Part B:

Subject Specific Skills

Chapter 6:

AI Reflection, Project Cycle and Ethics

6.1: AI Introduction

Since time immemorial, humans have tried to comprehend how our brains, which is a mere biological entity, can think, act, perceive, understand, predict, navigate and modify an external world far larger and more complicated than itself. Artificial Intelligence (AI) is a multidisciplinary field combining computer science, mathematics, neural and cognitive sciences amongst others focused on understanding and building intelligent systems capable of performing simple to complex tasks that typically require human intelligence. These can include problem-solving, data analysis, self-learning, intelligent reasoning, cognitive understanding, and decision-making. In general, AI systems work by taking in large volumes of labelled/unlabelled training data, performing modelling and analysis of the same, and using these output patterns to create intelligent future state predictions.

Historically, AI has been thought about in two fundamental ways. Some experts have defined it as how closely it resembles human performance, while others prefer defining it more abstractly as rationality— which loosely speaking, means doing the “right thing”. The methods used are necessarily different: the pursuit of human-like intelligence must be approached empirically connected to psychology and neuro-sciences, involving close observation and replication of actual human behaviour and thought processes; a rationalist approach, in contrast, involves a combination of applied mathematics, natural sciences, control systems.

1. Important focus areas in AI:

   1. Neural Networks: Area of AI that consists

      of interconnected layers of nodes (like neurons in the human brain) that process data in a way that simulates human intelligence and thought process. Deep learning, one of its specific branches, often involves deep (multi-layered) neural networks.

   2. Machine Learning (ML): Area of AI that

      involves training algorithms to automatically and continuously learn from data to recognize correlations and forecast predictions with minimal human intervention. ML can be of following types:

      • Supervised learning. Machines are trained with structured data for predictive analytics.

      • Unsupervised learning. Machines are trained with unstructured data for predictive analytics, with the model extracting information from the input to identify features and patterns where labels can be applied.

      • Reinforcement learning. Machines learn through trial-and-error iterations, using feedback loops to form actions.

   3. Expert Systems: Area of AI that

      trains computers to develop deep expertise in a single task to solve complex problems just like a human subject matter expert.

   4. Natural Language Processing

      (NLP): Area of AI that enables computers to comprehend, analyse, and create human language. Applications include chatbots, language translation tools, search engines, Call Center IVR (Interactive Voice Response) systems, text and sentiment analysis and many others.

   5. Computer Vision: Area of AI that

      enables machines to meaningfully analyse and understand visual inputs (like images or videos) and makes sense of the visual world. This is used in facial recognition, object detection, quality inspection, traffic management, autonomous driving and many other fields.

   6. Robotics: Area of AI that combines AI

   with mechanical engineering to create and operate robots (industrial or domestic) which in turn perform assigned or intelligent tasks such as navigating directions, providing security, operating in hazardous environments, and interacting with humans.

   
   

2. Applications of AI:

   
   

   • Healthcare: AI is extensively used to augment doctors and medical professionals for disease detection, prognosis and treatment as well as drug discovery in the pharmaceutical industry. Using Machine and deep learning algorithms AI can analyse rapidly and accurately patient health records, medical images and clinical data to predict diagnosis and clinical management. There are significant advances in tele-medicine that AI has contributed to.

     
     

     • Manufacturing: AI plays a key role in Industry 4.0 which is driving digitization of the

       manufacturing industry through intelligent design, manufacturing and supply chain optimization. For example iOT (Internet of things) is driven by AI to make manufacturing shop floors and assembly lines

       connected and intelligent to drive efficient productivity and operations. Imagine in the future cars can be designed by AI rather than human engineers at a fraction of the cost!

       • Finance: AI is extensively used across multiple areas of banking and financial services like credit rating, fraud and anomaly detection, Risk management, stock trading and financial market analysis. AI chatbots are extensively used by banks for customer service at front-end and process bots

         for transaction processing at back-end. For example while sanctioning a bank loan an AI bot can quickly analyse a person's credit rating and decide automatically to grant the same without any human intervention.

         
         

       • Retail: AI supports inventory management, demand forecasting, intelligent supply chain, and efficient store and warehouse operations. It is foundational in the e-commerce industry by providing customer behaviour insights thus enabling personalised customer engagement. Chatbots again are extensively used for automated customer service.

       • Transportation: AI isused in autonomous vehicles, optimises traffic management, route optimization and predictive maintenance of aeroplanes, ships, trains and automobiles. For example in the airline industry AI is increasingly used for air-network optimization and air traffic management which saves billions of dollars for the airline industry.

         
         

       • Entertainment: AI curates content for multiple entertainment channels like cinema, music, video games amongst others. Virtual and augmented reality supported by ML algorithms have also become common. It also plays a key role in customer sentiment analysis for the advertising industry.

   
   

3. Key areas of AI

   We will focus on 3 key areas of AI namely Computer Vision, Natural Language Processing and Data Statistics.

   
   

   1. Computer Vision: Computer Vision is a field of AI that enables computers to derive insights and information from all types of visual inputs (digital images, videos etc.) and take appropriate actions or make recommendations based on the analysed information. If AI enables computers to think, computer vision enables them to see, observe, and understand.

      Computer Vision has 5 stages of development, exactly how a human brain processes visual information that a human eye sees:

      1. Image Acquisition: Just like human eyes see light and create images, a computer uses cameras or sensors to capture images or videos. The process begins with image acquisition, which could be a single frame from a video, a photo from a camera, or even a live feed.

      2. Preprocessing: The computer leverages complex mathematical algorithms to pre-process these images. It breaks down the image into tiny pixels (small picture elements), each with specific colour and brightness values. The image may need to be converted to grayscale (black and white images), normalised (fit within a certain range), or resized to prepare for analysis.

      3. Feature Extraction: The system extracts key features (intensity of light, colour, depth etc.) in the image that are necessary for comprehending its content.

      4. Recognition: After feature extraction, the system compares the features to known labels to recognize objects or patterns. By comparing the distribution patterns of pixels, the computer can differentiate and recognize shapes, colours, and objects variations. For example, it can tell the difference between two people by analysing the shapes and colours of their faces at a pixel level – this is called facial recognition.

      5. Interpretation: The final step involves making sense of the recognized objects, often involving complex decision-making or learning over time. For example, in autonomous vehicles, the computer vision system guides the vehicle "see" the road, recognize other vehicles or obstacles, and decide when to halt or drive.

Computer Vision has multiple practical applications like facial recognition, object detection, medical imaging and robotics. The future of computer vision is quite promising, with advanced R&D pushing the boundaries into the future. Innovationsin deeplearning,particularly convolutionalneuralnetworks (CNNs), have significantly enhanced the predictability of computer vision systems. As technology continues to evolve, computer vision is expected to become more

integrated into human society, making human-machine interaction more natural and intuitive.

Quick Draw (based on Computer Vision-CV)

Quick, Draw is an online game developed by Google that combines fun with artificial intelligence (AI) learning. In this game, players are given a prompt to draw something within 20 seconds, and an AI attempts to guess what they are drawing based on its learned knowledge. This game is not only entertaining but also educational, as it introduces students to concepts in AI, machine learning, and pattern recognition in a fun and interactive way.

Objective: The goal is to draw six objects (one at a time) as accurately as possible within 20 seconds each, based on the prompts provided by the game. The AI will try to guess the object as the student draws it.

AI Guesses: As the student sketches, the AI makes real-time guesses, updating as the drawing evolves. The game ends when all six drawings are completed, and the AI displays whether it successfully identified each drawing.

Prompts: The prompts can range from simple items like a "cat" or "tree" to more complex ones like "microwave" or "skyscraper," requiring students to think quickly and creatively.

Click on https://startcoding.co.in/code Enter the code and click on

Learning Objectives:

Introduction to AI and Machine Learning: Students will experience firsthand how artificial intelligence works by observing how the AI recognizes patterns in their drawings.

Creative Thinking: The game pushes students to think creatively and find the simplest way to represent objects within a short time limit.

Time Management: Students must learn to manage their time effectively to finish their drawing before the 20-second timer runs out.

Pattern Recognition: The AI in Quick, Draw! recognizes common patterns in sketches, teaching students about how computers interpret visual information.

1. Natural Language Processing: Natural language processing (NLP) aims to make computers understand human language. NLP uses computational linguistics, which is the discipline of how language works, and various models based on statistics, machine learning, and deep learning. These technologies allow computers to analyse and process spoken or written languages, and to grasp their contextual meaning, including the speaker’s or writer’s intentions and emotions.

