CoGS: Causality Constrained Counterfactual Explanations using goal-directed ASP

TL;DR

The paper tackles the opacity of predictive models by enabling causality-aware counterfactual explanations. It introduces CoGS, a framework that combines rule-based learning (FOLD‑SE) with goal-directed Answer Set Programming (s(CASP)) to generate realistic, minimal counterfactuals that respect feature dependencies. By modeling two worlds (initial negative state and goal positive state) and using direct and causal actions, CoGS automatically discovers viable recourse paths from undesired to desired outcomes, with formal guarantees of soundness. The approach is demonstrated on the German, Adult, and Car Evaluation datasets, showing that causally informed explanations are attainable in reasonable time and can reveal actionable steps for users. These results suggest CoGS can improve transparency and fairness in decision-making systems while preserving interpretability and practicality.$f:X\rightarrow\{0,1\}$, CF$_f(x)=\{\hat{x}\mid f(x)\neq f(\hat{x})\}$, $G=\{s\in S_C\mid s\not\in S_Q\}$, etc.

Abstract

Machine learning models are increasingly used in areas such as loan approvals and hiring, yet they often function as black boxes, obscuring their decision-making processes. Transparency is crucial, and individuals need explanations to understand decisions, especially for the ones not desired by the user. Ethical and legal considerations require informing individuals of changes in input attribute values (features) that could lead to a desired outcome for the user. Our work aims to generate counterfactual explanations by considering causal dependencies between features. We present the CoGS (Counterfactual Generation with s(CASP)) framework that utilizes the goal-directed Answer Set Programming system s(CASP) to generate counterfactuals from rule-based machine learning models, specifically the FOLD-SE algorithm. CoGS computes realistic and causally consistent changes to attribute values taking causal dependencies between them into account. It finds a path from an undesired outcome to a desired one using counterfactuals. We present details of the CoGS framework along with its evaluation.

Figures (1)

• Figure 1: Top: Example 1 shows how John goes from being rejected for a loan to having his loan approved. Here the bank only considers the bank balance for loan approval. John does a direct action to increase his bank balance to $\$60000$. Bottom: Example 2 shows how John goes from being rejected for a loan to having his loan approved. Here the bank considers both bank balance as well as credit score for loan approval. While the bank balance is directly altered by John, altering the credit score requires John to directly alter his debt obligations first. After clearing his debt, the causal effect of having $\$0$debt increases John's credit score to $620\ point$. This is the causal action

Theorems & Definitions (19)

• definition thmcounterdefinition: State Space (S)
• definition thmcounterdefinition: Causally Consistent State Space ($S_C$)
• definition thmcounterdefinition: Decision Consistent State Space ($S_Q$)
• definition thmcounterdefinition: Initial State ($i$)
• definition thmcounterdefinition: Actions
• definition thmcounterdefinition: Transition Function
• definition thmcounterdefinition: Counterfactual Generation (CFG) Problem
• definition thmcounterdefinition: Goal Set
• definition thmcounterdefinition: Solution Path
• definition thmcounterdefinition: CFG Implementation