Fair Models in Credit: Intersectional Discrimination and the Amplification of Inequity

Intersectional discrimination is invisible in standard fairness assessments: The study found that while machine learning models may appear fair when evaluated for single characteristics (like gender or age alone), unfairness can actually worsen when examining people defined by multiple overlapping identities—for example, single mothers over 40 face compounded discrimination that wouldn't be detected by traditional fairness metrics.

Vulnerable groups experience disproportionate credit barriers: The research demonstrates that difficulties in accessing credit through automated systems are not randomly distributed but systematically affect vulnerable populations, particularly those at the intersection of multiple marginalized categories in the Spanish microfinance market.

Current fairness frameworks are insufficient: Existing approaches to algorithmic fairness that focus on individual protected characteristics (race, sex, age) miss critical patterns of inequity that emerge from the combination of multiple social categories, meaning current "fair" AI systems may still perpetuate hidden discrimination.

Machine learning amplifies rather than solves societal inequities: Because ML technology operates without awareness of social context or human welfare, it can inadvertently amplify existing inequalities in credit allocation unless explicitly designed with intersectional fairness in mind—fairness doesn't happen by default.

Intersectionality offers a path toward more authentic equity: By adopting an intersectionality lens that examines how multiple identities interact, organizations can develop more comprehensive fairness assessments and create genuinely equitable outcomes rather than superficially "fair" systems that mask deeper discrimination.