   NLP works as per 8 key steps as given below:

   1. Data Input and Collection

      Data Collection: Gathering text/verbal data from various sources.

      Data Storage: Storing the collected data in a structured format, such as a database or a file system.

   2. Text Preprocessing

      Preprocessing is necessary to cleanse and ready the raw data for further analysis. Common preprocessing steps include Tokenization (Splitting text into smaller units like words or sentences) and various other grammatical clean-ups like text standardisation, spell-checks etc.

   3. Text Representation

      Representing text as a collection of words and grouping of semantically similar words.

   4. Feature Extraction

      Extracting relevant features from the pre-processed data that can be used for various NLP tasks.

   5. Model Selection and Training

      Adopting and training a machine learning / deep learning model to perform specific NLP tasks. These could include Supervised Learning, Unsupervised Learning or Pre-trained Models.

   6. Model Deployment and Inference

      Deploying the trained model and using it to make predictions or extract insights from new text data.

   7. Evaluation and Optimization

      Evaluating the performance of the NLP algorithm using metrics such as accuracy, precision, recall and others.

   8. Iteration and Improvement

      Iteratively improving the model by adding new data, improving preprocessing techniques, adapting with different models, and optimising continuously.

      
      

      Semantris (based on Natural Language Processing - NLP)

      Semantris is an engaging word association game developed by Google that uses AI to interpret and predict word connections. The game helps students enhance their vocabulary, improve their critical thinking, and sharpen their language skills in a fun and interactive way. For grade 9 students, Semantris offers an opportunity to develop these skills while also introducing the concept of how machine learning can understand language.

      Objective:

      To improve vocabulary and word association skills.

      To develop critical thinking by choosing words that are semantically related.

      To introduce students to how artificial intelligence (AI) understands and processes language.

      
      

      Exploration of the Project: Click on https://www.startcoding.co.in/code Enter the code and click on Semantris

      
      

      
      

      Discussion and Reflection Class Discussion:

      After playing, ask students to reflect on the words they used. Did certain clues work better than others? Were some words harder to associate with clues than others?

      Talk about how the game’s AI was able to understand their word associations. This is a great chance to introduce students to the idea of natural language processing (NLP), the field of AI that helps machines understand and process human language.

      Critical Thinking:

      Ask students: What strategies did they develop while playing? Did they focus on synonyms, context, or broader concepts when giving clues?

      Discuss how quickly they had to think during Arcade Mode, and how this helped them strengthen their word association skills.

      
      

2. Data Statistics

   Statistics and applied mathematics play a foundational role in AI. We will study this in details in subsequent chapters but a brief overview is provided below:

   1. Data Collection and Analysis:

      AI systems need vast amounts of data for decision making.

      Statistics support experiment design, data collection and logical synthesis.

      Concepts like descriptive statistics (mean, median, mode, standard deviation) help summarise and characterise the data.

   2. Probability and Inference:

      Probability is a key mathematical concept that AI uses to deal with likelihood to make predictions. For instance, predictive analytics is based on probability assessment.

      Statistical inference allows AI to generalise from a sample of data to extrapolate to a larger population, which is essential in tuning AI models.

   3. Data Distribution and Sampling:

      Understanding data distribution (e.g., normal distribution, binomial distribution) is critical in AI for selecting appropriate models for accurate projections.

      Sampling methods are used to design smaller datasets that represent a larger population, which is important when working with big data in AI.

   4. Machine Learning:

      Machine learning, a major part of AI, is largely built on statistical methods. Commonly used Algorithms like linear regression, decision trees, and support vector machines (SVMs) use statistical techniques for pattern recognition and forecasting. Bayesian methods technique leverages probability to update existing models based on new data ingestion.

   5. Evaluation and Validation:

      Statistical methods are used to assess AI models accuracy. For example, a cross-validation method separates data between training and testing sets to determine how successfully an AI model will perform on unexplored data.

      Metrics like accuracy, precision, recall, and F1-score are statistical tools employed to measure AI model performance.

   6. Hypothesis Testing:

      In AI, hypothesis testing is used to prove the performance of models. Techniques like p-values and confidence intervals validate and improve the decision-making process.

      
      

      Acquiring and Analysing Data from a Data Set Objective:

      Students will learn how to acquire, organise, and analyse data from a pre- existing data set. This project will introduce them to basic data handling, including how to identify trends, perform simple calculations, and draw conclusions from the data.

      Skills Developed:

      • Data acquisition and handling Basic

      • data analysis and interpretation

      • Graphing and visualising data

      • Critical thinking and drawing conclusions

Select a Data Set

Provide students with a simple data set or allow them to select one from publicly available Data Sets (e.g., weather data)

Example Data Set: A data set containing daily temperatures over a one- month period from a weather station in a city. The columns may include:

Date

Daily high temperature Daily low temperature Precipitation (rainfall)

Explore and Understand the Data Set

In this example, explain how temperatures are recorded daily and what precipitation means.

Data Acquisition:

Students will learn how to access and acquire the data, either by downloading it from a source or by accessing it via a file provided by the teacher.

Data Organization

Have students organise the data in a table format, ensuring that they correctly label columns and rows. This can be done in a spreadsheet program like Google Sheets or Excel.

Data Analysis

Find the average daily temperature by averaging the high and low temperatures for each day.

Calculate the monthly average high and low temperatures. Identify the hottest and coldest days.

Create a line graph of daily high and low temperatures. Create a bar chart for total precipitation over the month. Use graphs to visually display the trends they discovered.

Based on their analysis, students will write a brief summary of their findings, addressing questions such as:

What were the warmest and coldest days of the month? What trends did you notice in the temperature data?

Was there a pattern in the precipitation?

1. : AI Project Cycle

   In the IT industry, real-life AI applications are built as a structured approach on how AI projects should be implemented across teams. This approach or methodology is called the AI Project Cycle having 6 steps as given in the diagram below. We will study in details each of these steps:

   
   

   1. Problem Scoping

      Problem scoping is the phase where you figure out what the project is about, what your goal is, and how you’re going to achieve it. It helps you clearly understand the problem you’re trying to solve, so you don’t get confused or lost halfway through the project.

      Any project to be successful, has to have a clear definition of what measures and metrics needs to be achieved to have a successful outcome. For example, in a climate change AI project, success might mean accurately predicting future temperatures or making sure AI helps reduce pollution by a certain amount.

      The scope of the project (what it includes but also importantly what it excludes) and the constraints also need to be clearly understood. For example, if you’re working on a school project about solar energy, you might decide to include research on how solar panels work, but exclude topics like wind energy or hydroelectricity. This helps you stay focused and not get distracted by things that aren’t part of your goal.

      Problem Scoping phase precisely does that to reach a common and clear understanding with all project stake-holders on the scope, constraints and measures of success of a project. In a project, constraints could be:

      • Time: You might only have a few weeks to finish.

      • Money: There’s a budget to stick to.

      • Resources: You might not have access to all the data or technology needed.

        Stakeholders are the people involved in the project. During Problem Scoping, everyone talks about the project to make sure they all understand what needs to be done and how success will be measured.

        Without this phase properly accomplished a project cannot be successful. Here’s what could go wrong:

      • Confusion: The team might not know what the real goal is.

      • Wasting time: If the scope isn’t clear, you might end up doing extra work that isn’t needed.

      • Failure: The project could fail because no one agreed on the challenges or how to deal with them.

        Here's a detailed breakdown of the problem-scoping phase: Problem Identification

      • Assess Business Context: Talk to everyone involved to understand the problems they face and where AI could help. Engage with ALL stakeholders (people who play a key role in the outcome of the project) to assess the business scenarios, existing processes, and challenges they are facing and opportunities they may have. Given the challenges/opportunities identify where AI can add value.

      • Stakeholder Interviews: Ask the stakeholders questions to gather details about their challenges and opportunities. Standard questionnaires can be created and used for the stakeholder interviews.

      • Core Problem Definition: Based on what we learned, figure out the

        main problem that AI will fix.

        Define the problem :

        The 4Ws Problem Canvas is a tool used for problem scoping by breaking down a problem into four main components: What, Why, Who, and Where. This canvas helps identify the problem in depth, the stakeholders involved, the reasons the problem exists, and the areas or regions impacted.

        Here’s a 4Ws Problem Canvas applied to the issue of Climate Change and Its Impact:

        
        

        4Ws Problem Canvas

        The 4Ws Problem canvas helps in identifying the key elements related to the problem. The 4Ws are:

        1. Who

        2. What

        3. Where

        4. Why

        Let’s understand the 4Ws by analysing the “climate problem” on Earth.

