Class 8 AI | Unit 3: Data and Fairness in AI
Is AI Always Fair? ⚖️
We often think computers are "neutral," but because AI learns from Human Data, it can inherit our human mistakes and prejudices. In this chapter, we learn how to keep AI fair for everyone.
1. What is Data Bias? 🔍
Bias happens when the data used to train an AI is not balanced. If the data is "skewed" (one-sided), the AI will make unfair decisions.
Example: The Photo App
Imagine an AI trained to recognize "Beautiful Landscapes." If the AI is only shown photos of snowy mountains from Europe, it might not recognize the beauty of the Deccan Plateau or the Thar Desert. This is a bias in data representation!
2. Promoting Fairness in AI
To promote fairness, developers must ensure that the AI treats every individual or group equally, regardless of their race, gender, or background.
- Diverse Data: Collect data from all types of people and regions.
- Testing: Test the AI with "edge cases" (situations that are not common) to see if it fails.
3. What is Inclusivity? 🤝
Inclusivity means designing AI so that everyone can use it, including people with different abilities.
Example: A voice assistant that can understand many different Indian accents (Telugu, Hindi, Bengali) is Inclusive. If it only understands "Perfect English," it excludes millions of people.
🌟 The "Fairness Test" Activity
Let's try to understand how bias creeps in with a simple mental activity:
- Imagine you are building an AI to pick the "Best Student."
- If you only give the AI data about **Marks**, it will ignore students who are great at Sports or Art.
- The Solution: To be fair, you must give the AI data about sports, leadership, and behavior too!
Want to see more? Follow the link to the official handbook for case studies.
Download Class 8 AI Handbook
Comments
Post a Comment