Skip to main content

Bias and Fairness in AI

Switch to German

Bias & Fairness

in algorithmic decision systems

IN algorithmic decision systems

Please note that this learning module has been designed as a desktop version.

For an optimal learning experience, we suggest completing it on a laptop or PC.

 Gain a deep, application-based understanding of bias & fairness in artificial intelligence (AI) systems using this site.


For the best learning experience:
Set the page zoom to 50%.

01

basics

Learn the basics about the concepts of “bias” and “fairness” in AI systems.

02

online courses

Build on the basics with practical exercises.

03

teaching case

Apply your new skills to a real teaching case.

Welcome to A
free learning experience

Do you want to understand the impact that the bias of an algorithmic decision can have on business and society?

Do you want to know what types of bias there are and what algorithmic fairness really means?

Do you want to understand how to explore, detect and mitigate bias and fairness in AI systems?

Then you've come to the right place!

Then you’ve come to the right place!

 

Why is it important to take a closer look at bias & fairness in AI systems?

Bias in algorithmic decisions already affects humans to a large extent!
Click through the slideshow to find out more!
When Artificial Intelligence (AI) hits the headlines
it is all too often because of problems with bias and resulting unfair, discriminatory decisions. Click on the arrow to the right to read some of the more notorious headlines from recent years.
Amazon: Job application robot discriminates against women
In 2014, Amazon developed an algorithm to rank incoming job applications. In 2018, it came to light that the application robot was discriminating against women. The algorithm was trained on data from hired applicants and was supposed to learn which characteristics Amazon preferred. In recent years, however, it has mainly hired men.
Read more about this headline.
Facial analysis software biased by gender and skin type
The results of a study show that some well-known AI facial recognition systems are much worse at determining a person's gender when they are women, especially darker-skinned women. The study shows an error rate of 0.8 per cent for light-skinned men and 34.7 per cent for dark-skinned women.
Read more about this headline.
US justice: algorithms systematically disadvantage African-Americans
In the US, a software called COMPAS has been used for years to calculate defendants' risk of reoffending based on various factors. It turns out that COMPAS makes different mistakes depending on ethnicity. African-Americans, for example, were more likely to be overestimated as being at risk of reoffending.
Read more about this headline.
Ad discrimination on Facebook
Meta enabled and encouraged advertisers to segment their housing ads based on race, color, religion, sex, disability, familial status, and national origin, thereby controlling which Facebook users received housing ads and which did not.
Read more about this headline.
The business and social impact of AI systems can be huge!
If left unchecked, AI systems could harm us and our society. AI systems must therefore be designed, developed and monitored with fairness in mind.

We want to explain in detail how it is possible for an AI system to make unfair decisions.
Getting to the basics

Why is it important to take a closer look at bias & fairness in AI systems?

Bias in algorithmic decisions already affects humans to a large extent!
Click through the slideshow to find out more!
When Artificial Intelligence (AI) hits the headlines
it is all too often because of problems with bias and resulting unfair, discriminatory decisions. Click on the arrow to the right to read some of the more notorious headlines from recent years.
Amazon: Job application robot discriminates against women
In 2014, Amazon developed an algorithm to rank incoming job applications. In 2018, it came to light that the application robot was discriminating against women. The algorithm was trained on data from hired applicants and was supposed to learn which characteristics Amazon preferred. In recent years, however, it has mainly hired men.
Read more about this headline.
Facial analysis software biased by gender and skin type
The results of a study show that some well-known AI facial recognition systems are much worse at determining a person's gender when they are women, especially darker-skinned women. The study shows an error rate of 0.8 per cent for light-skinned men and 34.7 per cent for dark-skinned women.
Read more about this headline.
US justice: algorithms systematically disadvantage African-Americans
In the US, software called COMPAS has been used for years to calculate defendants' risk of reoffending based on various factors. But it turns out that COMPAS makes different mistakes depending on ethnicity. African-Americans, for example, were more likely to be overestimated as being at risk of reoffending.
Read more about this headline.
The business and social impact of AI systems can be huge!
If left unchecked, AI systems could harm us and our society. AI systems must therefore be designed, developed and monitored with fairness in mind.

We want to explain in detail how it is possible for an AI system to make unfair decisions.
Getting to the basics

Do you have any questions or comments?
We welcome your feedback.

It is only with your feedback that we will be able to continue to improve our learning units in a targeted way.

Our training combines theoretical knowledge transfer with practical exercises.

The duration of the session, including practical exercises,
is approximately 90 minutes.
Let's go

Our training combines theoretical knowledge transfer with practical exercises. 

The duration of the session, including practical exercises, is approximately 90 minutes.

Let's go