Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Implications of the AI Act for Non-Discrimination Law and Algorithmic Fairness

Luca Deck, Jan-Laurin Müller, Conradin Braun, Domenique Zipperling, Niklas Kühl

TL;DR

The paper addresses the gap between post-hoc non-discrimination law and design-time algorithmic fairness within the EU context. It analyzes how the EU AI Act could bridge these approaches by enforcing fairness considerations at the model development stage and translating legal concepts into technical requirements, while also highlighting enforcement and compliance challenges. Three contributions are offered: (1) mapping the misalignment between legal and technical fairness discourses, (2) arguing that the AI Act can serve as a bridge, and (3) sketching practical implications for bias detection and correction and the challenges regulators and developers will face. The work provides a path toward harmonizing legal and technical fairness, emphasizing the need for cross-disciplinary guidance to specify and verify concrete requirements for high-risk AI systems.

Abstract

The topic of fairness in AI, as debated in the FATE (Fairness, Accountability, Transparency, and Ethics in AI) communities, has sparked meaningful discussions in the past years. However, from a legal perspective, particularly from the perspective of European Union law, many open questions remain. Whereas algorithmic fairness aims to mitigate structural inequalities at design-level, European non-discrimination law is tailored to individual cases of discrimination after an AI model has been deployed. The AI Act might present a tremendous step towards bridging these two approaches by shifting non-discrimination responsibilities into the design stage of AI models. Based on an integrative reading of the AI Act, we comment on legal as well as technical enforcement problems and propose practical implications on bias detection and bias correction in order to specify and comply with specific technical requirements.

Implications of the AI Act for Non-Discrimination Law and Algorithmic Fairness

TL;DR

The paper addresses the gap between post-hoc non-discrimination law and design-time algorithmic fairness within the EU context. It analyzes how the EU AI Act could bridge these approaches by enforcing fairness considerations at the model development stage and translating legal concepts into technical requirements, while also highlighting enforcement and compliance challenges. Three contributions are offered: (1) mapping the misalignment between legal and technical fairness discourses, (2) arguing that the AI Act can serve as a bridge, and (3) sketching practical implications for bias detection and correction and the challenges regulators and developers will face. The work provides a path toward harmonizing legal and technical fairness, emphasizing the need for cross-disciplinary guidance to specify and verify concrete requirements for high-risk AI systems.

Abstract

The topic of fairness in AI, as debated in the FATE (Fairness, Accountability, Transparency, and Ethics in AI) communities, has sparked meaningful discussions in the past years. However, from a legal perspective, particularly from the perspective of European Union law, many open questions remain. Whereas algorithmic fairness aims to mitigate structural inequalities at design-level, European non-discrimination law is tailored to individual cases of discrimination after an AI model has been deployed. The AI Act might present a tremendous step towards bridging these two approaches by shifting non-discrimination responsibilities into the design stage of AI models. Based on an integrative reading of the AI Act, we comment on legal as well as technical enforcement problems and propose practical implications on bias detection and bias correction in order to specify and comply with specific technical requirements.
Paper Structure (12 sections, 2 figures)

This paper contains 12 sections, 2 figures.

Table of Contents

  1. Introduction
  2. Non-discrimination law vs. algorithmic fairness
  3. Legal Context: Non-discrimination law and its shortcomings.
  4. Technical context: algorithmic fairness and its shortcomings.
  5. Implications of the AI Act
  6. Enforcement "by design"?
  7. Enabling "bias detection and correction"?
  8. Practical challenges for compliance
  9. Defining bias: what are "appropriate" fairness metrics?
  10. Detecting and correcting bias: when are biases "likely to lead to discrimination"?
  11. Conclusion
  12. Appendix

Figures (2)

  • Figure 1: Type 2 error for increasing the disparity of the acceptance rate for two groups
  • Figure 2: Type 2 error for the same disparity in acceptance rate for two groups but for different acceptance rate values