AI Ethics Needs Good Data

Ethics alone is insufficient for responsible AI: The paper argues that focusing only on ethical principles—a common approach in AI governance—is too limited because it depoliticizes important questions about power and who benefits from AI systems. The authors contend that "Good Data" frameworks must explicitly address political economy and power dynamics rather than relying on abstract ethical language.

Four pillars should guide AI development: Rather than adding more ethics principles, the authors propose building AI on four foundational pillars: community, rights, usability, and politics. This framework moves beyond the standard "FAT" (fairness, transparency, accountability) approach to include considerations of who has power in AI systems and how communities are involved in their creation.

AI's value should be measured by impact on marginalized groups: The paper defines AI as "good" based on whether it increases societal wellbeing and—critically—whether it increases the power and agency of "the most marginalized and disenfranchised." This shifts the evaluation focus from abstract principles to concrete outcomes for vulnerable populations.

Data protection and privacy are necessary but not sufficient: While acknowledging the importance of data protection and privacy concerns, the authors argue these alone cannot ensure responsible AI. A broader political economy perspective is needed to address underlying power imbalances in how AI systems are developed and deployed.

Current AI governance frameworks lack political accountability: The paper critiques depoliticized language in AI ethics discourse, suggesting that without explicit attention to power structures and political economy, responsible AI initiatives may inadvertently reinforce existing inequalities rather than challenge them.