How unfair is AI? Prof. Dr. Staab, co-speaker of IRIS, gave a lecture on questions of fairness at the Ludwigsburg Art Association.

July 29, 2024

How unfair is AI?

Prof. Dr. Staab explored this question last Friday evening in his lecture at the Ludwigsburg Art Association. The core of his lecture was about decisions that affect people, be it access to a degree, a loan, or a job. Both people and AI can make unfair decisions. If an AI learns from data based on unfair decisions made by people, the AI ​​will imitate this unfair behavior. Using numerous examples, Prof. Dr. Staab explained three fairness concepts that deal with the questions of equality and justice with different focuses and applications:

Individual fairness: This concept ensures that similar cases are treated consistently and equally. This is primarily about the fair treatment of individuals in comparable situations to avoid discrimination. The core idea is that people who only differ in protected characteristics (e.g., skin color, religion, origin) should also be treated equally. Unfortunately, this concept is often not applicable because you cannot always compare people, or a decision can sometimes only affect one person.

Libertarian fairness: In this libertarian version of fairness, statistics are compiled. Only the achievements of people are to be assessed. People's starting opportunities play no role. The guiding principle is the equal treatment of people.
Egalitarian fairness: This concept aims at an equal distribution of opportunities among all those involved. It is based on the idea that all people should have equal opportunities. Egalitarian fairness can be achieved through redistribution measures to compensate for inequalities due to unequal starting conditions.

Prof. Dr. Staab concluded that fairness must be discussed repeatedly because the perceptions of what is fair have changed over the decades. Big data makes it possible to measure the different types of fairness. This can be used to prove discrimination - regardless of whether discriminatory decisions were made by people or by an AI. In principle, a human can have a much more extensive understanding of an individual situation than an AI and use this understanding to make fair decisions. However, this fundamental possibility is often not realized in practical circumstances, e.g., in a government agency or a company.

Understanding fairness is an ongoing task for society as a whole.

The evening ended with a question and answer session with visitors and the opportunity to tour the art association's current exhibition.

Click here to watch the entire lecture.

Link to the original article

To the top of the page