Path-Specific Counterfactual Fairness

By Silvia Chiappa. 2018

399 citations recorded by the source corpus.

We consider the problem of learning fair decision systems from data in which a sensitive attribute might affect the decision along both fair and unfair pathways. We introduce a counterfactual approach to disregard effects along unfair pathways that does not incur in the same loss of individual-specific information as previous approaches. Our method corrects observations adversely affected by the sensitive attribute, and uses these to form a decision. We leverage recent developments in deep learning and approximate inference to develop a VAE-type method that is widely applicable to complex nonlinear models.

Subfields: Safety Research, Sociology and Political Science, Safety, Robustness & Interpretability, AI Ethics, Fairness & Society

How researchers see this paper

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

In Safety Research

In Sociology and Political Science

In Safety, Robustness & Interpretability

In AI Ethics, Fairness & Society

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).”
Counterfactual Fairness in Text Classification through Robustness (2018, background): “By extending these definitions to path-specific counterfactual fairness, it is possible to specify which uses of identity are acceptable [2].”
A Survey on Bias and Fairness in Machine Learning (2019, methodology): “In [32] authors introduce the path-specific counterfactual fairness definition which is an extension to counterfactual fairness definition [87] and propose a method to achieve it further extending the work in [116].”
Explanation Methods in Deep Learning: Users, Values, Concerns and Challenges (2018, background): “Work is in progress to mitigate or remove the inﬂuence of sensitive variables that should not aﬀect decision outcomes by embracing causal inference procedures (Chiappa and Gillam, 2018).”
Wasserstein Fair Classification (2019, methodology): “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).”
A Causal Bayesian Networks Viewpoint on Fairness (2018, neutral): “ŷ = 〈Yā(Da)〉p(Yā(Da)|A=a,Q=qn,D=dn) — the resulting model is said to satisfy path-specific counterfactual fairness [11].”
Causal Reasoning for Algorithmic Fairness (2018, methodology): “A number of recent works use causal approaches to address fairness [1, 7, 21, 23, 35, 44], which we review in more detail in Section 5.”
On the Fairness of Causal Algorithmic Recourse (2020, methodology): “However, it has been argued from causal (Chiappa 2019; Wu et al. 2019) and noncausal (Dwork et al. 2012) perspectives that fairness is fundamentally an individual-level concept:6 group-level fairness still allows for unfairness at the level of the individual, provided that positive and negative…”
Algorithmic recourse under imperfect causal knowledge: a probabilistic approach (2020, background): “The importance of causal reasoning for an ethical and socially beneficial use of ML-assisted technology has also been stressed in a number of recent works in the field of explainability and fair algorithmic decision making [29, 42, 24, 63, 64, 10, 57, 15].”
PC-Fairness: A Unified Framework for Measuring Causality-based Fairness (2019, background): “In [3], the author introduces the term “path-specific counterfactual fairness”, which states that a decision is fair toward an individual if it coincides with the one that would have been taken in a counterfactual world in which the sensitive attribute along the unfair pathways were different.”
Fairness in Machine Learning: A Survey (2020, methodology): “References [65, 173, 219] transform the output of a classifier in accordance to the identification of unfair causal pathways, either by averaging [219], constraining the conditional distribution of the decision variable [173], or through counterfactual correction [65].”
Causal Analysis of Agent Behavior for AI Safety (2021, methodology): “To estimate the effect of a given pathway, one can chain together the effects of the individual mechanisms along the path (Shpitser, 2013; Chiappa, 2019).”
Fairness Violations and Mitigation under Covariate Shift (2019, background): “(a) Disease risk scoring under population shift and treatment policy shift [66] (b) Credit scoring under population shift [10].”
Towards Out-Of-Distribution Generalization: A Survey (2021, background): “Many works define their fairness and propose corresponding fair algorithms, from which the definition of fairness can be divided into three types: individual fairness [59, 293], group fairness [89, 119, 270], and causality-based fairness notions [120, 36].”