Search⌘ K
AI Features
Home
Courses
Fundamentals of AI Fairness

Formal Definition and Fairness in Regulations

Explore formal definitions of AI fairness and the importance of regulatory guidelines such as the European GDPR and the US AI Bill of Rights. Understand how laws and principles guide the prevention of bias and discrimination in AI systems, highlighting the need for fair design, assessment, and ongoing monitoring to ensure equitable outcomes.

We'll cover the following...

Definition

Before proceeding with specific examples and techniques, we need to specify fairness more formally. We already know the intuitive definition; however, according to the Cambridge Dictionary, fairness is:

The quality of treating people equally or in a way that is right or reasonable

According to Mehrabi et al. (2021)Mehrabi, N., Morstatter, F., Saxena, N., Lerman, K., & Galstyan, A. (2021). A survey on bias and fairness in machine learning. ACM Comput. Surv., 54 (6):

The absence of any prejudice or favoritism toward an individual or a group based on their inherent or acquired characteristics.

So, things are simple: Just treat everyone equally, and it will be okay. However, important details can be often overlooked. Understanding terms equally can be very different for different people. We will see examples of this later in the course.

Example performance standards for army candidates
Example performance standards for army candidates
1.

Take a look at the table. It shows a part of the requirements for army candidates split for males and females. The numbers are not equal. Should they be? Is it fair or not? Think about it.

Show Answer
Did you find this helpful?

For the purpose of fair machine learning, we will be most interested in metrics that help us evaluate them. However, it is very important to keep in mind that the system is built for humans, so we can’t blindly follow the numbers.

Regulations

Some countries have developed specific laws to prevent algorithmic bias. In the European Union, the General Data Protection Regulation addresses the topic:

the [sic] controller should use appropriate mathematical or statistical procedures … that prevents, inter alia, discriminatory effects on natural persons on the basis of racial or ethnic origin, political opinion, religion or beliefs, trade union membership, genetic or health status or sexual orientation, or that result in measures having such an effect.

In the United States, there is no federal act regulating this topic. However, there are some guidelines and recommendations. An example is the AI Bill of Rights, which is a guide for designing and using AI systems. One of the topics covered is “Algorithmic Discrimination Protections.” Example rules from the document are:

  • Proactive assessment of fairness during the design

  • Representative and robust data

  • Constant monitoring and mitigation

  • Ensuring accessibility for people with disabilities

It also provides various examples of use cases that can be affected by fairness issues.

We need to keep in mind that regulators actively work, and our work might be impacted by them.