Page 61 - CT_AI_Class-8

DATA AND

3 FAIRNESS IN AI

PRIMARY PREVIEW

Role of Data Understanding Bias
Ensuring Fairness

A group of artists in Seoul, Korea, collected photographs of clouds that looked a bit like human
faces. They decided to feed these pictures into an AI face-detection program. To their surprise,
the AI identified all the cloud images as faces. As humans, we sometimes imagine faces in the
clouds too, but this project highlighted an issue with the AI system's training data. The AI’s face
recognition system was trained using more images of light-skinned men, which caused it to

incorrectly identify the cloud images as faces. This shows how AI can make mistakes if the data it
learns from is not balanced or diverse.

The example makes us realise that:
 AI looks for patterns it has learned from its
training data.

 The training data influences what AI detects,
which means if the data is unbalanced, the
AI's predictions can be biased or incorrect.

 Given the widespread use of AI across many
fields, it is essential to carefully examine and
ensure the balance and diversity in the training
data to avoid errors and ensure fairness in AI systems.

ROLE OF DATA

Imagine training an AI program to recognise different kinds of balls, like a football and basketball
from images. The AI will be provided with many pictures of each type of ball to help it learn how
to identify them. It studies these images carefully, looking for patterns such as the round shape
and texture of a football and the larger size and bounciness of a basketball. Over time, the AI
learns to distinguish between the different balls based on these patterns. After learning from

these examples, the AI can then try to correctly identify new images of balls it hasn’t seen before.

Data and Fairness in AI 59