2026/05/31 by Holli Sargeant · 1 voice
Decision Sciences · #Impact of AI and Big Data on Business and Society
paper · doi:10.1111/1468-2230.70045
Machine learning (ML) systems, increasingly deployed in high-stakes decisionmaking, inherently produce uncertain outputs that can lead to unlawful discrimination. This article examines the legal implications of predictive uncertainty in ML systems under UK antidiscrimination law. Employing a decision-theoretic framework, the article distinguishes between aleatoric uncertainty, stemming from irreducible randomness, and epistemic uncertainty, arising from incomplete knowledge or deliberate model design choices. It argues that intentional modelling decisions introduce epistemic uncertainty, which can directly and indirectly cause discriminatory outcomes. Instead of evaluating ML as a 'black-box' or solely by its outputs, greater legal importance should be placed on the policy and design choices of ML systems embedded within these systems than has been traditionally acknowledged. While identifying and justifying aleatoric uncertainty presents a unique challenge, given its reflection of underlying, irreducible risk external to the model, the practical difficulty of distinguishing these two forms of uncertainty poses pressing issues for discrimination law. The article contends that aspects of current anti-discrimination frameworks are ill-equipped to address these probabilistic harms and advocates for enhanced interdisciplinary interpretation of anti-discrimination doctrine with support from proactive oversight and regulatory measures beyond individual litigation. This article provides the first legal analysis of predictive uncertainty and unlawful discrimination.