        1. What?

           What is the problem?

           Core Problem: The Earth's climate is rapidly changing, with rising temperatures, increasing extreme weather events, and alterations in ecosystems and biodiversity

        2. Why?

           Why does this problem exist?

           Human Causes:

           Deforestation: Cutting down forests for agriculture and urban development reduces the Earth's capacity to absorb CO2.

           Industrial Processes: Factories and industries contribute to pollution and release harmful gases that trap heat in the atmosphere.

           Natural Causes:

           Volcanic activity, changes in solar radiation, and natural variability also influence the climate but at a much smaller rate compared to human activities.

        3. Who?

           Who is impacted by the problem? Who is involved in solving the problem?

           Impacted Stakeholders:

           Global Population: Extreme weather events affect everyone, but vulnerable populations like those in coastal regions, low-income communities, and island nations are most at risk.

           Wildlife and Ecosystems: Animals and plants face habitat loss, food scarcity, and shifts in their natural environments.

           Governments and Economies: Climate change increases disaster recovery costs, threatens infrastructure, and can destabilise economies.

        4. Where?

        Where is this problem happening? Global Impact:

        Worldwide: Climate change is a global issue, affecting every part of the world, with its impacts being felt differently across regions.

        Developing Countries: Often more vulnerable to climate change due to less resilient infrastructure, fewer resources for adaptation, and economies reliant on agriculture.

        Local Impact:

        Urban Areas: Cities face higher temperatures, heat islands, and increased air pollution.

        Rural Areas: Farmers experience changing weather patterns that affect crop production and water availability.

        
        

   2. Data Acquisition

      Data acquisition means collecting the data that an AI system needs to learn and make decisions. It’s the first step where you gather data from different sources to use in your project.

      In short, data acquisition is about gathering the right data you need to start working on a project.

      Just like a student needs textbooks and practice problems to learn, an AI needs data to understand and make predictions. For instance, if we want an AI to recognize different animals, we need to show lots of pictures of those animals.

      AI uses the data it learns from to make guesses or predictions. For example, an AI trained with pictures of fruits can predict if a new picture shows an apple or a banana.

      More and better data helps the AI make more accurate predictions. If the data is incomplete or wrong, the AI might make mistakes.

      1. Steps in Data Acquisition:

         1. Data Gathering and Organization:

            • Gathering Data: Initiate the process of gathering and downloading from all the pre-existing and new data that you have identified as data sources (surveys, interviews, focus groups, experiments, online analytics tools, sensor data, document analysis, social media analysis, observations, etc.)

            • Data Organization: The data gathered has to be assembled and organised in a meaningful and consistent manner so that it's easy to reference it for further analysis. For example , similar objects could be grouped into a file with that object name.

         2. Data Collection:

            Once the sources are identified, the next step is gathering the data.

         3. Data Integration:

            After collecting data from different sources, the data often needs to be merged or integrated.

         4. Data Cleaning:

            Raw data is often incomplete, inconsistent, or noisy. Data cleaning ensures the data is accurate and suitable for analysis.

            Tasks include:

            • Handling missing values.

            • Removing duplicates.

            • Correcting errors and outliers.

            • Standardising formats (e.g., dates, categories).

         5. Data Storage

            • Save Data: Data has to be stored and secured carefully like individual on- premise Database servers or external Cloud Data Centers.

            • Backup Data: Since loss of data could impact the entire project, it needs to be ensured that there is secondary backup of the primary data storage. For example all Cloud service providers provide multiple redundancies to ensure there is always backup of primary data storage.

         6. Data element Documentation

            • Mark Data Sources: Identify and document each data source mapped to each data element as it provides a way to trace back if needed and explain to stakeholders.

              • Catalogue Data: Catalogue each data element which is like a data dictionary (also called Metadata i.e. Data about Data!).

         7. Analysis and Feedback

            • Request Feedback: Continuously and periodically review with your stakeholders to ensure that data is accurate and complete before further processing.

            • Rectify and fix: If data is missing or incomplete collect additional data to make it complete or do data cleansing to rectify existing data.

         
         

         Case Study : Data Acquisition for AI

         1. Decide What You Want to Do:

            Let’s say we want to create an AI that can identify different types of pets from pictures, like dogs, cats, and rabbits.

         2. Collect Data:

            You need to gather pictures of the pets you want the AI to recognize. You might take your own photos or find pictures online. Use methods like downloading images from the internet, taking pictures yourself, or using a dataset that already has labelled pet images.

         3. Prepare the Data:

            Make sure the images are clear and correctly labelled. For example, remove blurry photos or fix any mistakes in labelling. Tag each photo with the correct label, such as “dog,” “cat,” or “rabbit,” so the AI knows what it’s looking at.

            
            

         4. Organise the Data:

            Divide your data into different groups:

            • Training Data: The bulk of the data that the AI will learn from.

            • Validation Data: A smaller set to check how well the AI is learning.

            • Test Data: A final set to see how well the AI performs on new, unseen data.

         5. Using the Data:

         Feed the training data into the AI so it can learn from it. After training, test the AI with the test data to see how well it can recognize pets it hasn’t seen before.

      2. Ethical judgement

         • Data Privacy: Ensure data collection meets all legal and ethical requirements. For example if personal data is collected from people their prior approval should be taken and it should be stored securely.

         • Holistic Representation: Ensure there is no sampling bias in the data collected so that it represents the target population holistically and fairly. For example while analysing voting sentiments, the people sampling should be done across regional, language and socio-economic segments.

      3. System Maps

      System Maps help us to find relationships between different elements of the problem which we have scoped. Identify the elements of the system. Let us discuss on The delayed effects of climate change.

      A system map for the delayed effects of climate change captures how climate change triggers long-term consequences that manifest over time. These delays are due to the slow response of natural systems (like oceans and ice sheets), infrastructure, and ecosystems to changing environmental conditions. The map would visualise the inputs driving climate change, the immediate impacts, and the feedback loops that lead to delayed, often compounded, effects.

      Let’s find the elements to draw a system map for the delayed effects of climate change:

      1. Inputs (Drivers of Climate Change)

         • Greenhouse Gas Emissions (GHGs):

           CO₂, Methane(CH₄), Nitrous Oxide (N₂O), etc., from human activities (burning fossil fuels, deforestation, industrial agriculture).

         • Land Use Changes:

           Deforestation, urbanisation, and agriculture, altering natural carbon sinks and increasing emissions.

         • Pollutants and Aerosols:

           Airborne particles that can affect atmospheric temperatures and cloud formation.

         • Industrialization:

         Increased energy use, transportation, and production contributing to emissions.

      2. Immediate Climate Impacts

         Rising Global Temperatures: Warming of the atmosphere and oceans.

         Melting of Ice and Glaciers: Immediate loss of ice due to warming temperatures.

         Increased Frequency of Extreme Weather Events: More frequent and severe storms, droughts, heatwaves, and flooding.

         Ocean Acidification: Absorption of excess CO₂ by oceans, affecting marine ecosystems.

      3. Delayed Effects (Long-term Impacts) mentioned as Big Problems

         1. Delayed Physical and Environmental Effects

            Sea-Level Rise: Melting glaciers and thermal expansion of seawater due to warming cause gradual sea-level rise, impacting coastal areas over decades.

            Permafrost Thawing: Melting permafrost slowly releases trapped methane and CO₂, further accelerating warming in a delayed feedback loop.

            Ocean Circulation Changes: Slowing of thermohaline circulation (e.g., the Atlantic Meridional Overturning Circulation), which can alter global climate patterns but maytake decades to manifest.

            Species Migration and Extinction: Ecosystem changes that unfold gradually as species either adapt, migrate, or face extinction due to shifting habitats and food sources.

            Changes in Agricultural Productivity: Gradual shifts in growing seasons, crop viability, and water availability leading to food insecurity.

            Desertification: Land degradation in arid regions as a slow result of warming, leading to loss of fertile land over time.

         2. Delayed Societal and Economic Effects

            Climate Refugees: Long-term displacement of populations due to sea- level rise, extreme weather, and agricultural collapse. The social effects may take decades to fully manifest.

            Infrastructure Strain: Gradual degradation of infrastructure due to extreme weather, sea-level rise, and heat stress (roads, buildings, water systems).

            Economic Losses: Cumulative long-term economic impacts from damage to agriculture, fisheries, tourism, and insurance costs related to extreme weather and sea-level rise.

            Public Health Impacts: Delayed onset of diseases like malaria, respiratory issues, and heat-related illnesses as climates change over time.

