Unit 1: Revisiting AI Project Cycle & Ethical Frameworks

Artificial Intelligence doesn't just "happen." To build a successful AI project, developers follow a structured path called the AI Project Cycle. Think of it as a roadmap or a recipe that ensures you don't miss any important steps while trying to solve a problem using technology.

1.1 AI Project Cycle: Overview of the Six Stages

The AI Project Cycle is a cyclical process followed to complete an AI project. Just as our minds subconsciously make plans for daily tasks—like deciding how to make a birthday card—this cycle provides a framework to reach a specific goal.

Six Stages of the AI Project Cycle

1. Problem Scoping
2. Data Acquisition
3. Data Exploration
4. Modelling
5. Evaluation
6. Deployment

Stage 1: Problem Scoping

This is the starting point where you set the goal for your project.

• What you do: State the problem you want to solve and look at the parameters that affect it to make the picture clearer.
• Key Tool: Developers use the 4Ws Problem Canvas — Who, What, Where and Why.
• Handbook Example: Stating that you want to help researchers identify penguin species based on their physical features.

Stage 2: Data Acquisition

Data is the "food" that makes an AI intelligent.

• What you do: Collect data that will become the base of your project.
• Important Rule: Data must be collected from reliable and authentic sources.
• Handbook Example: Collecting a dataset containing information on the flipper length, beak size and body mass of penguins.

Stage 3: Data Exploration

Raw data is often messy and difficult to understand simply by looking at a spreadsheet.

• What you do: Use visual representations such as graphs, flow charts, maps or databases to interpret patterns in the data.
• Goal: To identify trends and relationships within the acquired data.

Stage 4: Modelling

This is where the actual AI model is developed.

• What you do: Decide on the type of model needed, such as a Decision Tree or Neural Network, and develop the algorithm around it.
• Action: Research and test different models to find the one that gives the most efficient and suitable output.

Stage 5: Evaluation

Just like a student takes a test, an AI model must also be evaluated.

• What you do: Test the completed model on newly fetched or unseen data that it has not seen before.
• Result: Evaluation helps check the performance and accuracy of the model and identify areas for improvement.

Stage 6: Deployment

Deployment is the final stage where the AI solution is integrated into a real-world environment.

• What you do: Integrate and operate the AI solution in a real-world environment.
• Impact: It ensures that the AI solution actually delivers value and helps stakeholders solve the problem identified during Problem Scoping.

Why is the AI Project Cycle Important?

The AI Project Cycle provides a structured roadmap for developing Artificial Intelligence solutions.

• Efficiency: It helps teams build better AI solutions faster with less effort because every stage follows a planned sequence.
• Modularity: Large and complex problems are divided into smaller and manageable tasks. If an issue occurs, developers can review the affected stage instead of rebuilding the complete project.
• Clarity: Every member of the development team clearly understands the project objective and works towards the same goal.

Structuring an AI Problem Statement: 4Ws Problem Canvas

Before developing an AI solution, the problem should be clearly understood using the 4Ws Problem Canvas.

Question	Purpose
Who?	Identify the stakeholders who are facing the problem.
What?	Define the actual problem using evidence such as reports, surveys or news articles.
Where?	Identify the place or situation where the problem exists.
Why?	Explain how solving the problem will benefit stakeholders and society.

Problem Statement Template

Real-Life Example: AI for Pest Management in Cotton Crops

The AI Project Cycle can be understood through the example of Pink Bollworm infestation, which affects nearly 6 million cotton farmers in India.

AI Project Cycle Stage	Real-Life Mapping
Problem Scoping	Pink Bollworms are difficult to identify with the naked eye and damage cotton crops. The objective is to build an AI application that helps farmers decide when and how much pesticide should be sprayed.
Data Acquisition	Collect images of Pink Bollworms, farmer names, village locations, farm size and previous pesticide usage.
Data Exploration	Clean the collected data by resizing images, removing unnecessary information and identifying relationships between pesticide quantity and infestation.
Modelling	Develop a Computer Vision model capable of recognising Pink Bollworms from crop images.
Evaluation	Test the AI model using new crop images. Developers compare different models using evaluation metrics such as ROC Curves and choose the most accurate one.
Deployment	Deploy the final AI model as a mobile application that allows farmers to scan crops and receive immediate pest management recommendations.

