PART B ▪ UNIT 1 · AI Reflection, Project Cycle & Ethics

  root@vm-learning
  ~
  $
  open
  ch-b1
  

PART B ▪ UNIT 1

AI Reflection, Project Cycle & Ethics

Welcome to the World of Artificial Intelligence

Artificial Intelligence (AI) is a branch of computer science that deals with creating machines (computers, robots, software) that can think, learn, and take decisions like humans.

The term "Artificial Intelligence" was first used by John McCarthy in 1956, who is known as the "Father of AI". Today AI is present everywhere – from mobile phones and cars to hospitals and homes.

SUB-UNIT 1: AI REFLECTION

Learning Outcome 1: Identify and appreciate AI and describe its applications in daily life

1.1 Introduction to Artificial Intelligence

Artificial Intelligence (AI) is the ability of a machine to perform tasks that usually require human intelligence, such as understanding language, recognizing images, making decisions, solving problems, and learning from experience.

🔹 What Makes a Machine "Intelligent"?

A machine is called "intelligent" when it can do any one or more of the following:

Think and reason like humans.
Learn from data and experience.
Understand natural human language.
Recognize images, faces, and sounds.
Take decisions based on information.
Solve problems on its own.
Improve its performance over time.

🔹 What is NOT AI?

Not every machine or device is AI. A machine is considered AI only if it can learn, adapt, and take decisions. Examples of things that are not AI:

A calculator (just follows fixed rules).
A washing machine with preset cycles.
A simple alarm clock.
A remote control that only sends signals.

1.2 Applications of AI in Daily Life

AI is present all around us. Some common examples of AI in our daily life:

Area	AI Applications
Smartphones	Voice assistants (Siri, Google Assistant, Alexa), face unlock, camera filters, auto-correct.
Social Media	Friend suggestions, face tagging, personalized feeds, content recommendations.
Entertainment	Netflix / YouTube / Spotify recommendations based on what you watch and listen.
Shopping	Product recommendations on Amazon, Flipkart; chatbots for customer support.
Banking	Fraud detection, online banking chatbots, credit scoring.
Healthcare	Disease diagnosis, medical image analysis, robotic surgery.
Transport	Self-driving cars, Google Maps (traffic), Ola/Uber matching.
Education	Personalized learning apps, AI tutors, language translation.
Home	Smart TVs, smart lights, robotic vacuum cleaners.
Agriculture	Crop monitoring, smart irrigation, yield prediction.

Learning Outcome 2: Recognize the three realms (domains) of AI

1.3 The Three Domains of AI

AI has three main domains / realms based on the type of data it works with:

📊

1. Data Science

Works with numbers and statistics. Uses data to find patterns and make predictions.

👁️

2. Computer Vision (CV)

Works with images and videos. Helps computers "see" and understand visual data.

💬

3. Natural Language Processing (NLP)

Works with human language – text and speech. Helps machines understand and respond in our language.

📊 1. Data Science (Data Statistics)

Data Science is the domain of AI that deals with analyzing numerical data (numbers, statistics) to find patterns, relationships, and useful insights for decision-making.

🔹 Applications of Data Science

Weather forecasting
Cricket score prediction
Business sales analysis
Stock market predictions
Customer behaviour analysis
Disease outbreak tracking

Game: Rock, Paper, Scissors (rockpaperscissors.ai) – The AI studies the pattern of your moves (data) and predicts your next move.

👁️ 2. Computer Vision (CV)

Computer Vision is the domain of AI that enables computers to see, recognize, and understand visual information from images and videos, just like humans.

🔹 Applications of Computer Vision

Face recognition (phone unlock, airport security)
Self-driving cars (detecting pedestrians, signals)
Medical imaging (X-ray, MRI analysis)
Google Lens
Optical Character Recognition (OCR)
Quality check in factories
Security cameras detecting intruders

Game: Quick Draw (quickdraw.withgoogle.com) – You draw a sketch and the AI tries to guess what it is. It uses computer vision to analyse your drawing.

💬 3. Natural Language Processing (NLP)

Natural Language Processing is the domain of AI that enables computers to understand, interpret, and respond to human language in both text and voice form.

