[Paper Review] There is an elephant in the room: Towards a critique on the use of fairness in biometrics

Ana María Cuadros Valdivia, Júlia Corbera-Serrajòrdia+1 more|arXiv (Cornell University)|Dec 16, 2021

COVID-19 Digital Contact Tracing4 citations

TL;DR

This paper critiques the growing focus on algorithmic fairness in biometric systems, arguing that fairness metrics are mathematically incompatible and cannot resolve the deeper political and historical injustices embedded in biometrics used for migration control. It demonstrates theoretically and empirically that fair biometric systems are impossible due to conflicting fairness definitions and biased training data, revealing that fairness discourse distracts from the systemic racism and colonial legacy of border technologies.

ABSTRACT

In 2019, the UK's Immigration and Asylum Chamber of the Upper Tribunal dismissed an asylum appeal basing the decision on the output of a biometric system, alongside other discrepancies. The fingerprints of the asylum seeker were found in a biometric database which contradicted the appellant's account. The Tribunal found this evidence unequivocal and denied the asylum claim. Nowadays, the proliferation of biometric systems is shaping public debates around its political, social and ethical implications. Yet whilst concerns towards the racialised use of this technology for migration control have been on the rise, investment in the biometrics industry and innovation is increasing considerably. Moreover, fairness has also been recently adopted by biometrics to mitigate bias and discrimination on biometrics. However, algorithmic fairness cannot distribute justice in scenarios which are broken or intended purpose is to discriminate, such as biometrics deployed at the border. In this paper, we offer a critical reading of recent debates about biometric fairness and show its limitations drawing on research in fairness in machine learning and critical border studies. Building on previous fairness demonstrations, we prove that biometric fairness criteria are mathematically mutually exclusive. Then, the paper moves on illustrating empirically that a fair biometric system is not possible by reproducing experiments from previous works. Finally, we discuss the politics of fairness in biometrics by situating the debate at the border. We claim that bias and error rates have different impact on citizens and asylum seekers. Fairness has overshadowed the elephant in the room of biometrics, focusing on the demographic biases and ethical discourses of algorithms rather than examine how these systems reproduce historical and political injustices.

Motivation & Objective

• To challenge the assumption that fairness can resolve ethical issues in biometric systems used for migration control.
• To demonstrate that fairness definitions in machine learning are mathematically mutually exclusive, making unbiased biometric systems theoretically impossible.
• To empirically show that biometric systems remain unfair across gender, age, and race groups despite claims of bias mitigation.
• To expose how biometric datasets reproduce racialized and gendered hierarchies through outdated, offensive categories.
• To argue that the focus on fairness distracts from the political reality that biometric systems are designed to enable border control and discrimination.

Proposed method

• Theoretical analysis translating fairness definitions from machine learning into biometric contexts to prove mathematical incompatibility.
• Reproduction of experiments from prior biometric fairness studies to empirically test fairness criteria across demographic groups.
• Evaluation of fairness using multiple criteria—demographic parity, equal opportunity, and equalized odds—across age, gender, and race.
• Analysis of decision thresholds and their impact on fairness outcomes in biometric authentication systems.
• Critical examination of biometric training datasets for racial and gender categorization, highlighting archaic and exclusionary constructs.
• Integration of critical border studies and decolonial theory to frame biometric systems as tools of historical and political marginalization.

Experimental results

Research questions

• RQ1 Can biometric systems be truly fair when multiple fairness definitions are mathematically incompatible?
• RQ2 To what extent do existing biometric fairness studies accurately assess demographic bias, given flawed evaluation methods?
• RQ3 How do biometric datasets reproduce racial and gendered hierarchies through their categorization practices?
• RQ4 What are the political consequences of framing biometric systems as 'fair' when they are deployed in contexts of migration control?
• RQ5 Why does the focus on fairness obscure the deeper structural injustices embedded in biometric systems at borders?

Key findings

• Multiple fairness definitions in machine learning—such as demographic parity, equal opportunity, and equalized odds—are mathematically mutually exclusive, making it impossible for any biometric system to satisfy all simultaneously.
• Empirical replication of prior biometric fairness studies shows significant disparities in false acceptance and rejection rates across gender, age, and race groups, even at optimal decision thresholds.
• Biometric training datasets systematically underrepresent non-Western populations and rely on outdated, offensive racial and gender categories that reinforce colonial and patriarchal hierarchies.
• The decision threshold in biometric systems critically affects fairness outcomes, yet most studies fail to account for this in their evaluation framework.
• Despite claims of 'bias-free' performance in some biometric systems (e.g., finger vein recognition), these claims are undermined by flawed statistical evaluation methods that ignore intersectionality and threshold sensitivity.
• The EU’s proposed AI regulation fails to address large-scale biometric databases used for migration control, allowing systems with inherent political and racialized functions to remain unregulated.