Example 2: Monitoring Student Drowsiness in the Classroom

This project helps teachers identify students who are becoming drowsy during long classroom sessions so that learning effectiveness can be improved.

AI Project Cycle Stage	Real-Life Mapping
Problem Scoping	Stakeholders: Teachers and Students. Problem: Students sometimes fall asleep or lose concentration during classes. Goal: Develop an AI system that detects drowsiness and alerts the teacher.
Data Acquisition	Collect images and videos of students in different conditions: Alert Blinking Drowsy (half-closed eyes) Sleeping (eyes fully closed)
Data Exploration	Clean and standardize images by maintaining similar lighting conditions and image size. Study patterns such as eye closure duration and head movement.
Modelling	Use Computer Vision algorithms capable of facial landmark detection to monitor eye movement and head position.
Evaluation	Test the AI model using new classroom videos. Developers check: True Positives: Correctly detecting a sleeping student. False Positives: Incorrectly identifying an attentive student as sleeping.
Deployment	Integrate the AI model with classroom cameras and the teacher's smart device so that a vibration or notification is generated whenever a student is detected as drowsy.

Example 3: Reducing Food Wastage in the School Canteen

This project predicts the quantity of food required each day so that unnecessary food wastage can be reduced.

AI Project Cycle Stage	Real-Life Mapping
Problem Scoping	Stakeholders: School administration and canteen staff. Problem: Large quantities of food are wasted daily. Goal: Predict daily food demand to prepare only the required quantity.
Data Acquisition	Collect statistical data such as: Student attendance Daily menu Special school events Weather conditions Daily food waste in kg
Data Exploration	Use graphs to identify trends such as: Days with maximum wastage Dishes liked most by students Relationship between menu type and food waste
Modelling	Develop a Statistical Data (Data Science) model that predicts the amount of food required based on historical data.
Evaluation	Compare AI predictions with the actual quantity of food consumed. Fine-tune the model if the prediction error is high.
Deployment	Create a No-Code Dashboard using Tableau or MS Excel that displays the recommended quantity of food to be prepared every morning.

Case Study: The Canteen Queue AI

Competency-Based Questions

1. Multiple Choice Question

Q: What is the primary purpose of defining the problem statement during the Problem Scoping stage?

1. To collect as much data as possible.
2. To understand the aim and objective of the project.
3. To select the final programming language.
4. To train the neural network nodes.

Answer: b) To understand the aim and objective of the project.

2. Assertion and Reasoning

Assertion (A): Data Exploration involves using visual representations like graphs and maps.

Reason (R): Visualization makes it easier for humans to interpret patterns and trends in large quantities of acquired data.

1. Both A and R are true and R is the correct explanation of A.
2. Both A and R are true but R is not the correct explanation of A.
3. A is true but R is false.
4. A is false but R is true.

Answer: a)

3. Application Identification

Scenario: A developer is using the "4W Problem Canvas" to identify stakeholders and the context of a problem.

Q: Name the stage of the AI Project Cycle the developer is currently in.

Answer: Problem Scoping

---

1.2 Introduction to AI Domains

Artificial Intelligence becomes "intelligent" based on the training it receives through datasets. Depending on the type of data—numbers, images or text—AI is broadly classified into three domains.

1. Statistical Data (formerly Data Science)

This domain relates to data systems and processes where the system collects large amounts of data, maintains datasets and derives meaning or "sense" out of them to make decisions.

• Handbook Example: Price Comparison Websites such as PriceGrabber or Shopzilla use data from multiple vendors to help users find the best deals.
• Teacher/School Example: A school management system analyses the last five years of student attendance and marks to predict which students might need extra help before final examinations.
• CS/App Development Example: Targeted advertising uses algorithms that analyse a user's past buying behaviour to show relevant advertisements.

