AI2030 Responsible AI Tools Marketplace

Key Takeaways

One-size-fits-all AI safety frameworks are insufficient: The paper argues that formal taxonomies of harm created by industry, academia, and regulators—while well-intentioned—often fail to account for sector-specific contexts, values, and norms. Simply translating these frameworks into different sectors is inadequate for effective implementation.
"Vernacularization" offers a better approach than traditional translation: Rather than directly importing AI safety principles, the authors propose "vernacularization"—a participatory, decolonial practice that involves local communities in adapting harm taxonomies to their specific contexts, values, and discourses. This is distinct from top-down translation and better serves underresourced or high-risk sectors.
Education reveals how AI harms intensify in specific contexts: Using education as a case study, the paper demonstrates that deploying the same AI safety taxonomy reveals substantially different and intensified harms when applied to educational spaces compared to broader societal contexts, highlighting why sector-specific adaptation is critical.
Local knowledge and participation are essential for operationalization: The paper draws on anthropological theories of human rights to argue that effective AI safety requires meaningful participation from people within specific sectors who understand local norms, values, and potential vulnerabilities—not just external experts applying universal frameworks.
Vernacularization has broader applicability beyond education: The authors suggest this methodology could be generalized across multiple high-risk sectors, offering a scalable approach to making AI safety principles practically implementable in diverse organizational and cultural contexts.

Abstract

Operationalizing AI ethics and safety principles and frameworks is essential to realizing the potential benefits and mitigating potential harms caused by AI systems. To that end, actors across industry, academia, and regulatory bodies have created formal taxonomies of harm to support operationalization efforts. These include novel holistic methods that go beyond exclusive reliance on technical benchmarking. However, our paper argues that such taxonomies must also be transferred into local categories to be readily implemented in sector-specific AI safety operationalization efforts, and especially in underresourced or high-risk sectors. This is because many sectors are constituted by discourses, norms, and values that "refract" or even directly conflict with those operating in society more broadly. Drawing from emerging anthropological theories of human rights, we propose that the process of "vernacularization"--a participatory, decolonial practice distinct from doctrinary "translation" (the dominant mode of AI safety operationalization)--can help bridge this gap. To demonstrate this point, we consider the education sector, and identify precisely how vernacularizing a leading holistic taxonomy of harm leads to a clearer view of how harms AI systems may cause are substantially intensified when deployed in educational spaces. We conclude by discussing the generalizability of vernacularization as a useful AI safety methodology.

Accepted to the Proceedings of the Conference on AI Ethics and Society (AIES), 2024

Key Takeaways

One-size-fits-all AI safety frameworks are insufficient: The paper argues that formal taxonomies of harm created by industry, academia, and regulators—while well-intentioned—often fail to account for sector-specific contexts, values, and norms. Simply translating these frameworks into different sectors is inadequate for effective implementation.
"Vernacularization" offers a better approach than traditional translation: Rather than directly importing AI safety principles, the authors propose "vernacularization"—a participatory, decolonial practice that involves local communities in adapting harm taxonomies to their specific contexts, values, and discourses. This is distinct from top-down translation and better serves underresourced or high-risk sectors.
Education reveals how AI harms intensify in specific contexts: Using education as a case study, the paper demonstrates that deploying the same AI safety taxonomy reveals substantially different and intensified harms when applied to educational spaces compared to broader societal contexts, highlighting why sector-specific adaptation is critical.
Local knowledge and participation are essential for operationalization: The paper draws on anthropological theories of human rights to argue that effective AI safety requires meaningful participation from people within specific sectors who understand local norms, values, and potential vulnerabilities—not just external experts applying universal frameworks.
Vernacularization has broader applicability beyond education: The authors suggest this methodology could be generalized across multiple high-risk sectors, offering a scalable approach to making AI safety principles practically implementable in diverse organizational and cultural contexts.

Abstract

Accepted to the Proceedings of the Conference on AI Ethics and Society (AIES), 2024

Vernacularizing Taxonomies of Harm is Essential for Operationalizing Holistic AI Safety

Key Takeaways

Abstract

Paper Info

Responsible AI Topics

Links

Vernacularizing Taxonomies of Harm is Essential for Operationalizing Holistic AI Safety

Key Takeaways

Abstract

Paper Info

Responsible AI Topics

Links