Scholarlink

<p>Algorithmic fairness is commonly formalized through statistical notions of group fairness or distance-based measures of individual fairness, both of which are presented as objective measures mitigating discrimination. In this paper, we argue that these approaches, for different reasons, fail to capture the subjective and contextual nature of fairness. We propose a conceptual shift from the classical principle that <i>equals should be treated equally</i> to a subjective extension in which <i>perceived equals perceive themselves to be treated equally</i>. Within this perspective, similarity and fairness are not imposed ex ante by the decision-maker but are assessed by the individuals affected by the decision. We introduce a framework that models fairness as an aggregation of individual similarity judgments and outcome evaluations, thereby enabling the analysis of how fairness is perceived by those subject to algorithmic decisions. This approach provides insight into what individual fairness perceptions reveal about the system itself and offers a principled way to aggregate these perceptions into a collective measure, potentially informing both system evaluation and stakeholder oversight.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Subjective fairness in algorithmic decision-support

  • 期刊论文

  • 发表日期：

    2026年7月6日

查看全文
  • Sarra Tajouri,
  • Alexis Tsoukiàs

摘要

Algorithmic fairness is commonly formalized through statistical notions of group fairness or distance-based measures of individual fairness, both of which are presented as objective measures mitigating discrimination. In this paper, we argue that these approaches, for different reasons, fail to capture the subjective and contextual nature of fairness. We propose a conceptual shift from the classical principle that equals should be treated equally to a subjective extension in which perceived equals perceive themselves to be treated equally. Within this perspective, similarity and fairness are not imposed ex ante by the decision-maker but are assessed by the individuals affected by the decision. We introduce a framework that models fairness as an aggregation of individual similarity judgments and outcome evaluations, thereby enabling the analysis of how fairness is perceived by those subject to algorithmic decisions. This approach provides insight into what individual fairness perceptions reveal about the system itself and offers a principled way to aggregate these perceptions into a collective measure, potentially informing both system evaluation and stakeholder oversight.