2. Computer Vision (CV)

Computer Vision is the ability of a machine to "see" and analyse visual information such as photographs, videos or thermal images and make predictions or decisions. The goal is to teach machines to extract information from pixels.

• Agricultural Monitoring: Drones with cameras capture aerial images of farms to assess crop health, detect pests and estimate yields.
• Surveillance Systems: Cameras in public spaces can track individuals or vehicles and provide real-time alerts.
• Smart Attendance: A school camera can use facial recognition to mark students present automatically.
• Face Filters: Apps use Computer Vision to identify facial features and overlay digital masks or effects.

3. Natural Language Processing (NLP)

Natural Language Processing focuses on the interaction between computers and humans using natural language—the way humans speak and write. It extracts information from spoken or written words using algorithms.

• Email Filters: Spam filters identify words or phrases in an email to decide whether it belongs in the Inbox or Spam folder.
• Machine Translation: Tools such as Google Translate analyse sentence structures and translate them into another language.
• Automated Grading: An AI system can read student essays and analyse grammar, tone and whether the student stayed on topic.
• Voice Assistants: Voice assistants understand a user's intent and execute tasks such as setting an alarm.

Comparison of AI Domains

AI Domain	Type of Data	Main Purpose	Example
Statistical Data	Numerical / categorical data	Finding patterns and making predictions from data	Student performance prediction
Computer Vision	Images / Videos	Understanding visual information	Facial recognition
Natural Language Processing	Text / Speech	Understanding human language	Chatbots and voice assistants

Case Study: The Smart Library

A school library installs a system where students drop books into a bin. A camera identifies the book cover, a computer reads the title and summary to categorise it as Science or Fiction, and the system records the date and book ID in a database to track library usage trends.

Task	AI Domain	Reason
Identifying the book cover	Computer Vision	It analyses visual information from the book cover.
Reading title and summary	Natural Language Processing	It processes and understands textual information.
Tracking library usage trends	Statistical Data	It analyses stored data and usage patterns.

Competency-Based Questions

1. Multiple Choice Question

Q: Aditi developed a chatbot that clarifies the doubts of Economics students by replying instantly to their typed questions. Identify the AI domain.

1. Computer Vision
2. Statistical Data
3. Natural Language Processing
4. Data Science

Answer: c) Natural Language Processing

2. Assertion and Reasoning

Assertion (A): A self-driving car failing to detect a pedestrian at night indicates a failure in its Computer Vision system.

Reason (R): Computer Vision is the domain that enables AI to see, observe and make sense of visual data such as pixels.

1. Both A and R are correct and R is the correct explanation of A.
2. Both A and R are correct but R is not the correct explanation of A.
3. A is correct but R is incorrect.
4. A is incorrect but R is correct.

Answer: a)

3. Application Identification

Q: Which domain of AI would a developer use to build an app that predicts the future price of a used car based on its fuel type, kilometres driven and previous owners?

Answer: Statistical Data, as it involves analysing numerical features and historical trends.

---

1.3 Ethical Frameworks for AI

Ethical frameworks serve as the moral compass for developing and deploying Artificial Intelligence. They help ensure that powerful AI tools benefit society without causing unintended harm.

What are Frameworks?

A framework is a structured set of steps or a guide designed to help solve problems in an organised manner. Frameworks provide a common language for collaboration and ensure that relevant factors are considered during problem-solving.

What are Ethical Frameworks?

An Ethical Framework is a guide specifically designed to navigate complex moral dilemmas. It ensures that choices made by individuals or organisations align with their values and do not cause unintended harm to stakeholders.

Need for Ethical Frameworks for AI

As AI is increasingly used as a tool for decision-making and influencing, it is critical to ensure that its recommendations are morally acceptable.

