Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

From What Ifs to Insights: Counterfactuals in Causal Inference vs. Explainable AI

Galit Shmueli, David Martens, Jaewon Yoo, Travis Greene

TL;DR

The paper analyzes counterfactuals in two distinct data-science regimes: causal inference (CI) and explainable AI (XAI), proposing a unified definition that covers both input values and outcomes. It contrasts the aims, estimands, and evaluation criteria of CI (estimating $Y_i(1)-Y_i(0)$ and related averages) with XAI (finding $\tilde{\underline{x}}_i$ to alter $\hat{y}_i$) and discusses how model choice, data, and assumptions shape these counterfactuals. By surveying history and practice in both fields, the authors identify opportunities for cross-fertilization—grounding counterfactual explanations in causal theory and using explanatory counterfactuals to inform causal research, with examples in advertising, hiring, and medicine. The work argues for integrating domain knowledge with predictive models to enhance actionability, fairness, and decision-making, outlining concrete avenues such as counterfactual fairness and personalized interventions. Overall, the paper highlights a path toward combining CI and XAI to yield richer insights, robust explanations, and better-guided interventions across domains, underpinned by a common counterfactual framework that respects both outcomes and inputs.

Abstract

Counterfactuals play a pivotal role in the two distinct data science fields of causal inference (CI) and explainable artificial intelligence (XAI). While the core idea behind counterfactuals remains the same in both fields--the examination of what would have happened under different circumstances--there are key differences in how they are used and interpreted. We introduce a formal definition that encompasses the multi-faceted concept of the counterfactual in CI and XAI. We then discuss how counterfactuals are used, evaluated, generated, and operationalized in CI vs. XAI, highlighting conceptual and practical differences. By comparing and contrasting the two, we hope to identify opportunities for cross-fertilization across CI and XAI.

From What Ifs to Insights: Counterfactuals in Causal Inference vs. Explainable AI

TL;DR

The paper analyzes counterfactuals in two distinct data-science regimes: causal inference (CI) and explainable AI (XAI), proposing a unified definition that covers both input values and outcomes. It contrasts the aims, estimands, and evaluation criteria of CI (estimating and related averages) with XAI (finding to alter ) and discusses how model choice, data, and assumptions shape these counterfactuals. By surveying history and practice in both fields, the authors identify opportunities for cross-fertilization—grounding counterfactual explanations in causal theory and using explanatory counterfactuals to inform causal research, with examples in advertising, hiring, and medicine. The work argues for integrating domain knowledge with predictive models to enhance actionability, fairness, and decision-making, outlining concrete avenues such as counterfactual fairness and personalized interventions. Overall, the paper highlights a path toward combining CI and XAI to yield richer insights, robust explanations, and better-guided interventions across domains, underpinned by a common counterfactual framework that respects both outcomes and inputs.

Abstract

Counterfactuals play a pivotal role in the two distinct data science fields of causal inference (CI) and explainable artificial intelligence (XAI). While the core idea behind counterfactuals remains the same in both fields--the examination of what would have happened under different circumstances--there are key differences in how they are used and interpreted. We introduce a formal definition that encompasses the multi-faceted concept of the counterfactual in CI and XAI. We then discuss how counterfactuals are used, evaluated, generated, and operationalized in CI vs. XAI, highlighting conceptual and practical differences. By comparing and contrasting the two, we hope to identify opportunities for cross-fertilization across CI and XAI.
Paper Structure (18 sections, 1 figure, 2 tables)

This paper contains 18 sections, 1 figure, 2 tables.

Table of Contents

  1. Introduction
  2. Counterfactual Definition
  3. A Brief History of the Counterfactual Across Disciplines
  4. Causal Inference and Counterfactual Explanations in Practice
  5. When Worlds Collide
  6. Terminology and Notation
  7. Counterfactuals in CI vs. XAI
  8. Purpose: Estimate a Causal Effect vs. Explain a Predicted Outcome
  9. Causality vs. Correlation
  10. Quantity of interest: Treatment Effect vs. Feature Values
  11. Estimated Quantities: Outcomes vs. Inputs
  12. Aggregation Level: Sample vs. Individual Unit
  13. What is Modified? Model vs. Data
  14. Assumptions
  15. Performance Evaluation
  16. ...and 3 more sections

Figures (1)

  • Figure 1: Illustration of Counterfactuals in CI vs. XAI