On Explaining Proxy Discrimination and Unfairness in Individual Decisions Made by AI Systems
Belona Sonna, Alban Grastien
TL;DR
This paper investigates explaining proxy discrimination in AI decisions using abductive explanations, where an explanation is a minimal subset $XP \subseteq x$ that guarantees the same decision $\Delta(x)$. It formalizes proxy discrimination through a context $\Phi$ and background knowledge $K$, introducing proxy variables $q$ for protected attributes and aptitude-based mappings that align decisions across subgroups. A greedy algorithm computes minimal $XP$ and a fairness criteria based on equivalent aptitudes, revealing that BK-aware bias can exist even when explanations do not explicitly include the protected attribute, as shown on the German Credit Dataset. The approach yields interpretable, context-aware fairness audits for high-stakes domains and supports extensions to intersectional and non-binary settings.
Abstract
Artificial intelligence (AI) systems in high-stakes domains raise concerns about proxy discrimination, unfairness, and explainability. Existing audits often fail to reveal why unfairness arises, particularly when rooted in structural bias. We propose a novel framework using formal abductive explanations to explain proxy discrimination in individual AI decisions. Leveraging background knowledge, our method identifies which features act as unjustified proxies for protected attributes, revealing hidden structural biases. Central to our approach is the concept of aptitude, a task-relevant property independent of group membership, with a mapping function aligning individuals of equivalent aptitude across groups to assess fairness substantively. As a proof of concept, we showcase the framework with examples taken from the German credit dataset, demonstrating its applicability in real-world cases.