🔹 Applications of NLP

Voice assistants (Siri, Alexa, Google Assistant)
Google Translate / language translation
Chatbots for customer support
Spam filters in email
Grammar and spelling check (Grammarly)
Auto-caption in videos
Sentiment analysis on social media

Game: Semantris (research.google.com/semantris) – You type a word related to the shown word and the AI understands the connection between them using NLP.

SUB-UNIT 2: AI PROJECT CYCLE

Learning Outcome: Identify the AI Project Cycle framework

2.1 Introduction to AI Project Cycle

The AI Project Cycle is a step-by-step framework that guides us through the process of building an AI-based solution to a real-world problem.

The AI Project Cycle has 6 stages:

🔄 THE 6 STAGES OF AI PROJECT CYCLE

1Problem Scoping

2Data Acquisition

3Data Exploration

4Modeling

5Evaluation

6Deployment

2.2 Stage 1: Problem Scoping

Problem Scoping is the first step of the AI Project Cycle. It means clearly defining and understanding the problem we want to solve using AI.

🔹 Why is Problem Scoping Important?

Helps understand the exact problem.
Identifies who is affected by the problem.
Sets clear goals for the AI project.
Prevents waste of time and resources.
Makes sure the solution actually helps people.

📋 The 4Ws Problem Canvas

To scope a problem properly, we use the 4Ws Canvas – a simple template of four questions:

👥 WHO

Who is facing the problem?
Identify the stakeholders – the people affected by or involved in the problem.
Example: Farmers, students, elderly people.

❓ WHAT

What is the problem?
Describe the problem clearly and find evidence to show it really exists.
Example: Crops damaged by unexpected rain.

📍 WHERE

Where does the problem occur?
Identify the context / location / situation in which the problem happens.
Example: In village farms during monsoon.

💡 WHY

Why is it a problem? Why does it need to be solved?
Explain the importance of solving it and the benefits of solution.
Example: Financial loss and food shortage.

📄 Problem Statement Template

After filling the 4Ws canvas, write a clear problem statement in this format:

Our [Stakeholder – WHO] has a problem of [Issue – WHAT] while / when [Context – WHERE]. An ideal solution would be [Why solve it – WHY].

🎯 Goal Setting & Stakeholders

Goal: A clear target that we want to achieve by solving the problem.
Stakeholders: People who are directly or indirectly affected by the problem and the solution.

⚖️ Ethical Issues in Problem Scoping

While choosing a problem, we must think about ethical questions:

Will the solution harm any person or group?
Is data privacy being respected?
Is the solution fair to everyone?
Will it cause bias or discrimination?

2.3 Stage 2: Data Acquisition

Data Acquisition is the process of gathering relevant data from reliable sources that will be used to train the AI model.

🔹 What is Data?

Data means facts, figures, or information that is collected for reference or analysis. It can be in the form of numbers, text, images, audio, or video.

🔹 Sources of Data

Surveys and questionnaires
Interviews and observations
Sensors (temperature, motion, heart rate)
Cameras (images, video)
Websites / APIs / Databases
Government open-data portals
Social media platforms
News articles and research papers

🔹 Data Features

Data features are the individual pieces of information (attributes / parameters) that affect the problem and are used to train the AI model.

For predicting whether a student will pass an exam, data features could be: hours of study, attendance, previous marks, participation in class, amount of sleep.

🔹 System Maps

A System Map is a diagram that shows the relationship between different data features and how they affect the problem.

System maps use arrows (+ or –) to show if one feature increases or decreases another feature.

🔹 Important Questions for Data Acquisition

What data features are needed?
Where can I get the data?
How often do I have to collect the data?
Is the data reliable and valid?
What happens if I don't have enough data?
How does the analysis inform the action?

2.4 Stage 3: Data Exploration / Data Visualization

Data Exploration means analyzing and visualizing the collected data using charts and graphs to understand the patterns, trends, and insights hidden in the data.

🔹 Why Visualize Data?

