Counterfactual Fairness

By Matt J. Kusner, Joshua R. Loftus, Chris Russell, Ricardo M. A. Silva. Neural Information Processing Systems, 2017

1,918 citations recorded by the source corpus.

Machine learning can impact people with legal or ethical consequences when it is used to automate decisions in areas such as insurance, lending, hiring, and predictive policing. In many of these scenarios, previous decisions have been made that are unfairly biased against certain subpopulations, for example those of a particular race, gender, or sexual orientation. Since this past data may be biased, machine learning predictors must account for this to avoid perpetuating or creating discriminatory practices. In this paper, we develop a framework for modeling fairness using tools from causal inference. Our definition of counterfactual fairness captures the intuition that a decision is fair towards an individual if it the same in (a) the actual world and (b) a counterfactual world where the individual belonged to a different demographic group. We demonstrate our framework on a real-world problem of fair prediction of success in law school.

Subfields: Safety, Robustness & Interpretability

View Paper: https://arxiv.org/abs/1703.06856

How researchers see this paper

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

In Safety, Robustness & Interpretability

Citations in the corpus

Reducing Sentiment Bias in Language Models via Counterfactual Evaluation (2019, background): “, 2012), counterfactual fairness requires the same model predictions before and after intervention on sensitive attributes in data-generating causal graphs (Kusner et al., 2017; Kilbertus et al., 2017; Chiappa, 2019; Chiappa and Isaac, 2019).”
Shortcut learning in deep neural networks (2020, background): “eating similar individuals similarly while group fairness aims at treating subgroups no different than the rest of the population [133,134]. Fairness is closely linked to generalisation and causality [135]. Sensitive group membership can be viewed as a domain indicator: Just like machine decisions s”
The Capacity for Moral Self-Correction in Large Language Models (2023, methodology): “We do not advocate for the use of language models for admissions decisions—rather we treat this as a proxy experiment for discrimination that parallels previous work [18, 23, 30, 33, 48].”
Causal Mediation Analysis for Interpreting Neural NLP: The Case of Gender Bias (2020, background): “This conflicts with the principle of counterfactual fairness, which states that the model predictions should not be influenced by changes to a sensitive attribute such as gender (Kusner et al., 2017); for instance, a fair and unbiased model should equally associate gendered pronouns with professions”
Gender Bias in Coreference Resolution: Evaluation and Debiasing Methods (2018, methodology): “Kusner et al. (2017) propose the method based on causal inference to achieve the model fairness where they do the data augmentation under speciﬁc cases, however, to the best of our knowledge, we are the ﬁrst to propose data augmentation based on gender swapping in order to reduce gender bias.”
Counterfactual Fairness in Text Classification through Robustness (2018, background): “ncorporate any external understanding about what is causally acceptable in making predictions. The limitations of observational criteria can be addressed by counterfactual or individual fairness, see [3, 9, 12]. Social science literature on fairness raises arguments for counterfactual reasoning as w”
Advances and Open Problems in Federated Learning (2019, background): “airness requires that a user receive the same treatment as they would have if they had been a member of a different group (race, gender, etc), after taking all causally relevant pathways into account [250]. Federated learning raises several opportunities for fairness research, some of which extend p”
A Survey on Bias and Fairness in Machine Learning (2019, background): “ps should have equal true positive rates. Definition 3.(Demographic Parity). Also known as statistical parity. "A predictor Y satisfies demo-ˆ graphic parity if P(Y |A = 0) = P(ˆ Y|A = 1)"ˆ [43, 73]. The likelihood of a positive outcome [123] should be the same regardless of whether the pe”
Toward Causal Representation Learning (2021, background): “the general goal of modern machine learning to learn meaningful representations of data, where meaningful can include robust , explainable , or fair [130], [134], [142], [259], [275].”
Beyond Individual and Group Fairness (2020, background): “, 2016], counterfactual or demographic parity [Kusner et al., 2017], calibration [Guo et al.”
Prediction-Based Decisions and Fairness: A Catalogue of Choices, Assumptions, and Definitions (2018, background): “When Data = X (the insensitive variables), it is equivalent to Fairness through Unawareness [78], anticlassification [18], or treatment parity [82].”
FairSight: Visual Analytics for Fairness in Decision Making (2019, methodology): “We also include the in-processing model (Additive Counterfactual Fairness (ACF)) [26]) and the post-processing method (FA*IR [41]) as a way of mitigating bias ( R3 ).”
Generative Counterfactual Introspection for Explainable Deep Learning (2019, methodology): “With the pressing need to obtain causal understanding of model behavior, interpretation approaches [20]–[23] focusing on counterfactual reasoning have been proposed.”
On formalizing fairness in prediction with machine learning (2017, background): “(27) recently introduced one such measure which can be defined as follows: Definition 2 A predictor H is counterfactually fair, given Z = z and A = a, for all y and a 6= a, iff”
Statistical Equity: A Fairness Classification Objective (2020, background): “In counterfactual fairness [Kusner et al., 2017] the goal is to satisfy under any X = x and A = a for all y and for any value a (cid:48) feasible for A .”
Man is to Person as Woman is to Location: Measuring Gender Bias in Named Entity Recognition (2019, methodology): “hat this miscategorization affects more women than men. This serves as our deﬁnition of bias which considers the differences between gender groups following the statistical parity notion of fairness (Kusner et al. 2017; Dwork et al. 2012). The contributions of this paper are fourfold: 1.We introduce”
Wasserstein Fair Classification (2019, methodology): “arrio et al., 2019; Johndrow and Lum, 2019). The second approach consists in performing a post-processing of the model outputs (Chiappa, 2019; Doherty et al., 2012; Feldman, 2015; Hardt et al., 2016; Kusner et al., 2017). The third approach consists in enforcing fairness notions by imposing constrai”
Flexibly Fair Representation Learning by Disentanglement (2019, background): “This property relates to counterfactual fairness, which requires that models be robust with respect to counterfactuals along a (Kusner et al., 2017).”