The Causal Fairness Field Guide: Perspectives From Social and Formal Sciences

Area of Science:

• Interdisciplinary research at the intersection of computer science (machine learning), philosophy, sociology, and law.
• Focus on ethical AI and algorithmic fairness.

Background:

• Numerous methods for measuring causal fairness in machine learning have emerged.
• A significant gap exists in literature connecting causality-based fairness notions with social sciences.
• This disconnect hinders a comprehensive understanding of fairness in AI systems.

Purpose of the Study:

• To synthesize and explain causality-based fairness notions from both social and formal sciences.
• To bridge the gap between technical machine learning fairness metrics and social science perspectives.
• To guide practitioners in aligning fairness metrics with humanistic values and serving marginalized populations.

Main Methods:

• Compilation and exposition of causality-based fair machine learning notions.
• Explanation of mathematical backgrounds for popular causality-based fairness metrics.
• Analysis of the interplay between these notions and the fields of philosophy and law.
• Exploration of sociological and technical criticisms of current fairness approaches.

Main Results:

• Provides a foundational understanding of causality-based fairness in machine learning.
• Highlights the connections and tensions between technical fairness definitions and philosophical/legal concepts.
• Identifies key sociological and technical critiques of existing fairness methodologies.

Conclusions:

• Facilitates a deeper understanding of how causality-based fairness aligns with humanistic values.
• Offers insights for developing improved fairness metrics and methods in machine learning.
• Emphasizes the need for interdisciplinary approaches to ensure AI serves marginalized communities equitably.