Kann ein Algorithmus fair sein? Technische und rechtliche Perspektiven

Algorithmic fairness has become a central concern in the development and regulation of artificial intelligence, yet the concept of fairness differs substantially across technical and legal perspectives. This work bridges algorithmic fairness research and European non-discrimination law, examining how technical fairness concepts relate to legal notions of direct and indirect discrimination. We discuss how fairness metrics should be selected according to the decision context and the specific harms they seek to capture, emphasizing that such metrics provide evidence rather than legal definitions of fairness. We further examine practical challenges in evaluating algorithmic fairness, including statistical power for small and intersectional groups, long-term effects, and the possibility of gaming fairness metrics. Finally, we analyse the role of protected attributes in fairness mitigation and review the treatment of algorithmic discrimination under the EU AI Act, Digital Services Act, GDPR, the Digital Omnibus, and emerging European standards. We argue that determining what should be equalized across individuals or groups is ultimately a societal and legal choice rather than a purely technical one.