자율살상무기체계(LAWS)의 법적∙윤리적 판단을 위한 학습데이터셋 구축방안

This study proposes a framework for constructing a legal and ethical training dataset for Lethal Autonomous Weapons Systems (LAWS) by drawing on methods used in the civilian AI sector. The focus is on creating a text-based, unstructured dataset, but the proposed approach can also guide the use of other data types such as voice and video. This research examines legal and ethical standards for LAWS autonomous actions, particularly under International Humanitarian Law (IHL), International Human Rights Law, and international treaties. The four main principles of IHL-distinction, proportionality, military necessity, and the Martens Clause-are explained, with dataset construction examples provided for each. Additionally, the study reviews international trends and regulatory movements surrounding LAWS. Drawing insights from AI ethics datasets, bias-checking practices by companies like IBM and Google, and cases of harmful content dataset construction in South Korea, this study proposes a direction for developing war ethics training datasets. The process for utilizing these datasets includes securing raw data and converting them into datasets for training and learning.