Large amounts of data are hard to understand in number form.
Visualization makes data simple and clear.
Helps in finding patterns and trends quickly.
Makes presentation easy for everyone.
Helps in decision making.

🔹 Common Types of Graphs / Charts

📊Bar GraphCompare quantities

📈Line GraphShow trends over time

🥧Pie ChartShow parts of a whole

🔘Scatter PlotRelation between 2 variables

📉HistogramDistribution of data

📋TableOrganised rows & columns

🗺️Heat MapColor-coded intensity

📦Box PlotData spread & outliers

Reference: datavizcatalogue.com is a great website to learn about different types of graphs and when to use them.

2.5 Stage 4: Modeling

Modeling is the stage where we create an AI model that takes data as input, processes it using certain logic or learning methods, and gives an output / prediction.

🔹 Types of Modeling

📏 1. Rule-Based Model

The developer gives the machine fixed rules and relationships. The machine just follows those rules. The data and rules are fed in by humans and the machine does not learn on its own.

Example: A traffic light that follows a fixed time rule. Or, "If temperature > 35°C, turn ON AC."

🧠 2. Learning-Based Model

The machine learns from data on its own without being given fixed rules. The more data it gets, the smarter it becomes.

Example: Facebook's face recognition – it learns from the photos you tag. Or, email spam filter that learns from user's marking.

🔹 Difference between Rule-Based and Learning-Based

Rule-Based	Learning-Based
Rules written by humans	Machine learns rules from data
Does not improve with time	Improves with more data
Fixed and rigid	Flexible and adaptive
Simple to make	Complex to design
Needs little data	Needs large amounts of data

🔹 Sub-types of Learning-Based Models

Supervised Learning: Trained using labelled data (input + correct output). E.g., Spam filter.
Unsupervised Learning: Trained using unlabelled data; finds patterns itself. E.g., Customer grouping.
Reinforcement Learning: Learns by trial and error with rewards / penalties. E.g., Game-playing AI.

2.6 Stage 5: Evaluation

Evaluation is the stage where we test the performance of the AI model to check how accurate and reliable it is before using it in real life.

🔹 Why is Evaluation Important?

To check if the model works correctly.
To measure how many times it gives right or wrong answers.
To improve the model before deployment.
To compare with other models.

🔹 Key Evaluation Terms (Confusion Matrix)

During evaluation, an AI model's predictions fall into four categories:

✅ True Positive (TP)Prediction: YES & Reality: YES
Correct prediction of positive case.

❌ False Positive (FP)Prediction: YES but Reality: NO
Wrong prediction – "false alarm."

❌ False Negative (FN)Prediction: NO but Reality: YES
Wrong prediction – missed case.

✅ True Negative (TN)Prediction: NO & Reality: NO
Correct prediction of negative case.

Fire Alarm Example:
True Positive: There IS a fire and the alarm rings. ✅
False Positive: There is NO fire but alarm rings. ❌
False Negative: There IS a fire but alarm doesn't ring. ❌ (dangerous!)
True Negative: There is NO fire and alarm does NOT ring. ✅

2.7 Stage 6: Deployment

Deployment is the final stage where the AI model is released into the real world for actual use by the intended users.

🔹 Key Points About Deployment

The model moves from testing → real usage.
It should be user-friendly and work on devices users have.
Regular monitoring is needed – to check if it still works well.
Model needs updates when new data / conditions come.
Feedback from users helps improve the model further.

Case Study – Preventable Blindness: Google developed an AI model that scans eye images to detect diabetic retinopathy (a cause of blindness). After evaluation, the model was deployed in hospitals in India to help doctors screen patients early.

SUB-UNIT 3: AI ETHICS

Learning Outcome: Understand ethical issues, AI bias and AI access

3.1 What is AI Ethics?

AI Ethics refers to the moral principles and guidelines that must be followed while designing, developing, and using AI systems to make sure they are fair, safe, and do no harm.

🔹 Why AI Ethics is Important

AI affects the lives of millions of people.
Wrong decisions by AI can cause serious harm.
To ensure fairness for everyone.
To protect privacy and personal data.
To prevent misuse of AI (like deepfakes, fraud).
To maintain trust in AI technology.

