Algorithmic Recourse: from Counterfactual Explanations to Interventions

By Amir-Hossein Karimi, Bernhard Schölkopf, Isabel Valera. 2020

456 citations recorded by the source corpus.

As machine learning is increasingly used to inform consequential decision-making (e.g., pre-trial bail and loan approval), it becomes important to explain how the system arrived at its decision, and also suggest actions to achieve a favorable decision. Counterfactual explanations -"how the world would have (had) to be different for a desirable outcome to occur"- aim to satisfy these criteria. Existing works have primarily focused on designing algorithms to obtain counterfactual explanations for a wide range of settings. However, it has largely been overlooked that ultimately, one of the main objectives is to allow people to act rather than just understand. In layman's terms, counterfactual explanations inform an individual where they need to get to, but not how to get there. In this work, we rely on causal reasoning to caution against the use of counterfactual explanations as a recommendable set of actions for recourse. Instead, we propose a shift of paradigm from recourse via nearest counterfactual explanations to recourse through minimal interventions, shifting the focus from explanations to interventions.

Subfields: Artificial Intelligence, 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 Artificial Intelligence

In Safety Research

Citations in the corpus

On the Fairness of Causal Algorithmic Recourse (2020, methodology): “To capture causal relations between features, Karimi, Schölkopf, and Valera (2021) propose to approach the actionable recourse task within the framework of SCMs and to shift the focus from nearest CEs to minimal interventions, leading to the optimisation problem min θI∈F(xF) c(θI ;x F) subj. to…”
Algorithmic recourse under imperfect causal knowledge: a probabilistic approach (2020, methodology): “In this context, recent work [22] has argued for the need of taking into account the causal structure between features to find a minimal set of actions (in the form of interventions) that guarantees recourse.”
A survey of algorithmic recourse: definitions, formulations, solutions, and prospects (2020, background): “Approaching the recourse problem from a causal perspective within the structural causal model (SCM) framework [92], actions are considered as interventions of the form 𝒂 = do ({ X : = 𝑥 F 𝑖 + 𝜃 𝑖 } 𝑖 ∈I ) ∈ A( 𝒙 F ) , and 𝒙 SCF ( 𝒂 ; 𝒙 F ) denotes the structural counterfactual of 𝒙 F had…”
FOCUS: Flexible Optimizable Counterfactual Explanations for Tree Ensembles (2019, background): “Algorithmic recourse is a line of research that is closely related to counterfactual explanations, except that these methods include the additional restriction that the resulting explanation must be actionable [16, 19, 20, 34].”
Data-centric Artificial Intelligence: A Survey (2023, unknown): “Inference data development In-distribution Data slicing [53], algorithmic recourse [120] Out-of-distribution Generating adversarial samples [165], generating samples with distribution shift [125] Prompt engineeringManual prompt engineering [206], automated prompt engineering [238]”
Counterfactual Explanations for Machine Learning: A Review (2020, methodology): “As examples, gradient based algorithms can only handle differentiable models, the algorithms based on solvers require linear or piece-wise linear models [95, 107, 64, 65, 63], some algorithms are model-speciﬁc and only work for those models like tree ensembles [104, 63, 47, 79].”
Preserving Causal Constraints in Counterfactual Explanations for Machine Learning Classifiers (2019, background): “[20, 21, 17] also focus on the role of causality towards feasible counterfactual explanations but their approach requires full knowledge of the causal model.”
Causal Inference in Natural Language Processing: Estimation, Prediction, Interpretation and Beyond (2021, background): “Causal modeling can facilitate this by making it possible to reason about the causal relationships between observed features, thus identifying minimal actions which might have down-stream effects on several features, ultimately resulting in a new prediction (Karimi et al., 2021).”