How to Choose a Fairness Measure: A Decision-Making Workflow for Auditors
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Recent developments in Artificial Intelligence (AI) have greatly benefited society, but they alsocome with risks. One of those risks is that AI has the potential to discriminate against certaingroups of people. To address that risk, benchmark regulations such as the AI Act have been cre-ated, requiring AI systems to be fair and tasking auditors with ensuring their compliance. In orderto do so, auditors use fairness measures. However, selecting a specific definition of fairness fromthe various available options and choosing a fairness measure from the numerous possibilities com-plicates the auditing process, making it challenging for auditors to correctly assess AI fairness. Toassist them, we created a decision-making workflow that guides the auditor through the selectionprocess of the most appropriate measure and, consequently, the most suitable definition of fairness.To simplify the use of this workflow, we have integrated it into the open-source program JASP forAudit and demonstrated its functionality with two examples: the COMPAS recidivism and the DUOcase.
