Beyond Distributive Fairness in Algorithmic Decision Making: Feature Selection for Procedurally Fair Learning

By Nina Grgic-Hlaca, Muhammad Bilal Zafar, Krishna P. Gummadi, Adrian Weller. 2018

243 citations recorded by the source corpus.

With widespread use of machine learning methods in numerous domains involving humans, several studies have raised questions about the potential for unfairness towards certain individuals or groups. A number of recent works have proposed methods to measure and eliminate unfairness from machine learning models. However, most of this work has focused on only one dimension of fair decision making: distributive fairness, i.e., the fairness of the decision outcomes. In this work, we leverage the rich literature on organizational justice and focus on another dimension of fair decision making: procedural fairness, i.e., the fairness of the decision making process. We propose measures for procedural fairness that consider the input features used in the decision process, and evaluate the moral judgments of humans regarding the use of these features. We operationalize these measures on two real world datasets using human surveys on the Amazon Mechanical Turk (AMT) platform, demonstrating that our measures capture important properties of procedurally fair decision making. We provide fast submodular mechanisms to optimize the tradeoff between procedural fairness and prediction accuracy. On our datasets, we observe empirically that procedural fairness may be achieved with little cost to outcome fairness, but that some loss of accuracy is unavoidable.

Subfields: Cognitive Neuroscience, Safety Research

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Cognitive Neuroscience

In Safety Research

Citations in the corpus

Fairness and Abstraction in Sociotechnical Systems (2019, background): “That the mathematical definitions eliminate these nuances is a fact known to many fair-ML researchers [14, 24, 31, 53].”
Towards Robust Interpretability with Self-Explaining Neural Networks (2018, background): “Propublica’s study showing racial-biased scores sparked a flurry of interest in the COMPAS algorithm both in the media and in the fairness in machine learning community [25, 7].”
Blind Justice: Fairness with Encrypted Sensitive Attributes (2018, background): “While this has some intuitive appeal and justiﬁcation (Grgi´c-Hlaˇca et al., 2018), a signiﬁcant concern is that sensitive attributes may often be accurately predicted (“reconstructed”) from non-sensitive features (Dwork et al., 2012).”
You Shouldn't Trust Me: Learning Models Which Conceal Unfairness From Multiple Explanation Methods (2020, background): “This is sometimes considered a form of process fairness via unawareness [7, 12].”
Ethical Considerations in AI-Based Recruitment (2019, background): “3) Feature selection: Features used to improve the model may result in an unfair prediction [23].”
Human Perceptions of Fairness in Algorithmic Decision Making: A Case Study of Criminal Risk Prediction (2018, background): “ions around the above question for multiple reasons: First, people’s judgments about fairness of using features can be leveraged to learn fair algorithmic decision making, as shown in our recent work [28, 29]. Second, arXiv:1802.09548v1 [stat.ML] 26 Feb 2018 while the question, “is this feature fair”
Explainable AI: A Review of Machine Learning Interpretability Methods (2020, background): “[89] procedurally_fair_learning F G Agnostic tab 22 2018”
Trustworthy AI: From Principles to Practices (2021, background): “At each granularity, we might be concerned with distributive fairness or fairness of the outcome, or procedural fairness or fairness of the process (we refer the reader to Reference [137] for a more detailed discussion).”
A Review on Fairness in Machine Learning (2022, background): “[91] collect human judgments through surveys and introduce three fairness measures that take into account the perceived fairness of the features that are used for decision making.”
Explaining The Efficacy of Counterfactually-Augmented Data (2020, background): “Concerns about bias in algorithms stem in part from principles of procedural fairness (Blader & Tyler, 2003; Miller, 2017; Grgic-Hlaca et al., 2018; Lipton et al., 2018).”
Towards a Fair Marketplace: Counterfactual Evaluation of the trade-off between Relevance, Fairness & Satisfaction in Recommendation Systems (2018, background): “Recent work has explored the development of classification models with fairness-aware regularization [30], fairness aware decision making systems [9], prediction models for individual fairness [6], fair division of resources [2], auditing for fairness [17] and fairness in rankings [4, 25].”
Algorithmic bias: review, synthesis, and future research directions (2021, background): “…algorithm or algorithmic decision as being fair, whereas procedural fairness conceptua-lises fairness perceptions with respect to internal processes of data-driven systems such as the rules and logics incorporated into algorithms (Binns et al., 2018; Grgić-Hlača et al., 2018; Robert et al., 2020).”
Equity of Attention: Amortizing Individual Fairness in Rankings (2018, background): “Beyond the fairness of outcomes, researchers have looked into the fairness of process in the decision-making systems [18].”
On the Legal Compatibility of Fairness Definitions (2019, background): “ce In ML, the term “procedural fairness” has been co-opted to refer to identifying the input features thatleadtoaparticularmodeloutcome,asaproxyforthe“process”throughwhichthemodelmakes its prediction [21]. This is a narrow and misguidedview of what this legal term means. Rather than attemptingtospec”
Human Decision Making with Machine Assistance (2019, background): “Several studies also explored human perceptions of the fairness of machine decisions, in the context of self-driving cars [5, 12], targeted-advertising [59], loan approval [62], and bail decisions [36–38 Another series of studies has compared human and machine decision-making.”
FairFed: Enabling Group Fairness in Federated Learning (2021, methodology): “In classical centralized ML, common approaches for realizing group fairness can be classified into three categories: pre-processing (Grgić-Hlača et al. 2018; Feldman et al. 2015), in-processing (Kamishima et al. 2012; Zhang, Lemoine, and Mitchell 2018; Roh et al. 2021) and post-processing (Lohia…”
The Disparate Effects of Strategic Manipulation (2018, background): “Much of the previous work in algorithmic fairness has examined a particular undesirable proxy effect in which a classifier’s features may be linked to socially significant and legally protected attributes like race and gender, interpreting correlations that have arisen due to centuries of accumulate”