Fairness for Unobserved Characteristics: Insights from Technological Impacts on Queer Communities

Sexual orientation and gender identity are largely absent from fairness research: Current algorithmic fairness work has focused primarily on observable characteristics like race and legal gender, leaving queer communities' concerns about AI systems largely unaddressed across critical areas including privacy, employment, health, and online safety.

Unobserved characteristics require fundamentally different fairness approaches: Unlike race or legal gender, sexual orientation and gender identity are often missing, unknown, or impossible to measure in datasets. This means existing fairness methods—which typically rely on identifying and protecting observed groups—cannot be directly applied to protect queer individuals from algorithmic harms.

AI creates distinct risks and opportunities for queer communities: The paper identifies both positive and negative technological impacts on queer people, suggesting that fairness solutions must be tailored to address queer-specific concerns rather than applying one-size-fits-all approaches developed for other marginalized groups.

Participatory and context-sensitive research is essential: Addressing fairness for unobserved characteristics requires moving beyond technical fixes to include inclusive research processes that involve queer communities themselves, alongside attention to privacy preservation, process fairness, and broader sociotechnical impacts.

Current fairness assumptions are too narrow: The prevailing focus on observable characteristics in algorithmic fairness represents a significant limitation that excludes important populations and concerns, highlighting the need for the field to expand its foundational assumptions about what fairness means and how to achieve it.