The Neutrality Fallacy: When Algorithmic Fairness Interventions are (Not) Positive Action
Hilde Weerts, Raphaële Xenidis, Fabien Tarissan, Henrik Palmer Olsen, Mykola Pechenizkiy
TL;DR
The paper interrogates whether fairness interventions in algorithmic decision-making should be treated as EU law–driven 'algorithmic positive action' or strictly as measures to avoid discrimination. It introduces three neutrality fallacies—data neutrality, model neutrality, and decision-making neutrality—to argue that many fairness interventions are not inherently positive action and may carry legal and normative implications. The authors conclude that only a narrow subset of algorithmic interventions aligns with quota-like positive-action criteria, while most should be framed as discrimination avoidance, potentially supported by a broader move toward a positive obligation to prevent discrimination. They advocate reframing the normative framework toward a proactive 'do no harm' duty, potentially institutionalized via the AI Act and related policies, while preserving the essential role of structural positive action to transform inequitable baselines beyond algorithmic systems.
Abstract
Various metrics and interventions have been developed to identify and mitigate unfair outputs of machine learning systems. While individuals and organizations have an obligation to avoid discrimination, the use of fairness-aware machine learning interventions has also been described as amounting to 'algorithmic positive action' under European Union (EU) non-discrimination law. As the Court of Justice of the European Union has been strict when it comes to assessing the lawfulness of positive action, this would impose a significant legal burden on those wishing to implement fair-ml interventions. In this paper, we propose that algorithmic fairness interventions often should be interpreted as a means to prevent discrimination, rather than a measure of positive action. Specifically, we suggest that this category mistake can often be attributed to neutrality fallacies: faulty assumptions regarding the neutrality of fairness-aware algorithmic decision-making. Our findings raise the question of whether a negative obligation to refrain from discrimination is sufficient in the context of algorithmic decision-making. Consequently, we suggest moving away from a duty to 'not do harm' towards a positive obligation to actively 'do no harm' as a more adequate framework for algorithmic decision-making and fair ml-interventions.