ABSTRACT
Potential benefits of military AI are the advancement of the efficiency, effectiveness, and safety of the operations of armed forces. However, military AI also raises concerns about moral culpability, accountability, and value alignment between AI and human stakeholders.
The use of military AI forces us to think about what values are at stake and how we want to ensure these values are accounted for (value alignment). This chapter presents a Socio-Technical Feedback loop methodology to establish and maintain the required value alignment at the levels of governance, design, development, and operation of military AI throughout its life cycle. It takes an iterative, transdisciplinary, and multistakeholder approach, tailored to the prevailing objectives, context, and AI technology. Ethical, legal, and societal aspects as well as objectives for the human-AI system (interacting human and AI agents) in (harm-related) high-risk situations are made explicit, commensurable, and auditable (including the attribution of responsibility and accountability). An illustrative scenario and an example set of methods and functions for value alignment exemplify the methodology, which will be elaborated and evaluated in future case studies.
