An official website of the United States government
Identifying the causal pathways of unfairness is a critical objective for improving policy design and algorithmic decision-making. Prior work in causal fairness analysis often requires knowledge of the causal graph, hindering practical applications in complex or low-knowledge domains. Moreover, global discovery methods that learn causal structure from data can display unstable performance on finite samples, preventing robust fairness conclusions. To mitigate these challenges, we introduce local discovery for direct discrimination (LD3): a method that uncovers structural evidence of direct unfairness by identifying the causal parents of an outcome variable. LD3 performs a linear number of conditional independence tests relative to variable set size, and allows for latent confounding under the sufficient condition that all parents of the outcome are observed. We show that LD3 returns a valid adjustment set (VAS) under a new graphical criterion for the weighted controlled direct effect, a qualitative indicator of direct discrimination. LD3 limits unnecessary adjustment, providing interpretable VAS for assessing unfairness. We use LD3 to analyze causal fairness in two complex decision systems: criminal recidivism prediction and liver transplant allocation. LD3 was more time-efficient and returned more plausible results on real-world data than baselines, which took 46× to 5870× longer to execute.
more »
« less
This content will become publicly available on April 26, 2026
Fairness Testing Through Extreme Value Theory
Data-driven software is increasingly being used as a critical component of automated decision-support systems. Since this class of software learns its logic from historical data, it can encode or amplify discriminatory practices. Previous research on algorithmic fairness has focused on improving “average-case” fairness. On the other hand, fairness at the extreme ends of the spectrum, which often signifies lasting and impactful shifts in societal attitudes, has received significantly less emphasis. Leveraging the statistics of extreme value theory (EVT), we propose a novel fairness criterion called extreme counterfactual discrimination (ECD). This criterion estimates the worst-case amounts of disadvantage in outcomes for individuals solely based on their memberships in a protected group. Utilizing tools from search-based software engineering and generative AI, we present a randomized algorithm that samples a statistically significant set of points from the tail of ML outcome distributions even if the input dataset lacks a sufficient number of relevant samples. We conducted several experiments on four ML models (deep neural networks, logistic regression, and random forests) over 10 socially relevant tasks from the literature on algorithmic fairness. First, we evaluate the generative AI methods and find that they generate sufficient samples to infer valid EVT distribution in 95% of cases. Remarkably, we found that the prevalent bias mitigators reduce the average-case discrimination but increase the worst-case discrimination significantly in 35% of cases. We also observed that even the tail-aware mitigation algorithm—MiniMax-Fairness—increased the worst-case discrimination in 30% of cases. We propose a novel ECD-based mitigator that improves fairness in the tail in 90% of cases with no degradation of the average-case discrimination. We hope that the EVT framework serves as a robust tool for evaluating fairness in both average-case and worst-case discrimination.
more »
« less
- Award ID(s):
- 2317207
- PAR ID:
- 10651334
- Publisher / Repository:
- IEEE
- Date Published:
- Page Range / eLocation ID:
- 1501 to 1513
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Fairness Artificial Intelligence (AI) aims to identify and mitigate bias throughout the AI development process, spanning data collection, modeling, assessment, and deployment—a critical facet of establishing trustworthy AI systems. Tackling data bias through techniques like reweighting samples proves effective for promoting fairness. This paper undertakes a systematic exploration of reweighting samples for conventional Machine-Learning (ML) models, utilizing five models for binary classification on datasets such as Adult Income and COMPAS, incorporating various protected attributes. In particular, AI Fairness 360 (AIF360) from IBM, a versatile open-source library aimed at identifying and mitigating bias in machine-learning models throughout the entire AI application lifecycle, is employed as the foundation for conducting this systematic exploration. The evaluation of prediction outcomes employs five fairness metrics from AIF360, elucidating the nuanced and model-specific efficacy of reweighting samples in fostering fairness within traditional ML frameworks. Experimental results illustrate that reweighting samples effectively reduces bias in traditional ML methods for classification tasks. For instance, after reweighting samples, the balanced accuracy of Decision Tree (DT) improves to 100%, and its bias, as measured by fairness metrics such as Average Odds Difference (AOD), Equal Opportunity Difference (EOD), and Theil Index (TI), is mitigated to 0. However, reweighting samples does not effectively enhance the fairness performance of K Nearest Neighbor (KNN). This sheds light on the intricate dynamics of bias, underscoring the complexity involved in achieving fairness across different models and scenarios.more » « less
-
Machine Learning (ML) algorithms are increasingly used in our daily lives, yet often exhibit discrimination against protected groups. In this talk, I discuss the growing concern of bias in ML and overview existing approaches to address fairness issues. Then, I present three novel approaches developed by my research group. The first leverages generative AI to eliminate biases in training datasets, the second tackles non-convex problems arise in fair learning, and the third introduces a matrix decomposition-based post-processing approach to identify and eliminate unfair model components.more » « less
-
Many modern machine learning tasks require models with high tail performance, i.e. high performance over the worst-off samples in the dataset. This problem has been widely studied in fields such as algorithmic fairness, class imbalance, and risk-sensitive decision making. A popular approach to maximize the model’s tail performance is to minimize the CVaR (Conditional Value at Risk) loss, which computes the average risk over the tails of the loss. However, for classification tasks where models are evaluated by the 0/1 loss, we show that if the classifiers are deterministic, then the minimizer of the average 0/1 loss also minimizes the CVaR 0/1 loss, suggesting that CVaR loss minimization is not helpful without additional assumptions. We circumvent this negative result by minimizing the CVaR loss over randomized classifiers, for which the minimizers of the average 0/1 loss and the CVaR 0/1 loss are no longer the same, so minimizing the latter can lead to better tail performance. To learn such randomized classifiers, we propose the Boosted CVaR Classification framework which is motivated by a direct relationship between CVaR and a classical boosting algorithm called LPBoost. Based on this framework, we design an algorithm called alpha-AdaLPBoost. We empirically evaluate our proposed algorithm on four benchmark datasets and show that it achieves higher tail performance than deterministic model training methods.more » « less
-
As algorithms have become ubiquitous in consequential domains, societal concerns about the potential for discriminatory outcomes have prompted urgent calls to address algorithmic bias. In response, a rich literature across computer science, law, and ethics is rapidly proliferating to advance approaches to designing fair algorithms. Yet computer scientists, legal scholars, and ethicists are often not speaking the same language when using the term ‘bias.’ Debates concerning whether society can or should tackle the problem of algorithmic bias are hampered by conflations of various understandings of bias, ranging from neutral deviations from a standard to morally problematic instances of injustice due to prejudice, discrimination, and disparate treatment. This terminological confusion impedes efforts to address clear cases of discrimination. In this paper, we examine the promises and challenges of different approaches to disambiguating bias and designing for justice. While both approaches aid in understanding and addressing clear algorithmic harms, we argue that they also risk being leveraged in ways that ultimately deflect accountability from those building and deploying these systems. Applying this analysis to recent examples of generative AI, our argument highlights unseen dangers in current methods of evaluating algorithmic bias and points to ways to redirect approaches to addressing bias in generative AI at its early stages in ways that can more robustly meet the demands of justice.more » « less
