Kann ein Algorithmus fair sein? Technische und rechtliche Perspektiven [Can an Algorithm Be Fair? Technical and Legal Perspectives]

English title: Can an Algorithm Be Fair? Technical and Legal Perspectives.

Article written in German, published in ailex, no. 3 (2026), pp. 123-130, by MANZ.

Read the article on the publisher’s website.

TL;DR

We bridge algorithmic fairness and EU non-discrimination law in one paper, providing:

• Mapping causal path-specific effects onto direct and indirect discrimination, while being explicit about the limits of that analogy.
• Guidance on choosing a fairness metric based on the decision context and the harm it should capture, treating metrics as evidence rather than legal rules.
• A section on how to assess fairness in practice: sampling and statistical power for small groups, intersectional subgroups, long-term effects, and how metrics can be gamed.
• An account of why mitigation needs protected attributes, tied to positive action and the AI Act’s special-category data rules, including the Digital Omnibus.
• An up-to-date review of how EU law handles algorithmic discrimination in high-risk AI systems, general-purpose AI, and online platforms (AI Act, DSA, GDPR, Digital Omnibus, CEN-CENELEC standard), including where the gaps are.

The takeaway: choosing what to equalize is a societal and legal decision, not a technical one.