Page 78 - CT_AI_Class-8

BIAS AND FAIRNESS IN AI

Artificial Intelligence systems learn from data. The quality and nature of this data play a very
important role in how AI systems behave. If the data used for training is incomplete, unbalanced
or reflects existing inequalities, the AI system may also learn these unfair patterns. This problem
is known as bias in AI. Bias occurs when an AI system produces results that are unfair, unequal
or prejudiced toward certain individuals or groups.

Bias in AI is often unintentional. It usually happens because the data used to train the system
does not represent all groups equally. Since AI systems rely on patterns found in data, they may
repeat or even strengthen existing social biases. This can lead to unfair outcomes that may affect
people's opportunities, safety and daily lives.

Let us consider a simple example. Suppose a company uses an AI system to shortlist candidates
for a job. The system is trained using past hiring data. If, in the past, most employees hired were
from a particular background or gender, the AI system may learn to favour similar candidates. As
a result, equally qualified candidates from other backgrounds may be rejected unfairly.

Another example can be seen in facial recognition systems. If a system is trained mostly with
images of people from one region, it may not correctly recognise people from other regions. This
happens because the system has not learned from diverse data. As a result, the system performs

well for some groups but poorly for others.

Effects of Bias in AI

An unfair or biased AI system may create several problems, such as:
 Reject deserving candidates during hiring or college admissions

 Spread stereotypes about certain groups of people
 Ignore minority groups or underrepresented communities

 Make incorrect predictions or decisions
 Provide unequal services to different users

Such outcomes can negatively affect individuals and may also reduce trust in AI systems.

Reducing Bias in AI Systems

Developers and organisations take several important steps to reduce bias and ensure fairness in
AI systems, which are as follows:

 Use diverse and representative data: Training AI systems with data from different regions,
genders, age groups, languages and backgrounds helps create balanced models. When the
data represents a wide variety of people and situations, the system learns to make fair and
accurate decisions.

For example, if a speech recognition system is trained using voices of people from different age
groups, accents and languages, it will better understand a wider range of users. Otherwise, it
may only work well for a limited group.

76 Artificial Intelligence (CT & AI)-VIII