Rain Liivoja (The University of Queensland – T.C. Beirne School of Law; Asia-Pacific Institute for Law and Security) has posted Predictability in the Governance of Autonomous Weapon Systems (Australian Army Journal, forthcoming) on SSRN. Here is the abstract:
‘Predictability’ has become a central concept in multilateral efforts to regulate autonomous weapon systems (AWS). The Group of Governmental Experts on Lethal Autonomous Weapon Systems (GGE LAWS) has repeatedly considered language requiring AWS to be ‘adequately predictable’ but has failed to agree on it. National policies on military artificial intelligence (AI) largely avoid the term. This paper examines predictability as a possible normative standard for AWS and, by extension, for military AI. It traces the ‘predictability problem’ and surveys competing conceptions of predictability, including the distinction between technical and operational predictability. It then follows how, within the GGE LAWS process, the concept turned from a descriptive technical notion into a putative governance standard. The paper weighs the case for predictability, namely its grounding in existing international humanitarian law and its amenability to technical assessment, against its challenges. These are its open-endedness, the difficulty of setting criteria in advance, the military value of being unpredictable to the adversary, and its entanglement with reliability, traceability and explainability. It concludes that any future instrument should articulate what predictability substantively requires rather than adopt it as an undefined label.
To receive new posts from Legal Theory Blog by email, get a free subscription to Legal Theory Stack.
Lawrence Solum