AI自主性武器系統在國際法上適用之研析

An Analysis on the Application of AI Autonomous Weapon System in International Law

自主性武器系統及其管制規範體系，向來是國際人道法領域頗為棘手的議題。軍用無人系統因其零傷亡及對複雜任務的良好適應性，已被視為是近年的革命性武器，一旦廣泛使用將明顯改變現有的武裝衝突型態，甚至改變攻伐的本質。本文試圖聚焦日益仰賴巨量數據與演算法所驅動之自主性武器系統挑戰國際法秩序，提出四項觀察及評價：（一）演算法驅動之武器運作，恐會導致「自主性」意涵之認定產生浮動性；（二）武器系統可能誤判軍事目標，乃至科學證據的可信賴性與可解釋性存疑；尤有進者（三）人工智慧參與之自主性武器系統，如何符合日內瓦公約及其附加議定書規定意旨；（四）如何釐清國家責任成立與否的解釋論疑義，也是附帶衍生難題。本文主張自主性武器系統之研發與擴散，隨著AI的問世，為各國發展科技軍事武器及國際人道法帶來巨幅轉變，其內涵似非既有國際法規範及特定常規武器公約所能完整詮釋。基此，本文主張以限制無過失責任作為國家責任的主觀歸責門檻，透過此一高度歸責門檻之設定，期能敦促國家對AI新興科技軍事武器的使用更加審慎以對。

Autonomous Weapon Systems and their regulative systems have been knotty issues in the field of international humanitarian law. In recent years, unmanned military systems have been regarded as revolutionary arms due to their abilities for zero-casualty warfare and complex tasks. If deployed extensively, the systems would not only greatly change current forms of armed conflict, but also reshape the essence of military attacks. With growing dependency on big data and algorithms, autonomous weapon systems are challenging international law and order; this article unpacks the challenges by focusing on the following aspects: 1) Algorithm-driven weapons would cause the defining of“autonomy”to fluctuate. 2) The weapon systems could potentially misjudge military targets; and the scientific evidence’s reliability and explainability remain doubtful. 3) Under this circumstance, how could AI-based autonomous weapon systems comply with the Geneva Conventions and the additional protocols? 4) Furthermore, this would lead to derivative difficulties to determine as well as to clarify state responsibilities. To sum up, the burgeoning and proliferation of autonomous weapon systems, catalyzed by AI, have profoundly impacted both domestic military technology development and international humanitarian law. However, this impact, in my view, seems unable to be appropriately interpreted and tackled by the existing international legal regulations and the Convention on Certain Conventional Weapons. Thus, in order to urge the state to exploit the emerging AI-based weapons more prudently, I argue in favor of limited strict liability as an approach to establish a set of subjective imputation criteria for state responsibility.