            • Political Instability: Long-term pressure on governments due to resource shortages, displaced populations, and economic stress leading to potential conflict.

      4. Feedback Loops

         • Ecosystem Degradation: Loss of forests and wetlands reduces natural carbon sinks, further increasing atmospheric CO₂ levels over time.

         • Water Cycle Intensification: As warming continues, evaporation increases, leading to more intense but uneven rainfall, causing long-term shifts in water availability.

      5. External Factors

      • Social and Behavioural Change: Delayed societal shift in consumption habits, transportation, and lifestyle choices prolongs the reliance on fossil fuels and high-carbon practices.

      This system map outlines the cascading, delayed impacts of climate change and how they reinforce one another over time. The map here shows cause & effect relationship of elements with each other with the help of arrows. The arrow- head depicts the direction of the effect and the sign (+ or -) shows their relationship. If the arrow goes from one element to another with a + sign, it means that both are directly related to each other. If the arrow goes from one element to another with a – sign, it means that both the elements are inversely related to each other. To explore and draw System Maps click on the link below.

      
      

      
      

      Click https://www.startcoding.co.in/code Enter the code and Click on System Maps

      
      

   3. Data Exploration

      The Data Exploration phase in an AI project helps to derive a deep multi- dimensional understanding of the data collected before creating your AI model. The primary purpose of data exploration in the AI project cycle is to discover patterns and insights in data.

      
      

      1. Data Assessment

         • Basic Data Assessment: Initiate a high-level understanding of the data collected and arranged. This could be a basic classification of the data types, data quantities etc.

         • Data Types: These could be numeric, text, audio, video amongst many others.

           
           

      2. Data Visualisation

         • Charts and Graphs: Create simple infographics like charts or graphs for data visualisation and developing insights. For example, simple statistical tools like pie charts and bar charts will help visualise the distribution of a data set. These visual displays of information communicate complex data relationships and data-driven insights in a way that is easy to understand.

           Let’s see a tool to visualise data :

           
           

           Open https://www.startcoding.co.in/code Enter the code and click on

           Click on Data Visualisation Tool

           With this tool we will be exploring various types of Graphs. To analyse the data, you need to visualise it in some user-friendly format so that you can:

           Get a sense of the trends, relationships and patterns contained within the data.

           Define strategy for which model to use at a later stage. Communicate the same to others effectively.

           Some simple and common data visualisation techniques are :

           1. Bar Graphs

              A graph that uses rectangular bars to show the value of different categories.

              Use Case: Great for comparing data between different groups or categories.

              Example: Showing the number of students in different sports teams at school.

              
              

           2. Line Graphs

              A graph that uses lines to show how data changes over time.

              Use Case: Perfect for tracking trends or changes over a period of time.

              Example: Tracking temperature changes throughout the year.

              
              

           3. Pie Charts

              A circular chart divided into slices that show proportions of a whole.

              Use Case: Useful for showing how different parts contribute to a total.

              Example: Showing the percentage of students involved in different extracurricular activities.

              
              

           4. Histograms

              A graph that looks like a bar graph but is used to show the distribution of data.

              Use Case: Useful for showing how often data falls into different ranges (frequency).

              Example: Displaying the number of students scoring within certain grade ranges (like 60-70, 70-80).

              
              

           5. Scatter Plots

              A graph that shows individual points plotted on a grid to display the relationship between two variables.

              Use Case: Useful for identifying correlations or patterns between variables.

              Example: Plotting students' study hours versus their test scores to see if more study leads to better grades.

              
              

           6. Pictographs

              A graph that uses pictures or icons to represent data.

              Use Case: Ideal for younger audiences or simple data that needs a visual touch.

              Example: Using icons of books to represent the number of books read by students in different classes.

              
              

           7. Area Charts

              Similar to a line graph but the area below the line is filled with colour.

              Use Case: Good for showing how the total value changes over time.

              Example: Showing how the total amount of rainfall has changed each month.

              
              

           8. Tables

              A way of organising data into rows and columns.

              Use Case: Useful when you need to present exact numbers.

              Example: Displaying students' names along with their grades for each subject.

              
              

           9. Bubble Charts

              Similar to a scatter plot but uses bubbles (circles) where the size of each bubble represents a third variable.

              Use Case: Good for showing the relationship between three variables.

              Example: Plotting the number of students in different sports and using bubble size to show team size.

              
              

           10. Venn Diagrams

           A diagram that uses overlapping circles to show relationships between different sets.

           Use Case: Useful for showing how different groups have things in common.

           Example: Showing which students participate in basketball, soccer, and both sports.

           
           

           
           

           Click on www.startcoding.co.in/code Enter the code

           and click on

      3. Pattern Recognition

         • Find Patterns: Try to identify any obvious patterns or trends that come out from the data visualisation. For example, say certain data elements get repeated at a higher frequency than other data elements which may signify an underlying deeper pattern.

         • Outlier analysis: Identify any data that does not fit an existing pattern, like an image labelled as “car” that looks more like a bus or train. These are called outliers.

      4. Data mismatch and correction

         • Missing Data: Any missing data element needs to be identified. As an example if the data set was expected to have 1000 units of square shape, but has less or more than that the reason has to be identified.

         • Incorrect Data: This could be wrong labelling or duplicate entries. For example, if two words are repeated twice (by mistake), you might only need to retain one.

      5. Data Summary

         • Basic Statistics: Use statistical summary tools (Mean, Median, Mode, Standard Deviation etc.) to summarise data congruence. For example, what is Median of height of students in Class 9 or how many times in a week (Mode) there are traffic issues on a particular road?

         • Data Distribution: Distribution of Data looks at the frequency in which a data element is spread out over a certain range of values. This can be done using frequency tables as an example.

      6. Inductive Questions

         • Design simple questions to induce from your data and try to assess fundamental and underlying patterns that may exist. For example in a data set of colour patterns:

           Which colours have the highest frequency of occurrence? Are there any colours that are outliers?

           Is there any trending or shading from one colour type to another?

      7. Record Findings

         • Record Observations: Record all your observations that were found during your exploration phase. This could include the data sets, data elements within those sets, anomalies detected, cleansing and corrections made and initial patterns observed.

         • Visual Documentation: You can use simple software tools like MS Excel to create and document statistical tools like plotting, distribution etc and use them for future reference.

      8. Requesting Feedback

      1. Share Findings and request feedback: Share and request feedback on documented findings. Stakeholder feedback and peer review is important to ensure that nothing is missed out.

      2. Finetune and adjust: Based on feedback, if there are gaps, iteratively fix the same within the data sets.

      
      

      Case Study : Top 10 Song Prediction: Identify the data features, collect the data and convert into graphical representation.

      We can predict whether a song will make it to the Billboard Top 10 by analysing various features, such as the genre, tempo, song duration, and especially the popularity of the singer. Here's how the graphical representation of data can help in understanding trends and patterns to design a strategy for achieving the project’s goal.

      
      

      Key Data Features for Prediction:

      1. Current Trends in Music: Popular genres of music styles.

      2. Genre of Music: Different genres may have different success rates in reaching the Billboard Top 10.

      3. Tempo of the Song: The speed of the song (BPM- beats per minute) can influence its popularity.

      4. Duration of the Song: The length of the song may impact its radio playtime and streaming behaviour.

      5. Popularity of the Singer: More well-known artists are more likely to reach the Top 10.

         Using Graphs for Pattern Analysis

         1. Graph: Singer Popularity vs. Chart Success Graph Type: Bar Graphor Scatter Plot

            X-Axis: Popularity score of the singer (based on social media followers, streaming numbers, etc.).

            Y-Axis: Billboard ranking (or probability of making the Top 10).

            Purpose: This graph will help you identify if there is a direct relationship between the singer’s popularity and their likelihood of getting a song into the Billboard Top 10.

            Insight: If there’s a clear trend that more popular singers are consistently appearing in the Top 10, it shows how influential the artist’s popularity is. You can use this insight to give more weight to the popularity feature in your prediction model.

         2. Graph: Genre vs. Success Rate

            Graph Type: Pie Chart or Bar Graph

            X-Axis(if bar graph): Different genres (Pop, Hip-Hop, Rock, etc.).

            Y-Axis(if bar graph): Percentage of songs that made it to the Top 10.

            Purpose: To analyse which music genres have a higher chance of reaching the Top 10.

            Insight: Certain genres may be more likely to dominate the charts based on current trends, so this feature should be given more attention in your model.

         3. Graph: Tempo of Music vs. Chart Success Graph Type: Line Graph or Histogram

            X-Axis: Tempo of the song (BPM).

