Bias in AI systems: integrating formal and socio-technical approaches - PMC

Summary

This review article integrates formal mathematical and socio-technical approaches to understand bias in AI systems used in high-stakes domains like healthcare, finance, criminal justice, and employment. The authors categorize bias into four interrelated families—historical/representational, selection/measurement, algorithmic/optimization, and feedback/emergent—and illustrate these through case studies in facial recognition, large language models, credit scoring, and other applications. The paper examines bias origins, manifestations, and impacts while critically evaluating current mitigation strategies, providing a comprehensive framework for understanding how AI systems can reproduce and amplify structural inequities.