Taming the AI Monster: Monitoring of Individual Fairness for Effective Human Oversight

Abstract

This invited paper reviews a framework to assist in mitigating societal risks that software can pose. This is to promote effective human oversight, which is a central requirement enforced by the European Union’s upcoming AI Act. The paper advertises fragments of an upcoming journal publication, and as such is itself low in genuine originality. Yet it offers a specific perspective on that original work. Extrapolating earlier work on software doping, we report on the combination of established techniques for runtime monitoring and for probabilistic falsification to arrive at a black-box analysis technique for identifying undesired effects of software. We describe its application to high-risk systems that evaluate humans in a possibly unfair or discriminating way. The approach can assist humans-in-the-loop to make better informed and more responsible decisions. Our technical contribution is complemented by juridically …

@inproceedings{baumTamingTheAIMonster2024,
 author = {Kevin Baum and Biewer Sebastian and Hermanns Holger and Hetmank Sven and Langer Markus and Lauber-Rönsberg Anne and Sterz Sarah},
 booktitle = {Model Checking Software - 30th International Symposium - Proceedings. International Symposium on Model Checking Software (SPIN-2024), April 8-9, Luxembourg City, Luxembourg},
 doi = {10.1007/978-3-031-66149-5_1},
 editor = {Thomas Neele and Anton Wijs},
 month = {10},
 pages = {3-25},
 publisher = {Springer Nature, Switzerland},
 series = {Lecture Notes in Computer Science, LNCS},
 title = {Taming the AI Monster: Monitoring of Individual Fairness for Effective Human Oversight},
 volume = {14624},
 year = {2024}
}