> For the complete documentation index, see [llms.txt](https://jules-chancel.gitbook.io/discrimination-in-algorithmic-decision-making/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://jules-chancel.gitbook.io/discrimination-in-algorithmic-decision-making/abstract.md).

# Abstract

*The present paper is a copy of the master thesis I wrote as an LL.M. student. For further references, please consult the pdf version 👇*

{% file src="/files/NiELHNVq0YaJyoopf7VD" %}

In the last decade, algorithmic decision-making has improved fairness in health care, access to credit and employment. From ordering a meal to applying for a loan, our everyday life is already filled with automated decisions. The fact that algorithms are based on the rules of mathematics has long driven the impression that algorithms were neutral. However, a growing number of reports point out discrimination bias in widely used algorithms. Health market, employment, justice system, advertising: no economic field is exempt from a report highlighting algorithmic bias.

Because they are closely related, the present paper analyses both technical and legal challenges posed by algorithmic bias. From a technical point of view, this paper argues that using sensitive data in artificial intelligence training datasets inevitably leads to discrimination. Furthermore, at the age of big data, removing sensitive data from training datasets is not sufficient to avoid proxy discrimination. From a legal point a view, this paper argues that the directives prohibiting discrimination in the European Union coupled with the data protection rights are insufficient to tackle discrimination in solely automated decision making. A clear technical compliance framework regarding the production of algorithms is needed.