FairXAI - A Taxonomy and Framework for Fairness and Explainability Synergy in Machine Learning

Area of Science:

• Artificial Intelligence
• Machine Learning
• AI Ethics

Background:

• Explainable AI (XAI) and fair learning have advanced independently across domains like healthcare and recidivism prediction.
• Existing research highlights how explanations improve AI transparency and trustworthiness.
• A gap exists in understanding the intersection of fairness and explainability in AI systems.

Purpose of the Study:

• To systematically review the emerging field of FairXAI, exploring the synergy between fairness and explainability.
• To propose a novel taxonomy for FairXAI, using XAI techniques to address and assess bias.
• To provide a foundational framework and practical tools for researchers and practitioners in developing fair and transparent AI.

Main Methods:

• Conducted a systematic review of existing literature on FairXAI.
• Developed a taxonomy of FairXAI approaches for bias mitigation and evaluation.
• Outlined an interaction framework and a 'FairXAI wheel' for verifying core properties.
• Identified challenges and conflicts between fairness and explainability.

Main Results:

• Established a comprehensive taxonomy for FairXAI, categorizing methods based on XAI's role in bias management.
• Proposed a structured interaction framework and a practical evaluation tool (FairXAI wheel).
• Identified key challenges and conflicts in integrating fairness and explainability.

Conclusions:

• The review consolidates the nascent field of FairXAI, offering a structured approach for research and practice.
• The proposed taxonomy and framework provide essential tools for developing responsible AI.
• Further research is needed to address identified challenges and enhance AI system accountability.