Translating Intersectionality to Fair Machine Learning in Health Sciences.

Area of Science:

• Machine Learning
• Sociology
• Critical Theory

Background:

• Current machine learning fairness approaches primarily rely on performance metrics across demographic groups.
• This reliance overlooks the complex, intersecting systems of power and oppression that shape individual experiences.
• A need exists for more nuanced frameworks to address algorithmic bias.

Purpose of the Study:

• To propose a reframing of machine learning fairness.
• To shift the focus from solely performance metrics to a more contextual understanding of fairness.
• To introduce intersectionality as a theoretical lens for evaluating fairness in machine learning.

Main Methods:

• Conceptual analysis and theoretical reframing.
• Application of intersectionality, a Black feminist theoretical framework.
• Critique of traditional fairness metrics in machine learning.

Main Results:

• Demonstrates the limitations of purely metric-based fairness assessments.
• Highlights how intersectionality provides a richer understanding of fairness by considering systemic power dynamics.
• Advocates for a paradigm shift in how fairness is conceptualized and implemented in AI.

Conclusions:

• Machine learning fairness must move beyond group performance metrics.
• Intersectionality offers a vital framework for understanding and addressing algorithmic bias in a socially just manner.
• Integrating intersectional perspectives is crucial for developing equitable AI systems.