인공지능과 국가배상

초록

This paper examines the challenges of ensuring effective remedies for unlawful administrative actions involving artificial intelligence within the framework of the State Liability Act. While previous debates have focused on fully automated administrative decisions and the recognition of fault, the opacity and “black-box” nature of AI equally affect cases where human officials rely on AI-generated outputs in their decision-making. Against this backdrop, the paper examines the issues that may arise when a claimant seeks to hold the state liable on the ground that administrative actions involving AI are unlawful due to errors or bias in the outcome. The analysis highlights two principal difficulties: the evidentiary burden of establishing the unlawfulness of AI-based actions, particularly in areas such as policy evaluation and risk prediction; and the challenge of construing fault understood as subjective fault of individual officials. To address these problems, the article proposes reconceptualizing fault as organizational fault or systemic deficiency of the administration. This paper further explores the possibility of extending liability under Article 5 by treating AI systems as “public facilities.” Such an approach would shift the burden of proving fault to the administrative entity, but the difficulty of proving systemic errors or algorithmic bias would remain unresolved. Regardless of the approach taken, as AI becomes routine in administrative decision-making, the criteria for determining liability under both provisions are likely to develop in a converging direction.

키워드