🔹 Principles of AI Ethics

Fairness: AI should treat all people equally.
Transparency: Users should know how AI makes decisions.
Privacy: Personal data must be protected.
Safety: AI must not cause physical or mental harm.
Accountability: Someone must be responsible for AI actions.
Non-discrimination: AI should not be biased based on gender, religion, race, etc.

Explore Moral Machine (moralmachine.net) – A platform by MIT where users play the role of a self-driving car's AI and decide what is ethical in tough situations.

3.2 AI Bias

AI Bias happens when an AI system gives unfair or prejudiced results to certain groups of people because of problems in the data or the way the model was trained.

🔹 Causes of AI Bias

Biased data: If data represents only one group of people (e.g., only men, only one race).
Limited data: Not enough data about all types of users.
Bias of developers: Developers may have their own unconscious prejudices.
Wrong data labels: Incorrect or subjective labelling of training data.
Historical bias: Old data reflects past unfair practices.

🔹 Examples of AI Bias

Face recognition working poorly on dark-skinned faces.
Voice assistants not understanding certain accents.
Resume-screening AI favouring male candidates over female.
Loan approval AI rejecting people from certain areas.

🔹 How to Reduce AI Bias

Use diverse and balanced data.
Include people from different backgrounds in AI development team.
Test AI on different groups of users.
Make AI systems transparent.
Regularly check and audit AI decisions.

3.3 AI Access

AI Access means the equal availability of AI technology, tools, and knowledge to all people regardless of their income, location, or background.

🔹 The Problem of Unequal Access

People in rich countries / cities have better access to AI than people in rural / poor regions.
Lack of internet, electricity, and devices limits AI use.
Lack of AI education in some communities.
Poor understanding of local languages in AI tools.

🔹 How to Improve AI Access

Make AI tools free or affordable.
Provide AI education in all schools (like CBSE doing with this course!).
Develop AI in local languages.
Improve internet connectivity in rural areas.
Give training and digital literacy programmes.

3.4 Advantages and Disadvantages of AI

✅ Advantages of AI

Works 24×7 without tiredness
Very fast and accurate
Handles huge amounts of data
Reduces human errors
Can do dangerous or repetitive tasks
Helps in healthcare (disease diagnosis)
Improves daily convenience
Smarter decisions using data
Personalized services (recommendations)
Supports disabled people (voice assistants)

❌ Disadvantages of AI

Very expensive to develop
Can replace human jobs (unemployment)
Lacks creativity and emotions
Depends on data – bad data = bad output
Can be biased or unfair
Threat to privacy
Risk of misuse (fake news, deepfakes)
Makes people lazy / dependent
Cannot think morally like humans
Security risks (hacking, cyber-attacks)

Balloon Debate: Students divide into teams of 3 on a theme. One team argues in favour of AI (for its benefits) while another argues against (for its harms). This helps analyse both sides of AI.

Quick Revision – Key Points to Remember

AI = Machines that can think, learn, and decide like humans.
Father of AI: John McCarthy (coined the term in 1956).
3 Domains of AI: Data Science (numbers), Computer Vision (images), NLP (language).
AI Project Cycle (6 stages): Problem Scoping → Data Acquisition → Data Exploration → Modeling → Evaluation → Deployment.
4Ws Canvas: Who, What, Where, Why.
Data Features = attributes affecting the problem. System Map = diagram showing relationship between features.
Graphs: Bar, Line, Pie, Scatter, Histogram, Heat Map, Box Plot, Table.
Types of Models: Rule-Based (rules by humans) and Learning-Based (learns from data).
Learning types: Supervised, Unsupervised, Reinforcement.
Evaluation Matrix: True Positive (TP), False Positive (FP), False Negative (FN), True Negative (TN).
Deployment = releasing AI model for real-world use.
AI Ethics Principles: Fairness, Transparency, Privacy, Safety, Accountability, Non-discrimination.
AI Bias = unfair results due to biased data / developers.
AI Access = equal availability of AI to all.

🧠Practice Quiz — test yourself on this chapter→