The growing use of artificial intelligence (AI) in sensitive areas such as law enforcement, education, justice administration, and migration has raised concerns about algorithmic bias and discrimination. The EU adopted the Artificial Intelligence Act (AI Act), the first legal instrument to comprehensively regulate AI. The AI Act subjects high-risk AI systems to a distinct range of obligations under a risk-based framework.
This thesis analyzes the extent to which the AI Act effectively addresses algorithmic bias and discrimination in high-risk AI systems. First, it discusses the concepts of AI, algorithms, algorithmic bias, and algorithmic discrimination. Then, it provides a descriptive and critical analysis of the provisions most central to bias, namely Art. 9, 10, 14, 15, and 86. Finally, it assesses the Act’s overall approach to discrimination.
This thesis argues that the Act is best understood as a product-safety regulation designed to address bias, a concept rooted in computer science, through the technical obligations set out in the Act. However, the Act is ill-suited to regulate and prevent algorithmic discrimination as a legal, normative concept. Recurring ambiguities in the wording as well as a regulatory gap concerning emergent bias further leave its effectiveness undetermined. Therefore, the Act plays a supporting role in seeking to minimize the risk of algorithmic discrimination by regulating algorithmic bias, whereas the legal assessment of discrimination remains with the existing EU non-discrimination framework.