Research Summary: On the Societal Impact of Machine Learning

This research investigates the societal impact of machine learning (ML) systems, highlighting their increasing influence on consequential decisions and recommendations. It identifies the risk of discriminatory effects stemming from ML systems often developed without explicit fairness considerations. The thesis contributes methods for measuring fairness, decomposing systems to anticipate bias, and implementing interventions to mitigate algorithmic discrimination while preserving system utility.

Why it matters

The pervasive integration of machine learning into decision-making processes necessitates a deep understanding of its societal consequences, particularly concerning fairness and potential discrimination. Addressing these issues is critical for maintaining public trust, ensuring equitable outcomes, and fostering the responsible adoption of advanced technologies across industries.

Key insights

• Machine learning systems increasingly inform critical decisions and recommendations across various sectors.
• The absence of explicit fairness considerations during ML development carries a significant risk of discriminatory outcomes.
• Methodologies are developed to enhance the measurement of fairness within ML systems.
• A systematic approach is proposed for decomposing ML systems to better understand and predict bias dynamics.
• Effective interventions are identified that can reduce algorithmic discrimination while sustaining system performance and utility.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2510.23693