Non-Discrimination and Fairness in Artificial Intelligence

Sumit Tak Assistant Professor

DES Shri. Navalmal Firodia Law College, Pune

Slide Title

• What is discrimination?
• How can discrimination occur in the AI era?
• What are non-discrimination and fairness?

Discrimination (The Problem)

• Discrimination means treating someone unfairly because of certain characteristics such as race, gender, age, religion, or disability.
• It happens when a person or group is unfairly disadvantaged

compared to others.

• Discrimination can

be intentional (done on

because of existing

purpose) or systems or

unintentional (happens

practices).

• In AI:

• Discrimination occurs when an AI system gives worse results to a particular group of people. Example: If a resume-screening AI system rejects applications from older candidates, it is discriminating, even if the programmer did not intend it.

Non-Discrimination (The Basic Principle)

• Non-discrimination means not treating people unfairly based on

their personal characteristics.

• It is an important legal and ethical principle followed by many

organizations and governments.

• How it works:
• The idea is to avoid using sensitive information such as race,

gender, or religion while making decisions.

• In AI:
• Developers may remove variables like race or gender from

the data used to train AI systems.

• However, AI may still find indirect indicators (such as location, education, or spending habits) that relate to those characteristics.
• This means discrimination can still happen indirectly.

Fairness (The Goal)

• Fairness means making sure that outcomes are just, balanced, and equal for everyone.
• It goes beyond simply avoiding discrimination.
• How it works:
• Non-discrimination focuses on treating everyone the same (equality).
• Fairness focuses on ensuring equal opportunities for everyone (equity).
• In AI:
• Developers must carefully design and test AI systems to make sure their

results are fair for all groups.

• Example: An AI loan system should approve loans for men and women

at similar rates if they have similar qualifications.

• This helps correct unfair patterns that may exist in historical data.

Introduction

• Artificial Intelligence (AI) is now used in many areas like jobs, banking, healthcare, and the justice system, and it can affect important decisions in people’s lives.
• Many people think AI is completely objective and fair because it uses data and mathematical calculations.
• However, AI learns from historical data, which may already contain social biases and inequalities.
• If biased data is used, AI systems may repeat or even increase discrimination.
• This can turn discrimination into a hidden and large-

scale problem.

• Therefore, ensuring fairness and non-discrimination in

AI is important to protect human rights and equality

The Roots of the Problem: Types of AI Bias

Historical Bias

• Historical bias happens when AI learns from past data that already contains social inequalities or discrimination.
• The AI simply repeats what happened in the

past.

• Example: If an AI hiring system is trained on old company data where mostly men were hired, the system may start preferring male candidates over equally qualified women.

The Roots of the Problem: Types of AI Bias

Representation (Sampling) Bias

• Representation bias occurs when the data used to train AI does not represent all groups in society.
• Some groups may be underrepresented or

missing in the data.

• Example: If a facial recognition system is trained mostly on photos of lighter-skinned people, it may make more mistakes when identifying darker-skinned individuals.

The Roots of the Problem: Types of AI Bias

Measurement Bias

• Measurement bias happens when the data used to measure a problem does not accurately represent the real situation.
• The AI then makes decisions based on

incomplete or misleading data.

• Example: If a policing AI uses past arrest records instead of actual crime rates, it may send more police to neighborhoods that were already heavily policed, even if crime is not higher there.

The Roots of the Problem: Types of AI Bias

Algorithmic Bias

• Algorithmic bias occurs when the design of the algorithm itself creates unfair results.
• Even if the data is correct, the way the system processes the data may produce biased outcomes.
• Example: If a loan approval AI is designed to prioritize high income and long credit history, it may reject many young or low-income applicants even if they are capable of repaying the loan.

Strategies to Reduce Bias in AI

Using Diverse and Balanced Data

• The data used to train AI should represent different groups of people in society.
• Developers should check the data carefully to ensure it is inclusive and balanced.

Strategies to Reduce Bias in AI

Building Diverse Teams

• AI systems should be developed by teams with people from different backgrounds, experiences, and expertise.
• This helps identify possible problems or

biases early in the design stage.

Strategies to Reduce Bias in AI

Human-in-the-Loop Approach

• Humans should remain involved in

important AI decisions.

• Experts can review AI recommendations, especially in sensitive areas like healthcare, hiring, and criminal justice, to prevent unfair outcomes.

Strategies to Reduce Bias in AI

Following Rules and Ethical Guidelines

• Organizations should follow laws and

ethical guidelines for AI.

• These rules promote transparency, accountability, and regular checks to ensure AI systems are fair and safe.

Conclusion

1. Artificial Intelligence is becoming part of everyday life,

so it must be fair and non-discriminatory.

2. Algorithmic bias is not only a technical issue but also a

matter of protecting fundamental rights.

3. AI systems should be designed in a way that promotes

equality and fairness.

4. Developers, policymakers, and legal experts must work

together to apply constitutional values in digital systems.

5. Strong regulations and human oversight are necessary to

prevent discrimination in AI.

6. The aim is to ensure that future technologies do not

repeat past social injustices.