A. Introduction to AI Project Cycle
Artificial Intelligence (AI) is the ability of machines to perform tasks that usually require human intelligence, such as learning, decision-making, and problem-solving.
To create any AI solution, we follow a proper method known as the AI Project Cycle. It consists of five important stages:
B. Five Stages of the AI Project Cycle
1. Problem Scoping
Meaning: Identifying and understanding the real-world problem that we want to solve using AI.
Steps in Problem Scoping:
- Understand the problem.
- Identify the goal.
- Know the stakeholders (people who are affected by the problem).
- Define constraints (limitations like budget, time, tools).
SDGs (Sustainable Development Goals): While identifying problems, we can connect them to global goals like Zero Hunger, Clean Water, Quality Education, etc.
Example: In a school canteen, students waste a lot of food. We want to reduce this food wastage.
2. Data Acquisition
Meaning: Gathering the correct data that is related to the problem.
Sources of Data:
- Surveys
- Internet
- Sensors
- School records
- Social media
Types of Data:
- Structured data: Tables, Excel sheets, rows and columns (e.g., student marks)
- Unstructured data: Images, videos, audio, free text (e.g., photos, YouTube videos)
Example: Collect data of daily food cooked and wasted in the canteen over the last 30 days.
3. Data Exploration
Meaning: Understanding and preparing the data for training.
Main Tasks:
- Cleaning the data: Removing errors or missing values.
- Visualizing the data: Using graphs and charts to understand patterns.
- Identifying features and labels:
- Feature: Input value (e.g., day of the week)
- Label: Output value (e.g., amount of food wasted)
Example: Check which days food is wasted more and why. You may find that Mondays have more wastage than Fridays.
4. Modelling
Meaning: Using machine learning algorithms to train a model on the data so it can make predictions.
Steps in Modeling:
- Choose the type of model:
- Supervised learning (when data has labels)
- Unsupervised learning (when data has no labels)
- Feed the training data into the model.
- Train the model to learn from the data.
Example: Train the model to predict food requirement based on previous records of meals.
5. Evaluation
Meaning: Checking whether the AI model is giving correct and useful results.
Methods of Evaluation:
- Accuracy score
- Confusion matrix
- Precision and Recall
- Cross-validation
Example: Test if the food prediction model is actually reducing food wastage. If it works correctly in most cases, the model is successful.
C. Summary of the AI Project Cycle
| Stage | Description |
|---|---|
| Problem Scoping | Identify the real-world problem |
| Data Acquisition | Collect data related to the problem |
| Data Exploration | Understand and clean the data |
| Modelling | Train the AI model using data |
| Evaluation | Test the model and measure its performance |
D. Training and Testing Data
To train and test the model, we divide data into two parts:
- Training Data – Used to teach the model.
- Testing Data – Used to check how well the model performs.
Example: Out of 100 records, 80 can be used for training and 20 for testing.
E. Introduction to AI Ethics
AI must be used responsibly. It should not cause harm, cheat people, or invade privacy.
F. Principles of Ethical AI
- Bias – AI should not favor any group or individual unfairly.
- Fairness – Everyone should get equal and just treatment.
- Transparency – People should know how decisions are made by AI.
- Accountability – Humans should be responsible for what AI does.
- Privacy – Personal data should be protected and not misused.
G. Examples of Ethical Issues in AI
- A recruitment AI that always selects men is biased and unfair.
- A face-recognition app that uses people’s photos without permission violates privacy.
- A chatbot spreading fake news causes misinformation and harm.
H. Responsible Use of AI
Before creating or using any AI system, we must ask:
- Is it fair?
- Is it safe?
- Is it private?
- Can I explain how it works?
- Am I using the data legally?
I. Case Study Activity for Students (Suggested)
Situation: An AI system used by a school to award scholarships starts rejecting female students more than males.
Ask Students:
- Is the AI model biased?
- Which ethics principles are being violated?
- How would you correct the AI system?
J. Keywords to Remember
| Term | Meaning |
|---|---|
| Feature | Input variable used to train the model |
| Label | Output value we want the model to predict |
| Dataset | Collection of data used for AI |
| Model | The machine learning program trained on data |
| Evaluation | Measuring how well the model is working |
| Ethics | Rules about right and wrong behavior |
| Bias | Favoring or harming a group unfairly |