Oxford Handbook on AI Ethics Book Chapter on Race and Gender

AI systems can amplify existing social inequalities: Face recognition systems show significantly higher error rates for dark-skinned women compared to light-skinned men, and machine learning tools used to assess criminal recidivism are biased against African Americans. These aren't random failures—they systematically disadvantage already marginalized groups.

Bias in AI reflects bias in training data: Natural language processing tools trained on newspaper text reproduce societal stereotypes (such as associating "computer programmer" with men and "homemaker" with women), demonstrating that AI systems learn and perpetuate the prejudices present in their source materials.

Vulnerable populations face the most AI-based decision-making: People in lower socioeconomic classes are subjected to more automated decision-making tools than wealthy individuals, meaning those most harmed by biased AI systems are also those least able to avoid or challenge them.

Technical fixes alone are insufficient: While many technical solutions to reduce bias have been proposed, addressing AI bias requires a holistic, multidisciplinary approach that considers the broader sociopolitical context—not just engineering improvements.

High-stakes applications demand urgent attention: AI is increasingly used in consequential domains like criminal justice, healthcare diagnostics, and surveillance, making bias in these systems a matter of serious social concern that requires immediate investigation and action.