본문 바로가기
  • Home

A Study on Designing a Regulatory Sandbox for Professional-Occupation Substitutability by Artificial Intelligence (AI/LLMs) and Social-Pressure Dynamics

  • Journal of Regulation Studies
  • 2026, 35(1), pp.105~160
  • Publisher : 한국규제학회
  • Research Area : Social Science > Public Administration
  • Received : March 24, 2026
  • Accepted : June 24, 2026
  • Published : June 30, 2026

Han Byeol Yoo 1 Soung Wan Kim 2

1선문대학교
2한경국립대학교 사회통합학부

Accredited

ABSTRACT

This study examines how advances in generative artificial intelligence (AI) affect professional occupations (e.g., physicians and lawyers) through a dual lens of technical substitutability and social pressure, and proposes a Korean-style operational model for a regulatory sandbox aligned with these dynamics. Using national daily newspaper coverage from 2022 to 2026 to proxy social pressure, we combine this evidence with task-level substitution assessments generated via large language models (LLMs)—namely GPT, Claude, and Gemini— to assess technical substitutability across 14 professional occupations and estimate risk profiles for occupations with sufficient social-pressure data. To mitigate single-model dependence, we ensemble the assessments of the three models and validate inter-model measurement reliability using the intraclass correlation coefficient (ICC), Krippendorff's α, and Kendall's W. The results indicate that judicial scriveners and customs brokers exhibit high levels of technical substitutability yet receive relatively limited social attention. By contrast, physicians and lawyers fall into a high-conflict domain characterized by moderate technical substitutability but exceptionally high levels of social pressure. Drawing on capture theory and responsive regulation, we develop a four-stage regulatory sandbox tailored to occupational characteristics. In particular, we propose a differentiated strategy that (i) mandates human-in-the-loop intervention in high-risk domains while (ii) establishing an automation fast track for low-risk administrative tasks. Finally, the study derives policy implications for designing subordinate regulations under the Korean AI Basic Act, which came into effect in 2026.

Citation status

* References for papers published after 2024 are currently being built.