PuSH - Publikationsserver des Helmholtz Zentrums München

Lee, M.S.A.* ; Padh, K.* ; Watson, D.* ; Kilbertus, N. ; Singh, J.*

Fairness under uncertainty in sequential decisions.

In: (FAccT '26: The 2026 ACM Conference on Fairness, Accountability, and Transparency, 25-28 June 2026, Montreal QC Canada). 2026. 1784 - 1809 (ACM Facct 2026 Proceedings of the 9th Annual ACM Conference on Fairness Accountability and Transparency)
DOI
Fair machine learning (ML) methods support the identification and mitigation of the risk that algorithms will encode and/orautomate social injustices. While algorithmic approaches alone cannot resolve structural inequalities, these techniques cansupport socio-technical decision systems by surfacing unintended discriminatory biases, clarifying trade-offs, and enablinggovernance. Although fairness has been well studied in supervised learning, many real-life ML applications are onlineand sequential, with feedback from previous decisions informing future decisions. Each decision in such a setting is takenunder uncertainty due to unobserved counterfactual outcomes and finite samples, with especially dire consequences forunder-represented groups, who are often systematically under-observed due to historical exclusion and selective feedback.For example, a bank cannot know whether a denied loan would have been repaid, and it may have less data about previouslymarginalized and financially excluded populations.Towards this, this paper introduces a taxonomy of uncertainty in sequential decision-making—including model uncertainty,feedback uncertainty, and prediction uncertainty—to provide a shared vocabulary for assessing and governing sequentialdecision systems in which uncertainty is unevenly distributed across groups. We formalize the model uncertainty and feedbackuncertainty using counterfactual logic and reinforcement learning techniques. We illustrate the potential harms for both thedecision maker (unrealized gains and losses) and the decision subject (compounding exclusion and reduced access) of naïvepolicies that ignore the unobserved space. We illustrate our framework using simple algorithmic examples that demonstratethe possibility of simultaneously reducing the variance in outcomes for historically disadvantaged groups while preservinginstitutional objectives (e.g. expected utility) of the decision maker. Our experiments on data, simulated to include varyingdegrees of bias, illustrate how unequal uncertainty and selective feedback can produce disparities in sequential decisionsystems, and how uncertainty-aware exploration can alter observed fairness metrics. By providing a structured lens foridentifying where and how uncertainty arises, this framework equips researchers and practitioners to better diagnose, audit,and govern fairness risks in sequential decision systems. In sequential and online systems in which uncertainty is a core driverof unfair outcomes, not merely incidental noise, our results highlight the importance of explicitly accounting for uncertaintyin fair and effective decision-making.
Altmetric
Weitere Metriken?
[➜Einloggen]
Zusatzinfos bearbeiten [➜Einloggen]
Publikationstyp Artikel: Konferenzbeitrag
Schlagwörter Algorithmic Fairness ; Online Learning Algorithms ; Sequential Decision Making ; Uncertainty
ISSN (print) / ISBN [9798400725968]
Konferenztitel FAccT '26: The 2026 ACM Conference on Fairness, Accountability, and Transparency
Konferzenzdatum 25-28 June 2026
Konferenzort Montreal QC Canada
Quellenangaben Band: , Heft: , Seiten: 1784 - 1809 Artikelnummer: , Supplement: ,
Institut(e) Helmholtz Artifical Intelligence Cooperation Unit (HAICU)