Sangchul Park, Why Korea’s AI Framework Act Requires Amendment Before Implementation - A Structural and Textual Analysis, Journal of Korea Infomation Law, Vol.28, No.3(2024), pp.3-50.
<Abstract>
Korea’s AI Framework Act establishes a horizontal regulatory framework in which a single agency enforces a uniform set of regulations for “high-impact” AI systems. While the overall regulatory burden and penalty severity have been eased compared to the EU AI Act, aligning more closely with Canada’s proposed AI and Data Act, the framework fails to account for the heterogeneity of AI use cases and risks overregulation through a one-size-fits-all framework. Furthermore, as social issues arising from AI often intersect with existing societal challenges, the horizontal framework will result overlaps and conflicts with sector-specific regulations. If the Korea Communications Commission enacts an AI User Protection Act while automated decision-making regulations under the Personal Information Protection Act and the Credit Information Act are applied concurrently, a quadruple regulatory structure could emerge, stifling the development and adoption of AI. Leading AI-innovating nations have instead adopted sector- and context-specific regulatory frameworks, where domain-specific regulators adapt existing rules to diverse use cases. Following this precedent, Korea’s current horizontal framework should transition to a more cooperative system, in which the lead agency would develop regulatory guidelines tailored to specific categories, such as rights- or safety-impacting AI use cases, while individual agencies update their respective laws accordingly. Beyond these structural issues, the AI Framework Act’s provisions are inadequately drafted. Definitions of AI and AI systems rely on anthropomorphic interpretations that are technically inaccurate, rendering them legally inappropriate and disconnected from recent developments, such as the OECD’s 2024 revisions to the definition of AI. Additionally, the roles and obligations of participants along the AI value chain are poorly defined, with significant missteps in role allocation. Regulations for high-impact AI lack clear distinctions in responsibilities based on use cases or stakeholder roles. Similarly, obligations for generative models are vague, with errors in assigning responsibilities. The scope of regulations for high-performance general-purpose models are misaligned with ongoing international AI safety discussions. Investigation and enforcement mechanisms are also misplaced. In conclusion, as the deficiencies in the AI Framework Act are too substantial to be resolved through administrative rulemaking, Korea must amend it before it comes into effect.
<Keywords>
AI Framework Act, AI Law, Horizontal Regulatory Framework, Context-Specific Regulatory Framework, AI Governance.
Sangchul Park, Why Korea’s AI Framework Act Requires Amendment Before Implementation - A Structural and Textual Analysis, Journal of Korea Infomation Law, Vol.28, No.3(2024), pp.3-50.
<Abstract>
Korea’s AI Framework Act establishes a horizontal regulatory framework in which a single agency enforces a uniform set of regulations for “high-impact” AI systems. While the overall regulatory burden and penalty severity have been eased compared to the EU AI Act, aligning more closely with Canada’s proposed AI and Data Act, the framework fails to account for the heterogeneity of AI use cases and risks overregulation through a one-size-fits-all framework. Furthermore, as social issues arising from AI often intersect with existing societal challenges, the horizontal framework will result overlaps and conflicts with sector-specific regulations. If the Korea Communications Commission enacts an AI User Protection Act while automated decision-making regulations under the Personal Information Protection Act and the Credit Information Act are applied concurrently, a quadruple regulatory structure could emerge, stifling the development and adoption of AI. Leading AI-innovating nations have instead adopted sector- and context-specific regulatory frameworks, where domain-specific regulators adapt existing rules to diverse use cases. Following this precedent, Korea’s current horizontal framework should transition to a more cooperative system, in which the lead agency would develop regulatory guidelines tailored to specific categories, such as rights- or safety-impacting AI use cases, while individual agencies update their respective laws accordingly. Beyond these structural issues, the AI Framework Act’s provisions are inadequately drafted. Definitions of AI and AI systems rely on anthropomorphic interpretations that are technically inaccurate, rendering them legally inappropriate and disconnected from recent developments, such as the OECD’s 2024 revisions to the definition of AI. Additionally, the roles and obligations of participants along the AI value chain are poorly defined, with significant missteps in role allocation. Regulations for high-impact AI lack clear distinctions in responsibilities based on use cases or stakeholder roles. Similarly, obligations for generative models are vague, with errors in assigning responsibilities. The scope of regulations for high-performance general-purpose models are misaligned with ongoing international AI safety discussions. Investigation and enforcement mechanisms are also misplaced. In conclusion, as the deficiencies in the AI Framework Act are too substantial to be resolved through administrative rulemaking, Korea must amend it before it comes into effect.
<Keywords>
AI Framework Act, AI Law, Horizontal Regulatory Framework, Context-Specific Regulatory Framework, AI Governance.