행정 영역의 인공지능 활용과 법적 통제

With the rapid development of artificial intelligence (AI) technology, its application in public administration is expanding. Since the enactment of the Framework Act on Public Administration in 2021, South Korea has established legal foundations for automated dispositions and AI-based administration through the Electronic Government Act and Administrative Procedures Act. While “AI administration” can enhance efficiency and provide prompt services, concerns arise that the opacity and potential bias in AI decision-making processes may undermine the rule of law. This paper first examines the scope of permissible AI application in administration under current law and classifies AI-based administrative procedures according to modes of decision-making involvement and types of administrative actions. Furthermore, it analyzes the legal status of AI—distinguished as physical infrastructure of administration, dispositional standards and procedures, or part of administrative processes—and examines respective legal issues under positive law. The paper identifies limitations in traditional legal control mechanisms, such as reasoning requirements for dispositions, guarantees of opportunities to submit opinions, and legality reviews, which cannot effectively function in AI-based administrative procedures. Accordingly, this paper proposes new standards for the rule of law suited to the “AI administration” era: “functional transparency” and “algorithmic due process.” Functional transparency embodies a pragmatic approach that ensures verifiability of legality and correctability of unjust outcomes instead of complete technical transparency. Algorithmic due process encompasses a procedural framework that embeds fairness and legality from the system design stage and requires human intervention at the final stage. Based on these principles, the paper suggests stage-specific control measures according to the system life cycle. The design and development stage requires enhanced ex-ante controls including algorithmic impact assessments, stakeholder participation, and expert institutional reviews. The implementation stage demands electronic notification obligations, guarantees of explanation rights, human final verification, and continuous monitoring to ensure procedural rights. The remedy stage calls for establishing online appeal channels and introducing expedited review procedures and collective remedy mechanisms. AI-based administrative procedures signify a paradigm shift in administration, yet this innovation must not weaken the principles of rule of law and the protection of fundamental rights. Based on the new standards of functional transparency and algorithmic due process, urgent legislative reforms—including amendments to the Administrative Procedures Act and related statutes—are required to establish a comprehensive control system where ex-ante and ex-post controls operate in multiple layers. To realize “human-centered AI administration,” continuous institutional improvements must follow to harmonize technological innovation with administrative accountability.