            Y-Axis: Number of songs in the Top 10.

            Purpose: To explore whether faster or slower songs have a better chance of success.

            Insight: If you notice that songs within a certain tempo range are more commonintheTop10, your predictive model can incorporate this information to improve accuracy.

            
            

         4. Graph: Song Duration vs. Chart Success Graph Type: Box Plot or Scatter Plot

            X-Axis: Song duration (minutes). Y-Axis: Billboard ranking.

            Purpose: To analyse whether song length affects the likelihood of making the Top 10.

            Insight: If shorter or longer songs dominate the charts, you can tailor your predictions by paying attention to this feature.

         5. Graph: Current Trends in Music Graph Type: Line Graph

      X-Axis: Time (months/years).

      Y-Axis: Popularity of different genres or music styles over time.

      Purpose: To track how musical trends evolve over time and influence chart success.

      Insight: Understanding current trends will allow you to fine-tune your model based on what’s popular now, rather than relying on outdated patterns.

      Using graphical representations, like the ones mentioned above, helps to visually identify trends, relationships, and outliers in the data. By understanding these patterns, you can design a more accurate predictive model to forecast whether a song is likely to make it into the Billboard Top 10.

      
      

   4. Modelling

      In this phase the machine learning model is developed and trained using the data sets collated and organised from the previous phases. It consists typically of three phases:

      • Model Selection

      • Data Splitting (Training & Testing)

      • Training the Model

      1. Model Selection

         Model types: Models can be classified in the following way as below. We will discuss them in detail.

         
         

         Learning based Approach:

         In a learning-based approach the system is designed to "learn" from examples instead of following fixed rules. This type of AI, often called machine learning, can look at a lot of information (data) and find patterns or make decisions based on that data. It gets better over time by improving its understanding as it processes more information.

         Real life example: Voice Assistants like Siri or Google Assistant can recognize your voice and understand what you’re asking because they’ve been trained on millions of voice recordings.

         Machine Learning and Deep Learning are two alternate approaches within Learning based Approach, but they have different approaches, capabilities, and applications. The key differences between them are summarised below:

         • Machine Learning (ML):

           ML approach focuses on creating models that can learn from and predict potential outcomes or take decisions based on input data. ML algorithms iteratively improve and fine-tune their performance over time as they get exposed to repeated volumes of data.

           ML involves techniques like linear regression, decision trees, support vector machines (SVMs), and clustering methods like K-Means.

           • Deep Learning (DL):

             The DL approach uses neural networks (a mathematical approach that has multiple and interconnected nodes just like the neurons in Human brain) with embedded layers (hence "deep") to model complex data patterns.

             DL models, such as Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), are widely used for activities like image, speech recognition, and complex sentiment analysis.

             Machine Learning in turn are of 3 categories as follows:

             • Supervised Learning: Supervised Learning is a family of ML algorithms where the model is progressively trained using a labelled dataset. This essentially means that the model learns from questions and answers where the correct answers are already given. As an example for an examination all answers are labelled and categorised and the model by looking at the questions essentially matches (supervised) and learns from the labelled answers. Examples of supervised learning are spam detection, picture and image identification etc. The Model is given a dataset of inputs (like questions) and corresponding outputs (labelled answers) and the model learns to associate between both of them. After the model is trained it can be given new data sets where it can make predictions based on what it has learned. There are 2 types of Supervised Learning:

               • Regression: This is a mathematical tool used when the outcome is a number. For example predicting stock market price based on its underlying dependent variables could be done through Regression analysis.

               • Classification: This is a mathematical tool used when the outcome is a type or category (not a number). For example, classifying whether a surface is smooth or curved based on its texture is a Classification problem.

                 Supervised Learning is generally accurate if there is a lot of underlying data which is accurately labelled and is easy to as well. But it does not work well if large parts of the data input is not labelled or the data quantity is not large enough. It could also be time consuming to train a model on large data sets.

             • Unsupervised Learning: Unsupervised Learning is a family of ML algorithms that operate on data sets that are not labelled. Unlike supervised learning, where the model knows the correct answers (labelled answers) while learning, in unsupervised learning, the model researches the data through its own pattern recognition techniques. For example the problem statement could be that the model is given pictures of fruits without labels and it has to (on its own) identify patterns to correctly classify the fruits. The way it works is that the model tries to find similarities or differences and tries to cluster them together as groups or identify underlying data patterns. For example, in this case based on the picture texture of different types of fruits (not labelled) it can try to group or cluster similar looking fruits together. Once the clustering of different categories is completed the model can start differentiating different categories of fruits even though it was not known to it earlier.

               Examples include weather pattern recognition, anomaly detection, shopping behaviour analysis amongst others. Please note that in all these examples its not possible to answer the question beforehand as they are not known in advance.

               A few types of Unsupervised Learning include:

               • Clustering: Grouping or arranging data points together based on their similarity or differences so that eventually distinct clusters can be obtained from the data points.

               • Dimensionality Reduction: This is a technique to reduce less important parameters/dimensions of a data set while retaining the more important ones which simplifies the problem considerably.

             Unsupervised Learning has multiple advantages as it enables us to model on data sets that are not labelled as well as identify hidden patterns not previously understood by us.

             
             

             • Reinforcement Learning: Reinforced Learning is a family of ML algorithms where the model is made to operate against an environment (instead of against labelled data as in the case Supervised Learning) where the environment trains ”reinforces” the model based on a system of rewards and penalties (based on if the model is right or wrong) to eventually reach the correct decision or outcome. So, the two parties are the model is the “Agent” which learns from the “environment” by continuously interacting with it. A good example is how a child learns to differentiate between hot and cold objects. Touching a hot object imposes a painful “penalty” on the child and over time s/he recognizes to avoid touching the same. The steps involved typically include:

               • Defining the environment

               • Define action types by the agent which can explore (unknown actions) or exploit (known actions which give higher penalties)

               • Define the reward/penalty that the agent will face by performing action with the environment

               • Define the agent policy which will try to maximise rewards and minimise penalties

               • Train the agent by actioning on the environment and over repeated attempts learn to drive the right and desired action

             Examples include self-driving cars, Robotics and Human-AI games (like chess with computers).

             Based on the nature of the problem and data available start with an appropriate model that is easy to understand and implement.

             
             

             Rule Based Approach

             A rule-based approach in AI means that the system follows specific instructions or rules that have been set by a programmer. It’s like giving the AI a list of "if-then" instructions to follow for different situations. Each rule tells the system exactly what to do when it encounters a particular situation.

             
             

             • Example Rule: If the customer says “What’s your return policy?”, then

               respond with “Our return policy is 30 days.”

             • Example Rule: If someone asks “What time is it?”, then the AI checks the current time and responds with it.

             Decision Trees is a rule-based approach because it uses a tree-like model of decisions and their possible consequences, including chance event outcomes. Each node represents a feature (attribute), each branch represents a decision rule, and each leaf node represents an outcome.

             Let’s say you want to decide what to wear based on the weather. Your decision tree might look like this:

             
             

             
             

             Real life example: ATM Machines work based on rules. For example, if you enter the correct PIN, it lets you withdraw money. If the PIN is incorrect, it won’t give access.

      2. Splitting the Data

         The selected model has to use the data available both for training and then for testing, so the data is split into 2 parts:

         Training Set: Portion of the data used to train the model.

         Testing Set: Portion of the data used to evaluate model performance.

      3. Training the Model

         • Feed Data to the Model: Use the training portion of data to tutor the model. The model starts getting activated with the data ingestion and actions taken based on that.

         • Iterate on input Parameters: During training, based on training outcomes (success or failure) the input parameters need to be adjusted to make the training better and optimised.

      
      

      
      

      Case Study of Rule-based Vs Learning-based AI Models

      A company Xernia wants a chatbot on its website to answer customer questions about things like product details, order tracking, return policies, and troubleshooting. They have two options for building this chatbot: using a rule-based model or a learning-based model.

      Option 1: Rule-Based AI Model

      A rule-based chatbot follows specific rules or instructions set by a programmer. Every possible question or phrase has to be programmed, like “If a customer says ‘track order,’ show them the tracking link.”

      Example:

      For a question like, “How can I check my order status?” the rule-based bot recognizes “order status” and responds with a link or instructions for tracking the order.

      Pros:

      1. Predictable Responses: The bot’s answers are always the same, which makes them predictable.

      2. Simple Setup: This bot is easy to set up for straightforward questions that don’t change much.

      Cons:

      1. Limited Understanding: It may not understand questions if they’re phrased differently, like “Where’s my package?” instead of “order status.”

