# Algorithmic Recourse: from Counterfactual Explanations to Interventions

**Authors:** Amir-Hossein Karimi, Bernhard Schölkopf, Isabel Valera  
**arXiv:** [2002.06278](https://arxiv.org/abs/2002.06278)  
**Date:** 2020-02-14  
**Summary page:** https://www.summarizepaper.com/en/paper/2002.06278v4/

## Key points

- Growing need to explain how decisions are made and provide suggestions for favorable outcomes in machine learning
- Counterfactual explanations as a solution for explaining how the world would have needed to be different for a desirable outcome
- Existing research focused on generating counterfactual explanations, but overlooks using explanations as a means to help individuals take action
- Proposal to shift from recourse via nearest counterfactual explanations to recourse through minimal interventions
- Caution against relying solely on counterfactual explanations as recommended actions, advocate for actionable recommendations that consider fairness and ethics
- Importance of providing actionable recommendations alongside explanations in decision-making processes informed by machine learning algorithms

## Summary

In the field of machine learning, there is a growing need to explain how decisions are made and provide suggestions for achieving favorable outcomes. This is particularly important in consequential decision-making scenarios such as pre-trial bail and loan approval. Counterfactual explanations have emerged as a solution to meet these requirements by showing how the world would have needed to be different for a desirable outcome to occur. However, existing research has primarily focused on developing algorithms to generate counterfactual explanations across various settings, overlooking one crucial objective: using explanations as a means to help individuals take action rather than simply understand the decision-making process. To address this limitation, this work proposes a shift in paradigm from recourse via nearest counterfactual explanations to recourse through minimal interventions. By leveraging causal reasoning, the authors caution against relying solely on counterfactual explanations as a recommended set of actions for recourse. Instead, they advocate for moving the focus from explanations towards actionable recommendations that consider factors such as fairness and ethics beyond mere explanation and intervention. Overall, this paper highlights the importance of not only explaining decisions but also providing actionable recommendations for individuals seeking recourse. By shifting the focus from counterfactual explanations towards minimal interventions and considering broader societal implications, this work contributes valuable insights into improving decision-making processes informed by machine learning algorithms.