Bridging the Global Divide in AI Regulation: A Proposal for a Contextual, Coherent, and Commensurable Framework

Two major regulatory approaches are competing globally: The EU, Canada, and Brazil use a "horizontal" approach treating all AI similarly, while the US, UK, Israel, and Switzerland use "context-specific" approaches tailored to particular AI applications. This divide creates inconsistent international standards for AI governance.

Context-specific regulation may better address real-world AI risks: The paper argues that one-size-fits-all regulations miss important differences between AI use cases and their potential harms, suggesting that tailored rules for specific domains (like healthcare versus hiring) could be more effective at protecting people.

The paper proposes a new framework categorizing AI by how it's used: Rather than focusing on AI model types, the framework divides AI into three categories based on real-world impact—autonomous systems (making decisions alone), discriminative systems (making judgments about people in hiring, criminal justice, etc.), and generative systems (creating new content)—to create more targeted protections.

International AI regulation needs better coordination: The current patchwork of different national approaches creates confusion for companies and inconsistent protection for citizens; the paper proposes a "contextual, coherent, and commensurable" framework to help different countries' regulations work together more effectively.

The AI lifecycle matters for regulation: The framework distinguishes between the "learning phase" (when AI systems are being developed) and the "deployment phase" (when they're actually being used), recognizing that different risks and safeguards apply at each stage.