      2. No Learning: The bot won’t get better over time. Every new type of question needs a new rule to be added manually.

      
      

      Option 2: Learning-Based AI Model

      A learning-based chatbot uses machine learning, which means it learns from examples instead of fixed rules. It’s trained with lots of sample questions and answers to understand and recognize patterns in language.

      
      

      Example:

      For the question, “How can I check my order status?” the learning- based bot recognizes the question’s intent (wanting to track an order) and responds with the same instructions.

      It can also recognize different ways of asking the same question, like “Where’s my package?” or “Track my order.”

      
      

      Pros:

      1. Flexible Responses: It can understand questions even if they’re worded differently, which improves the experience.

      2. Learning Ability: The bot can get better over time by learning from new data.

      Cons:

      1. Complex and Time-Consuming: Training the bot takes more time, data, and resources.

      2. Less Control Over Answers: Responses may sometimes vary, so the bot might need some monitoring.

         
         

         Activity : Supervised Learning : Rock, Paper and Scissors

         Rock-Paper-Scissors game using AI can be a fun way to understand how computers can learn and make decisions. In this game, the AI’s goal is to guess or predict what the player will choose next and try to counter it with the winning move.

         Exploration of the Project:

         
         

         Click on www.startcoding.co.in/code Enter the

         code and click on

   5. Evaluation

      The Evaluation Phase is where after training of the model is completed the performance of the model is assessed across all dimensions. This phase is very important as it determines if the model is performing as per set criteria and whether it’s ready to be deployed into the real world.

      Reality = YES Reality = NO

      Prediction = YES TRUE POSITIVE FALSE POSITIVE

      Prediction = NO FALSE NEGATIVE TRUE NEGATIVE

      As an example, let’s consider a model trained to predict a type of fruit. Consider that the fruit is an Apple in reality, let’s see how the model could predict various scenarios as above:

      TRUE POSITIVE

      Prediction Image

      Reality Image

      TRUE NEGATIVE FALSE POSITIVE

      No image

      No

      No image

      image

      FALSE NEGATIVE

      No image

      Evaluation process consists of testing the Prediction (by the model) against Reality (actual or real results). Typically, you start by feeding data (which you know in reality will give 100 % desired results i.e. “Reality”) into the model and observe the “prediction” the model makes. The comparison can give 4 permutations:

      
      

      Confusion Matrix is a table which evaluates the performance of the Classification model. The rows and columns represent the Reality classes (what is actually true) and the predicted classes (what the model guessed). Each cell in the matrix shows the total number of predictions made by the model that fall into each category. A basic Confusion Matrix is shown below:

      
      

      Actual: Yes Actual: No

      Predicted: Yes

      No. of True Positives (TP) No. of False Positives (FP)

      Predicted: No

      No. of False Negatives (FN) No. of True Negatives (TN)

      
      

      The Evaluation of the Model is done using the Confusion Matrix along 4 parameters as given below:

      Accuracy: This measures the overall correctness of the model as per the formula below:

      Accuracy = (TP+TN)/(TP+TN+FP+FN)

      Precision: Ratio of positive predictions that were actually correct Precision = TP/(TP+FP)

      Sensitivity/Recall: Ratio of actual positives that were identified correctly by the model

      Sensitivity = TP/(TP+FN)

      F1 Score: Harmonic mean of Precision and Recall

      F1 Score = 2* (Precision* Recall)/(Precision + Recall)

      A Confusion matrix is an essential analytical tool to handle data which is not balanced by giving a more detailed understanding and hence improving the overall performance of the model.

      
      

      ROC Curve : Receiver Operating Characteristic Curve

      The ROC curve is a graph that shows the performance of a classification model at all possible thresholds( threshold is a particular value beyond which you say a point belongs to a particular class). The curve is plotted between two parameters

      • TPR – True Positive Rate

      • FPR – False Positive Rate

      ROC is a metric used to find the accuracy of a model

      
      

      Evaluation Case Study : Weather Forecast - Prediction Vs Reality Scenario:

      A local weather station predicts the weather for a small town over a period of three days. They rely on data from satellites, weather stations, and computer models to make the prediction.

      Day-by-Day Forecast vs Reality Day 1:

      • Prediction: Sunny with clear skies, temperature around 25°C.

      • Reality: Sunny and clear, temperature 26°C.

        The prediction was very close to reality. This is an example of how short-term weather predictions are often highly accurate.

        Day 2:

      • Prediction: Light rain in the afternoon, with cloudy skies throughout the day, temperature around 22°C.

      • Reality: Mostly cloudy, but no rain. Temperature 23°C.

        The prediction was partially correct (clouds), but the rain didn’t happen. Sometimes, weather systems weaken or shift direction, resulting in different outcomes than expected.

        Day 3:

      • Prediction: Heavy thunderstorms, strong winds, and cooler temperatures around 20°C.

      • Reality: Light drizzle in the morning, clearing up by noon with temperatures around 24°C.

        
        

        In this case, the prediction was more inaccurate. Thunderstorms were expected, but only light rain occurred. Weather models sometimes overestimate storm severity, and local factors (like winds or humidity) can change weather patterns.

        Why Did the Forecast Differ from Reality?

      • Atmospheric Complexity: Weather systems are dynamic and influenced by many factors like wind, air pressure, and moisture. A small change in any of these can alter the weather significantly.

      • Model Limitations: Weather forecasting models are based on historical data and simulations. While they are highly advanced, they cannot always predict how conditions will evolve, especially over longer periods.

      • Local Geography: Features like mountains, lakes, or oceans can influence local weather in unexpected ways.

      Images to Illustrate the Case Study

      To help students understand these concepts, let me create a simple image showing the forecast versus the actual weather over the three days:

      Day 1: Predicted sunny, reality sunny.

      Day 2: Predicted cloudy with rain, reality cloudy but no rain.

      Day 3: Predicted thunder storms, reality light drizzle with clear skies in the afternoon.

   6. Deployment

Deployment is the final stage of the AI process where the model which is trained and successfully evaluated can be deployed into the real world and start performing live functions. The software code for the AI model could be in a program, a robot or an iOT device. Once deployed the real world impact of the model vis-a-vis its original business objectives can truly be measured.

Steps in the Deployment Process

1. Selection of Deployment Environment:

   The deployment environment could be a web-page, desktop application or a smartphone device based on the business environment. Example: If your model does factory automation, the model could be deployed on shop floor robots.

2. Model Preparation and integration:

   Specific deployment environments need specific data formats hence the model needs to be converted to the same. This might involve exporting the model into the target environment file type or format.

   Integrate the model into the application or system where it will be used.

3. Deployment Testing:

   Before going live in a production environment, test the deployed model in a production like environment to ensure it performs as expected. This helps identify any issues that could be missed out during the initial evaluation and validation phases.

4. MonitortheModelinAction:

   After deployment, monitor the model to see how it performs over time. Sometimes models need updates or retraining as they encounter new types of data.

   
   

   Deployment Case Study: “AI Sowing App” deployed for Crop Health Monitoring by Microsoft and ICRISAT

   Context and Problem: In India, many farmers depend on rain and other natural factors to grow their crops. However, unpredictable weather, such as sudden rainfall or drought, can damage crops and lead to poor harvests. Additionally, many farmers need help deciding when to plant seeds or how to protect their crops from pests. This is especially challenging in rural areas with fewer resources.

   Objective: The goal was to leverage Artificial Intelligence (AI) to help Indian farmers maximize crop yield, reduce losses, and improve decision-making. The specific objectives were to:

   1. To help farmers make better decisions about planting and harvesting.

   2. To provide farmers with weather predictions to protect their crops.

   3. To reduce crop losses due to unexpected weather and pests.

Solution by Microsoft and ICRISAT: Microsoft partnered with the International Crops Research Institute for the Semi-Arid Tropics (ICRISAT) to develop an AI-driven platform called the AI Sowing App. This app provided farmers with timely advice on optimal sowing time, soil treatment, and irrigation.

Key AI components included:

1. Weather Data Analysis: Using AI algorithms to analyze historical and real- time weather data, the app predicted the best time for sowing based on rainfall, soil moisture, and weather patterns.

2. Crop Health Monitoring: The app used satellite imagery and computer vision to monitor crop health and detect signs of disease early.

3. Soil Health Recommendations: The system analyzed soil data and recommended specific treatments based on the soil composition & nutrient levels in each region.

4. Localized Advice: The app delivered insights in local languages to ensure farmers could access and understand the information easily.

   Implementation:

   • Data Collection: The project gathered historical climate, rainfall, soil, and weather data from government and private databases.

