Certifying Fairness of Artificial Intelligence : Emerging Trends in Internet Technologies, Summer Term 2021

Background: Artificial Intelligence (AI)-embedded systems offer many benefits, however, also come with fairness issues and risks of discrimination, such as racial bias. A promising approach for ensuring fairness are certifications as a means of (independently) assessing the fairness and non-discrimination of AI-embedded systems.
Objective: However, research and practice still struggle with designing and performing fairness certifications. In this study, we clarify how to certify fairness of AI-embedded systems.
Methods: We enhance the understanding by performing a literature review on fairness in AI and conducting expert interviews.
Results: Our study discusses three key structural building blocks for fairness certifications: fairness criteria to be assessed (certification content), potential issuers and auditors (certification sources), and auditing methods to evaluate fairness (certification process).
Conclusion: With this study, we provide a first conceptualization of important building blocks for a fairness certification to highlight its feasibility and usefulness in the domain of AI.