expected returns and risk factors.
 A self-driving taxi optimizes fuel efficiency, speed, and passenger comfort rather than
just reaching the destination.
Advantage: More flexible decision-making, especially when there are multiple competing
objectives.
7. Learning Agent
A learning agent improves its performance over time by learning from experiences and adjusting
its behavior accordingly. It consists of:
1. Performance element – Executes actions.
2. Learning element – Adapts based on feedback.
3. Critic – Evaluates performance.
4. Problem generator – Explores new actions.
Example:
 ChatGPT improves responses by learning from user interactions.
 AlphaGo (AI playing Go) improves by playing thousands of matches and adjusting its
strategies.
Advantage: Can adapt to new environments without being explicitly programmed for every
scenario.

Question 6
Example: Intelligent Medical Diagnosis System
A Medical Diagnosis Expert System (MDES) assists doctors by suggesting possible diseases
based on symptoms and patient data.

Phase 1: Problem Assessment

 Objective: Build an AI system that helps doctors diagnose diseases based on symptoms,
test results, and medical history.
 Challenges:
o Need a large dataset of patient records.

o Diagnosing diseases requires expert medical knowledge.

o Must ensure high accuracy and reliability.

  Feasibility:
o AI techniques like Machine Learning (ML) and Rule-Based Expert Systems are
well-suited for medical diagnosis.
o Collaboration with medical professionals is required to validate diagnoses.

Example:
A hospital wants an AI system to help diagnose respiratory diseases based on symptoms like
cough, fever, and X-ray images.

Phase 2: Data and Knowledge Acquisition

 Gather medical knowledge:
o Work with doctors to understand symptom-disease relationships.

o Review medical textbooks, research papers, and clinical guidelines.

 Collect patient data:

o Historical patient records, lab test results, and imaging data.

o Datasets from hospitals or public sources (e.g., WHO, CDC).

 Types of Knowledge Acquired:

o Rule-based: "If fever + cough + difficulty breathing → Possible pneumonia."

o Case-based: Previous patient cases with correct diagnoses.

Example:
 Collect 10,000 patient records from hospitals, including symptoms, diagnoses, and lab
results.

Phase 3: Development of a Prototype System

 Build a basic version of the system with limited features.
 Use Machine Learning models (e.g., Decision Trees, Neural Networks) to predict
diseases.
 Implement a simple rule-based system for initial testing.
Example:
  A prototype system takes input (e.g., "fever, cough, fatigue") and suggests possible
diseases like flu or pneumonia.
 Doctors test the system and provide feedback.

Phase 4: Development of a Complete System

 Expand the knowledge base to include more diseases.
 Improve accuracy by training on a larger dataset.
 Add a user-friendly interface for doctors and nurses.
 Implement Natural Language Processing (NLP) to allow input in natural language.
 Ensure security and compliance (e.g., HIPAA for patient data).
Example:
 The final system can diagnose 50+ diseases and integrates with hospital databases to
access patient history.
 Doctors can upload X-ray images for AI-based analysis.
Phase 5: Evaluation and Revision of the System
 Test system accuracy on new patient cases.
 Compare AI predictions with real doctor diagnoses.
 Get feedback from doctors to improve system recommendations.
 Fix errors and biases in the model.
Example:
 AI diagnoses 95% of flu cases correctly but struggles with rare diseases.
 Developers adjust the algorithm to improve detection of rare conditions.
Phase 6: Integration and Maintenance of the System
 Deploy the system in hospitals and clinics.
 Provide regular updates with new medical research.
 Train hospital staff on how to use the system.
 Ensure long-term support and bug fixes.
Example:
The system is used in 100+ hospitals, with monthly updates based on new medical studies.