   • Machine Learning Models: The data was used to train machine learning models on Microsoft’s Azure cloud platform. These models could predict weather patterns, rainfall, and crop yield estimates.

   • Mobile App Deployment: The AI Sowing App was made available to farmers via smartphones, providing them with real-time insights and advice in local languages.

Impact: The AI Sowing App demonstrated a significant impact on agriculture in India:

1. Higher Crop Yields: Farmers using this AI system saw higher crop yields, meaning they harvested more food from the same land.

2. Reduced Crop Losses: Knowing when rain or pests were coming helped farmers protect their crops in advance.

3. Wider Reach: The AI system helped small-scale farmers in rural areas who often lack access to agricultural experts.

   Challenges and Learnings:

   • Internet and Connectivity: In remote villages, internet access is limited, which can make it harder for farmers to access the app.

   • Building Trust: Some farmers were initially hesitant to trust AI-based advice, as they were used to traditional farming methods.

   • Language and Cultural Barriers: Ensuring the app was user-friendly and available in local languages made it more accessible, but cultural training on technology adoption was still necessary.

Since we have completed all the stages of the AI Project cycle, let us now see a case study on AI Project Cycle.

Project : Case Study on AI Project Cycle: Climate Change and Its Impact

1. Problem Scoping

   Objective: Problem related to climate change : rising temperatures, increased extreme weather events, impacts on local agriculture.

   4W Canvas:

   1. What?

      Definition of Climate Change: Climate change refers to significant and lasting changes in the Earth's climate, particularly an increase in temperature and changes in weather patterns over extended periods. While climate change can occur naturally, the current rapid changes are largely driven by human activities.

      Key Effects of Climate Change:

      • Rising Global Temperatures: Average global temperatures have increased by about 1.2°C since the late 19th century.

      • Melting Ice Caps and Glaciers: Polar ice sheets are losing mass, contributing to rising sea levels.

      • Increased Frequency of Extreme Weather Events: There has been a rise in the intensity and frequency of hurricanes, droughts, and floods.

      • Ocean Acidification: Increased CO2 levels are leading to higher acidity in oceans, affecting marine life.

   2. Why?

      Causes of Climate Change:

      • Greenhouse Gas Emissions: Human activities, particularly the burning of fossil fuels (coal, oil, natural gas), release significant amounts of greenhouse gases such as carbon dioxide (CO2) and methane (CH4).

      • Deforestation: Cutting down forests reduces the number of trees that can absorb CO2 from the atmosphere, leading to higher concentrations of greenhouse gases.

      • Industrial Activities: Factories and power plants emit large amounts of pollutants and greenhouse gases.

        Importance of Addressing Climate Change:

      • Environmental Protection: Safeguarding ecosystems and biodiversity is crucial for maintaining natural balance.

      • Public Health: Reducing pollution can lower the incidence of respiratory and cardiovascular diseases.

      • Economic Stability: Addressing climate change can prevent damage from extreme weather, which can lead to significant economic losses.

   3. Who?

      Stakeholders Involved in Climate Change:

      • Governments: National and local governments are responsible for creating and enforcing environmental policies and regulations.

      • Businesses: Corporations play a crucial role in emissions reduction by adopting sustainable practices and investing in clean technologies.

      • Scientists and Researchers: They study climate systems and develop solutions to mitigate climate change.

        Roles of People:

      • Governments: Implement climate policies, invest in renewable energy, and engage in international agreements (like the Paris Agreement).

      • Businesses: Transition to sustainable practices and reduce waste and emissions.

      • Individuals: Adopt eco-friendly habits, advocate for policy changes, and participate in community initiatives.

   4. Where?

      Global and Local Impacts of Climate Change:

      • Global Impacts: Climate change affects ecosystems, economies, and communities worldwide, resulting in phenomena like sea-level rise that threaten coastal cities and increased natural disasters that disrupt lives and livelihoods.

      • Local Impacts: Each community may experience different effects:

   Weather Changes: Communities may notice shifts in weather patterns, such as longer droughts or more intense storms.

   Impact on Agriculture: Farmers might experience crop failures or reduced yields due to changing rainfall patterns and extreme weather.

2. Data Acquisition

   Objective: Collect relevant data to understand the scope of the climate change issue being studied.

   Types of Data to Collect:

   • Historical Climate Data:

     Temperature, precipitation, and extreme weather events over the past decades.

     Sources: National Weather Service, NASA, NOAA.

   • Local Impact Data:

     Data on crop yields, health statistics, and biodiversity changes in your area.

     Sources: Local agricultural offices, health departments, and environmental organisations.

   • Public Opinion Data:

     Surveys or questionnaires assessing community awareness and perceptions of climate change.

     Methods of Data Collection:

   • Online databases and government resources.

   • Surveys distributed in the community.

   • Interviews with local experts and stakeholders.

     1. Data Exploration

        Objective: Analyse the collected data to uncover patterns, trends, and relationships related to climate change impacts.

        Activities:

        
        

   • ​

     
     

   • ​

     
     

   • ​

     
     

     Tools:

   • ​

Descriptive Analysis:

Calculate averages, trends, and changes over time for temperature and precipitation data.

Visualisations:

Create graphs and charts to illustrate trends in climate data, such as rising temperatures or increasing frequency of extreme weather.

Correlational Analysis:

Explorerelationships between climate variables (e.g., temperature and crop yields) to identify potential impacts.

Software like Excel, Google Sheets, or data visualisation tools such as Tableau or Python libraries (e.g., Matplotlib, Seaborn).

1. Modelling

   Objective: Develop predictive models to simulate the impact of climate change on specific areas of interest.

   Activities:

   • Selection of Modelling Techniques:

     Choose appropriate models based on the problem and data available. Common techniques include:

     • Regression analysis for predicting crop yields based on temperature changes.

     • Climate models to project future temperature and precipitation scenarios.

   • Model Implementation:

     Use statistical software (like R, Python) to implement the chosen models.

   • Scenario Analysis:

     Run different scenarios (e.g., varying levels of greenhouse gas emissions) to see potential future impacts.

     Outputs:

   • Predictive models showing expected changes in temperature, precipitation, and impacts on local agriculture and ecosystems.

2. Evaluation

   Objective: Assess the effectiveness of the models and findings.

   Activities:

   • Model Validation:

     Compare model predictions against actual data to determine accuracy.

   • Sensitivity Analysis:

     Evaluate how sensitive your model is to changes in input variables (e.g., how much temperature change affects crop yield).

   • People Feedback:

     Present findings to local people (farmers, policymakers) to gather their input and insights.

     Metrics for Evaluation:

   • Accuracy of predictions, stakeholder satisfaction, and applicability of findings to local conditions.

3. Deployment

   Objective: Implement the findings and recommendations derived from the project.

   Activities:

   • Communication of Results:

     Prepare reports, presentations, or infographics summarising the findings and implications of the study.

   • Actionable Recommendations:

     Suggest specific actions for people responsible based on model outputs, such as adopting climate-resilient crops or implementing water conservation strategies.

   • Community Engagement:

     Organise workshops or community meetings to discuss the findings and engage local residents in climate action initiatives.

   • Monitoring and Feedback:

Establish a system to monitor the impacts of implemented strategies over time and gather feedback for future projects.

Activity:

Implement the AI project cycle to develop an AI Model for Indian Healthcare – Detecting Eye Disease

In India, many people face eye health issues, especially in rural areas where access to doctors can be limited. Diabetic Retinopathy is an eye disease caused by diabetes that can lead to blindness if not treated early. However, diagnosing this disease requires a lot of time and effort from doctors, especially because they need to examine many images of patients’ eyes.

Activity Guidelines:

• Understand the problem.

• Learn the various aspects and developments in the field.

• Add the steps of the AI Project Cycle for the problem given above.

6.6: Ethical issues in AI

Ethics is a branch of philosophy which gives us a set of moral principles as to how humans and society should function. This specifically speaks about duties, obligations, morality, fairness and equity. Given the vast and almost revolutionary impact of AI across all aspects of human society, ethical and moral considerations play an important role in the shaping of AI policies.

There are key ethical issues in AI where active debate is on-going.

Self-driving cars are becoming more common. Imagine an AI-controlled car has to make a quick decision to avoid an accident. It might have to choose between hitting a pedestrian who suddenly crosses the road or swerving and risking a collision with another car.

• Ethical Question: Should AI make life-and-death decisions like this?

• Discussion Point: Who is responsible if an accident happens—the car company, the programmer, or no one?

