"Building a Rights-Based Approach to Regulating AI Development"

A rights-based approach to regulating AI development is proposed, which requires AI developers to provide documentation proving they have met goals to protect people's rights throughout the development and deployment process. The approach involves four pillars of AI development: data collection, model training, model evaluation, and deployment. Each pillar affects rights and should give rise to regulatory artifacts. The article focuses on the third pillar, "Model Evaluation & Analysis," and discusses the use of "Model Cards" as a transparency artifact. By prioritizing rights-oriented goals and requiring regulatory artifacts, governments can incentivize innovation while minimizing negative impacts of AI.