Ferrara's thesis challenges fairness auditing with structural context

Antonio Ferrara's doctoral thesis, 'Statistical and Structural Approaches to Algorithmic Fairness,' takes aim at the foundations of how we measure and mitigate bias in machine learning. The paper argues that as ML systems evolve from isolated predictors into mediators of human opportunity, they inherently absorb structural inequalities from their environments. Existing fairness methods rely on fragile simplifications—particularly deterministic point estimates during audits—which fail to capture the dynamic, systemic nature of bias. Ferrara also criticizes the prevalent practice of treating individuals as independent entities, ignoring the social and relational contexts that shape outcomes.

To address these limitations, the thesis proposes a reframing of fairness that incorporates structural analysis and probabilistic reasoning. Instead of static, one-number audits, Ferrara advocates for auditing methods that account for interdependencies and systemic effects. The work is relevant for researchers and practitioners building fairness toolkits, regulators crafting AI oversight policies, and engineers deploying models in high-stakes domains like hiring, lending, and criminal justice. By exposing the gaps in current approaches, the thesis signals a needed shift toward more robust, contextually aware fairness evaluation.