Risk-based governance has become the dominant regulatory paradigm for artificial intelligence,yet existing scholarship largely treats risk categories as neutral assessments of technological harm that precede regulator...Risk-based governance has become the dominant regulatory paradigm for artificial intelligence,yet existing scholarship largely treats risk categories as neutral assessments of technological harm that precede regulatory intervention.This article challenges that assumption by arguing that risk categories function less as assessments of harm than as social architectures—institutional boundary-making devices that reorganize innovation capacity,compliance burdens,and inequality across jurisdictions and organizational types.Drawing on Beck’s risk society thesis,Jasanoff’s co-production framework,and Gieryn’s concept of boundary work,the article develops a diagnostic framework—Risk as Social Architecture—and formalizes a causal loop mechanism linking category definition,innovation channeling,compliance burden,perception lock-in,and inequality reproduction.A Cobb–Douglas formalization and three testable propositions operationalize the framework’s core variables.Applying this framework to three jurisdictional cases—the European Union’s AI Act(2021–2024),China’s Algorithmic Governance Provisions(2022–2023),and Singapore’s Model AI Governance Framework—demonstrates that identical AI systems receive fundamentally different risk classifications,with variance explained by political negotiation rather than technical properties.The article contributes to governance scholarship by reframing risk classification as a primary site of institutional power rather than a secondary policy tool.More broadly,it demonstrates how governance through categories reshapes inequality not only in artificial intelligence but across regulatory domains—including biotechnology,finance,and climate risk—wherever classification systems mediate access to markets,legitimacy,and innovation.展开更多
文摘Risk-based governance has become the dominant regulatory paradigm for artificial intelligence,yet existing scholarship largely treats risk categories as neutral assessments of technological harm that precede regulatory intervention.This article challenges that assumption by arguing that risk categories function less as assessments of harm than as social architectures—institutional boundary-making devices that reorganize innovation capacity,compliance burdens,and inequality across jurisdictions and organizational types.Drawing on Beck’s risk society thesis,Jasanoff’s co-production framework,and Gieryn’s concept of boundary work,the article develops a diagnostic framework—Risk as Social Architecture—and formalizes a causal loop mechanism linking category definition,innovation channeling,compliance burden,perception lock-in,and inequality reproduction.A Cobb–Douglas formalization and three testable propositions operationalize the framework’s core variables.Applying this framework to three jurisdictional cases—the European Union’s AI Act(2021–2024),China’s Algorithmic Governance Provisions(2022–2023),and Singapore’s Model AI Governance Framework—demonstrates that identical AI systems receive fundamentally different risk classifications,with variance explained by political negotiation rather than technical properties.The article contributes to governance scholarship by reframing risk classification as a primary site of institutional power rather than a secondary policy tool.More broadly,it demonstrates how governance through categories reshapes inequality not only in artificial intelligence but across regulatory domains—including biotechnology,finance,and climate risk—wherever classification systems mediate access to markets,legitimacy,and innovation.