• Preventing Bias: Frameworks help identify and eliminate biases, such as those found in hiring algorithms that might unfairly disadvantage specific groups.
• Avoiding Unintended Outcomes: Applying ethical frameworks during development can help prevent harmful consequences before they occur.

Factors Influencing Our Decision-Making

Our decisions—and the frameworks we build—are often influenced by factors that we may not even realise, including:

• Culture and Religion: Whether a decision aligns with personal or societal beliefs.
• Values and Intuition: The internal sense of whether a choice feels right.
• Identity and Location: We may unconsciously favour recipients based on proximity or identity.
• Value of Humans vs. Non-humans: How we prioritise different forms of life.

Types of Ethical Frameworks

Ethical frameworks for AI are broadly classified into two main categories:

Category	Meaning	Examples / Types
Sector-based Frameworks	Frameworks tailored to the unique needs of specific industries.	Finance, Education, Transportation, Agriculture and Bioethics.
Value-based Frameworks	Frameworks based on fundamental moral philosophies guiding human behaviour.	Rights-based, Utility-based and Virtue-based.

A. Sector-Based Frameworks

Sector-based frameworks are tailored to the unique needs of specific industries.

• Finance
• Education
• Transportation
• Agriculture
• Bioethics — a prominent sector-based framework focusing on healthcare and life sciences.

B. Value-Based Frameworks

Value-based frameworks focus on fundamental moral philosophies guiding human behaviour. They are divided into three types.

Framework	Focus
Rights-based	Prioritises human rights and dignity, values human life and respects individual autonomy.
Utility-based	Focuses on maximising the overall good by seeking outcomes that provide the greatest benefit to the largest number of people while minimising harm.
Virtue-based	Centres on the character and intentions of the decision-maker and asks whether actions align with virtues such as honesty, compassion and integrity.

Bioethics and its Principles

Bioethics is the study of ethical issues arising from biology and medicine. When AI is applied to healthcare, four core principles help ensure that technology remains ethical.

Principle	Meaning
Respect for Autonomy	Users, including doctors and patients, should be fully aware of how AI functions and how it makes decisions.
Do No Harm (Non-maleficence)	Harm to any being must be avoided. If a difficult choice must be made, the path of least harm should be chosen.
Ensure Maximum Benefit (Beneficence)	Actions should not merely avoid harm but should actively strive to produce the best possible outcomes for all involved.
Justice	Benefits and burdens of an AI solution must be distributed fairly across people regardless of their background or region.

Use Case: AI in Hospital Resource Allocation

The handbook provides a real-world scenario that demonstrates why ethical frameworks are important when AI is used in healthcare.

The Scenario

A company created an AI algorithm to identify high-risk patients so that hospitals could allocate resources more effectively.

The Problem

The algorithm was found to be biased. Patients from a specific Western region were categorised at the same risk level as other patients even though they were actually much sicker.

The Cause

The AI was trained using healthcare expense data rather than actual physical illness data. Since historically less money had been spent on patients in that region, the AI incorrectly interpreted them as being healthier than they actually were.

The Ethical Solution

Applying Bioethics principles could help address the problem.

Principle	Application to the Case
Justice	Ensure that the model understands social determinants of health and does not use flawed financial data as a proxy for actual health.
Beneficence	Use unbiased datasets that accurately reflect the needs of patients from all races and regions.
Transparency / Autonomy	Make data labels and predictions accessible to patients for re-verification.

Quick Revision: Unit 1 at a Glance

Topic	Key Points to Remember
AI Project Cycle	Problem Scoping → Data Acquisition → Data Exploration → Modelling → Evaluation → Deployment
4Ws Problem Canvas	Who, What, Where and Why
Statistical Data	Works mainly with numerical/categorical data and patterns.
Computer Vision	Enables machines to understand images and videos.
NLP	Enables computers to process human language.
Sector-Based Ethics	Ethical frameworks designed for specific sectors.
Value-Based Ethics	Rights-based, Utility-based and Virtue-based.
Bioethics	Autonomy, Non-maleficence, Beneficence and Justice.

Important Board Examination Keywords