Algorithmic recourse under imperfect causal knowledge: a probabilistic approach

Abstract

Recent work has discussed the limitations of counterfactual explanations to recommend actions for algorithmic recourse, and argued for the need of taking causal relationships between features into consideration. Unfortunately, in practice, the true underlying structural causal model is generally unknown. In this work, we first show that it is impossible to guarantee recourse without access to the true structural equations. To address this limitation, we propose two probabilistic approaches to select optimal actions that achieve recourse with high probability given limited causal knowledge (e.g., only the causal graph). The first captures uncertainty over structural equations under additive Gaussian noise, and uses Bayesian model averaging to estimate the counterfactual distribution. The second removes any assumptions on the structural equations by instead computing the average effect of recourse actions on individuals similar to the person who seeks recourse, leading to a novel subpopulation-based interventional notion of recourse. We then derive a gradient-based procedure for selecting optimal recourse actions, and empirically show that the proposed approaches lead to more reliable recommendations under imperfect causal knowledge than non-probabilistic baselines.

arXiv commentCamera ready version (NeurIPS 2020 spotlight)

References in corpus (19)

• Auto-Encoding Variational Bayes
• Explaining Machine Learning Classifiers through Diverse Counterfactual Explanations
• Counterfactual Fairness
• Actionable Recourse in Linear Classification
• Causal Effect Inference with Deep Latent-Variable Models
• On the Identifiability of the Post-Nonlinear Causal Model
• Identifiability of Gaussian structural equation models with equal error variances
• The Hidden Assumptions Behind Counterfactual Explanations and Principal Reasons
• Causal Discovery from a Mixture of Experimental and Observational Data
• Preserving Causal Constraints in Counterfactual Explanations for Machine Learning Classifiers
• Towards Realistic Individual Recourse and Actionable Explanations in Black-Box Decision Making Systems
• Reliable Decision Support using Counterfactual Models
• Causal Discovery from Changes
• Algorithmic Recourse: from Counterfactual Explanations to Interventions
• Gaussian Process Networks
• A survey of algorithmic recourse: definitions, formulations, solutions, and prospects
• Conditional Density Estimation with Bayesian Normalising Flows
• Equalizing Recourse across Groups
• Fair Decisions Despite Imperfect Predictions

Cited by in corpus (23)

• Counterfactual Shapley Additive Explanations
• A survey of algorithmic recourse: definitions, formulations, solutions, and prospects
• Towards Robust and Reliable Algorithmic Recourse
• Counterfactuals and Causability in Explainable Artificial Intelligence: Theory, Algorithms, and Applications
• GAM Coach: Towards Interactive and User-centered Algorithmic Recourse
• From Explanation to Recommendation: Ethical Standards for Algorithmic Recourse
• Improvement-Focused Causal Recourse (ICR)
• A Causal Perspective on Meaningful and Robust Algorithmic Recourse
• Setting the Right Expectations: Algorithmic Recourse Over Time
• Synthesizing explainable counterfactual policies for algorithmic recourse with program synthesis
• Algorithmic Recourse in the Wild: Understanding the Impact of Data and Model Shifts
• Consistent Counterfactuals for Deep Models
• On the Connection between Game-Theoretic Feature Attributions and Counterfactual Explanations
• Disaggregated Interventions to Reduce Inequality
• Towards the Unification and Robustness of Perturbation and Gradient Based Explanations
• Exploring Counterfactual Explanations Through the Lens of Adversarial Examples: A Theoretical and Empirical Analysis
• Flexible and Robust Counterfactual Explanations with Minimal Satisfiable Perturbations
• Counterfactual Explanations for Machine Learning: Challenges Revisited
• Counterfactual Situation Testing: From Single to Multidimensional Discrimination
• Linear Classifiers that Encourage Constructive Adaptation
• Towards User Guided Actionable Recourse
• Estimation of Counterfactual Interventions under Uncertainties
• Learning Models for Actionable Recourse