  
  

  Bias and Discrimination:

  AI systems can sometimes be biassed if the original data they are trained on is itself biassed or skewed based on the data collection. This can lead to unfair discrimination of disadvantaged and marginalised populations.

  Example: Facial recognition AI has been found to be less accurate for people with darker skin tones. For example, some systems have a higher error rate in identifying Black or Asian individuals compared to lighter-skinned individuals. This happens because the AI was trained mostly on images of lighter-skinned faces, which leads to inaccuracies and unfair treatment.

• Impact: This bias can lead to misidentifications, which can be particularly serious when facial recognition is used by law enforcement. People could be falsely accused or stopped by police due to inaccurate identification.

  AI Bias can happen due to both bias inbuilt into data (biassed or incomplete data) and the algorithm (developer has induced bias into the model algorithm). Since AI is all pervasive across diverse segments like policing, employment, financial inclusion AI bias can lead to significant negative outcomes for society impacting fairness, equity, diversity and inclusion.

  AI bias can be addressed by:

• Representative and Diverse data

• Regular testing for bias detection

• Publicly shared algorithms

• Ethical policies for AI development

  Data Privacy Concerns:

• Since AI deals with huge volumes of data, often personally sensitive, it has to be ensured appropriate permissions are taken for rightful usage and the data stored securely to avoid malafide and fraudulent usage.

• Example: An AI application that analyses financial information of customers could be misused for fraudulent intent.

  Job Displacement:

  For many routine and standard jobs, AI could replace them since it is more cost effective and hence could lead to large scale unemployment.

  Example: Automated systems or BOTs in call centres/service centres might replace call centre workers, leading to job losses. Similarly industrial robots can impact job losses in the factory shop floor.

  Autonomous Decision-Making:

• AI systems can make decisions with minimal or no human involvement, which can be challenging if the decisions are wrong or harmful.

• Example: A self-driving car must make split-second decisions in emergencies, and those decisions could have serious consequences.

Activity: Ethical Decision-Making with AI

The Moral Machine is an online experiment created by researchers at MIT to explore how people from different cultural backgrounds will make ethical decisions involving autonomous vehicles (self-driving cars). It is designed as a large-scale interactive study where participants are presented with various moral dilemmas related to hypothetical accident scenarios and have to decide which group of people (or animals) should be saved or sacrificed.

Tool: Open www.startcoding.co.in/code Click on Objective:

This activity helps students explore how AI systems make moral decisions and the ethical dilemmas involved, particularly in situations like autonomous vehicles making life-and-death choices.

Steps

:

1. Interactive Activity:

   • Visit the Moral Machine tool and engage with the scenarios presented.

   • In each scenario, decide which action the self-driving car should take in life-threatening situations (e.g., whether to save passengers or pedestrians).

2. Discussion:

   • After completing the scenarios, students will participate in a group discussion.

   • Questions to consider:

     • How did they make their decisions?

     • What factors influenced their choices (e.g., age, number of people)?

     • Do they think AI should be allowed to make such decisions?

       
       

3. Reflection Project:

   • Write an essay on the ethical challenges of AI decision-making, including:

   • An example scenario from Moral Machine.

   • Their thoughts on whether machines can be trusted to make moral decisions.

   • Possible solutions to ensure fairness in AI decision-making systems.

Classroom Activity : Balloon Debate

The Balloon Debate is a thought experiment and discussion exercise often used in educational settings to explore themes of ethics, decision-making, and value judgments. It typically involves a hypothetical scenario where participants must make difficult choices about who to save in a life-or- death situation.

The participants are asked to imagine that they are in a hot air balloon that is losing altitude and may crash. To lighten the load and ensure the survival of the balloon, the group must decide which member of the party to throw overboard. The participants are usually given specific roles or identities, each with their own background and justification for why they should be kept in the balloon.

Common roles may include:

• A scientist

• A doctor

• A politician

• An artist

• A child

• An elderly person

  Each participant must present arguments for why they should not be thrown overboard. This involves persuading the group based on their role’s perceived value to society, their potential contributions, or emotional appeals.

  
  

  After all participants have made their case, the group must come to a consensus on who will be sacrificed to save the others. This process encourages discussion about moral and ethical considerations, such as:

• Value of human life

• Contribution to society

• Emotional connections

• Random selection vs. merit-based decisions

  Learning Outcomes: The balloon debate serves several educational purposes:

• It fosters critical thinking and ethical reasoning.

It encourages participants to consider different perspectives & values.

It highlights the complexities of moral decision-making in crisis situations.

It can reveal biases or assumptions about worth and value in society.

Chapter 6: Exercise

1. What are 3 types of ML techniques? Tick all that apply.

   1. Supervised Learning

   2. Unsupervised Learning

   3. Random Learning

   4. Reinforcement Learning

2. Which of the following are NLP applications. Tick all that apply

   1. Chatbots

   2. IVR systems

   3. AI generated paintings

   4. Language translation tools

3. Robotics combines which 2 disciplines. Tick all that apply

   1. Computer Science

   2. Medicine

   3. Civil Eng

   4. Mechanical Eng

4. Which is the granular breakdown of images that Computer Vision does?

   1. Pixar

   2. Pixel

   3. Picture

   4. Atoms

5. Computer Vision isused in the following areas.Tick all that apply

   1. Face Recognition

   2. Language translation

   3. Medical imaging

   4. Object detection

6. Tokenization during NLP achieves the following:

   1. Splitting words into letters

   2. Expanding letters into words

   3. Splitting texts into words and sentences

   4. Creating new words

7. Which of the following is not part of NLP steps:

   1. Text preprocessing

   2. Model selection and training

   3. Robotics

   4. Feature extraction

8. Which of the following is not a ML model

   1. Decision Tree

   2. Support Vector Machines

   3. Linear Regression

   4. Anomaly detection

9. Which of the following is not a tool for AI evaluation

   1. Accuracy

   2. Precision

   3. Confidence interval

   4. F1 score

10. KPIs defined during problem scoping should have the following attributes

    1. Measurable

    2. Objective

    3. Random

    4. Variable

11. Data Cleansing is done if the Data quality is poor. True or False 12. Data categories can include. Tick all that apply

    1. Text

    2. Audio

    3. Video

    4. Paint

1. Data can be stored in the following. Tick all that apply

   1. Relational Database

   2. File systems

   3. Cloud servers

   4. Algorithms

2. Which of the following are Data Visualization Tools? Tick all that apply

   1. Charts

   2. Graphs

   3. Tableau

   4. Document

Questions from papers

1. An application lets you search what you see, get things done faster and understand the world around you – using just your camera or a photo. Which domain does this app belong to?

   1. Natural Language Processing

   2. Data Sciences

   3. Computer Vision

   4. Artificial Language Processing

2.                helps us to summarise all the key points into one single Template so that in future, whenever there is a need to look back at the basis of the problem, we can take a look at this and understand the key elements of it.

   1. 4W Problem canvas

   2. Problem Statement Template

   3. Data Acquisition

   4. Algorithm

      
      

3. The 4Ws of the 4W problem canvas are:

   
   

   1. Who, What, Where and Why

   2. Who, What, When and Why

   3. Who, What, Where and When

   4. Who, Where, When and Why

      
      

4. Which of the following is incorrect?

   1. In a rule based approach, the relationship or patterns in data are defined by the developer.

   2. Decision tree looks like an upside-down tree.

   3. Pixel It activity is an example of how computers see images, process them and classify them.

   4. In a learning based approach, the relationship or patterns in data are defined by the developer.

5. Which of the following is an example of rule based approach?

   1. Pixel it activity

   2. Decision trees

   3. Histogram

   4. Illustration diagram

      
      

6. Which of the following is not valid for Data Acquisition?

   1. Web scraping

   2. Surveys

   3. Sensors

   4. Announcements

      
      

7. Which of the following comes under Problem Scoping?

   1. System Mapping

   2. 4Ws Canvas

   3. Data Features

   4. Web scraping

      
      

8. The primary purpose of data exploration in AI project cycle is

   
   

   1. To make data more complicated

   2. To simplify complex data

   3. To discover patterns and insights in data

   4. To visualise data

   
   

9. Artificial Intelligence and machine learning systems can display unfair behaviour if not trained properly.

(True/False)

Project for students : AI Project Cycle

Impact of Unsegregated waste and how AI can help solve this issue :

1. Write the Problem scoping

2. Who are getting affected directly or indirectly due to it.

3. Determine the nature of the problem. What is the problem and how do you know that it is a problem?

4. What is the context in which the problem arises, and the locations where it is prominent.?

5. Why will this solution be of value to the people affected by it?