Paper Number

2401

Paper Type

Short

Abstract

Artificial intelligence (AI) and deep learning techniques excel at making accurate predictions for hard problems. However, the lack of transparency in these black-box models presents significant challenges. Explainable AI (XAI) aims to tackle these challenges by developing methods that provide meaningful explanations for humans to comprehend. However, explanations offered by an XAI method may be inadequate or misleading, causing mistrust and lack of confidence in the AI technology. How to address the risk of misleading explanations has not been studied in the XAI literature. In this paper, we propose a novel method to provide risk-sensitive counterfactual explanations for the predictions of AI models. The proposed method provides robust counterfactuals to mitigate the risk of weak counterfactuals on the one hand and vigilant counterfactuals to reduce the risk of non-responsive counterfactuals on the other hand. We validate the proposed method in an empirical evaluation study using real-world data.

Comments

10-AI

Share

COinS
 
Dec 15th, 12:00 AM

Risk-Sensitive Counterfactual Explanations for AI Model Predictions

Artificial intelligence (AI) and deep learning techniques excel at making accurate predictions for hard problems. However, the lack of transparency in these black-box models presents significant challenges. Explainable AI (XAI) aims to tackle these challenges by developing methods that provide meaningful explanations for humans to comprehend. However, explanations offered by an XAI method may be inadequate or misleading, causing mistrust and lack of confidence in the AI technology. How to address the risk of misleading explanations has not been studied in the XAI literature. In this paper, we propose a novel method to provide risk-sensitive counterfactual explanations for the predictions of AI models. The proposed method provides robust counterfactuals to mitigate the risk of weak counterfactuals on the one hand and vigilant counterfactuals to reduce the risk of non-responsive counterfactuals on the other hand. We validate the proposed method in an empirical evaluation study using real-world data.

When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.