Algorithmic Fairness: Choices, Assumptions, and Definitions

By Shira Mitchell, Eric Potash, Solon Barocas, A. D'Amour, Kristian Lum. Annual Review of Statistics and Its Application, 2021

600 citations recorded by the source corpus.

A recent wave of research has attempted to define fairness quantitatively. In particular, this work has explored what fairness might mean in the context of decisions based on the predictions of statistical and machine learning models. The rapid growth of this new field has led to wildly inconsistent motivations, terminology, and notation, presenting a serious challenge for cataloging and comparing definitions. This article attempts to bring much-needed order. First, we explicate the various choices and assumptions made—often implicitly—to justify the use of prediction-based decision-making. Next, we show how such choices and assumptions can raise fairness concerns and we present a notationally consistent catalog of fairness definitions from the literature. In doing so, we offer a concise reference for thinking through the choices, assumptions, and fairness considerations of prediction-based decision-making.

Subfields: Artificial Intelligence, Safety Research

View Paper: https://www.semanticscholar.org/paper/0508cc8d1a9f499a7276b62e2e7b788ab46f296a

How researchers see this paper

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

In Artificial Intelligence

In Safety Research

Citations in the corpus

Decolonizing Content Moderation: Does Uniform Global Community Standard Resemble Utopian Equality or Western Power Hegemony? (2023, background): “, the choices, assumptions, dataset selection, and fairness considerations—that shape predictive content moderation systems [96].”
Algorithmic Bias in Education (2021, background): “In the current article, we focus on studying algorithmic bias in terms of situations where model performance is substantially better or worse across mutually exclusive groups (i.e. Gardner et al., 2019; Mehrabi et al., 2019; Mitchell et al., 2021).”
The Fallacy of AI Functionality (2022, background): “However, this intuition is only valid when certain conditions and assumptions are met [67, 124, 198], such as the measurement validity of the data and labels.”
Regression diagnostics meets forecast evaluation: conditional calibration, reliability diagrams, and coefficient of determination (2021, background): “, 2020; Allen, 2021), sensitivity (Fissler and Pesenti, 2022), and fairness (Pleiss et al., 2017; Mitchell et al., 2021).”
Coarse race data conceals disparities in clinical risk score performance (2023, background): “Performance metrics Past work in algorithmic fairness has used numerous metrics to evaluate whether algorithms perform equally well across groups (Kleinberg et al., 2016; Chouldechova, 2017; Narayanan, 2018; Corbett-Davies and Goel, 2018; Chen et al., 2021b; Zink et al., 2023; Mitchell et al., 2021;”
Counterfactual Fairness Is Basically Demographic Parity (2022, background): “It follows a surge of papers over the past decade that borrow intuition from moral philosophy to formally deﬁne what makes an algorithm fair (Dwork et al. 2012; Feldman et al. 2015; Hardt, Price, and Srebro 2016; Mitchell et al. 2021).”