Responsible Use of AI in Military Systems (Jan Maarten Schraagen)（Z-Library）

【One-Line Pitch】 A transdisciplinary field guide to building, deploying, and governing military AI responsibly—covering ethics, law, human factors, engineering, and policy in one coherent volume. Best for defense analysts, AI ethicists, systems engineers, policymakers, and military planners who need the full landscape rather than a single-discipline take. 【Book Arc】 - **Opening (~0%–10%)**: Frames the core problem—biased data, opaque algorithms, and eroded trust—then introduces practical implementation models: the SOTEF socio-technical feedback loop, the E-AID air-defense demonstrator, and a qualitative Risk Hierarchy for AI-enabled systems. - **Early (~10%–35%)**: Moves into engineering and ethics of human-machine teaming, including cognitive vs. volitive machines, "excarnation" of human perception and action, and design principles ensuring a distinct responsible "somebody" always remains accountable. - **Middle (~35%–50%)**: Deepens the risk-based approach—distinguishing conventional IHL-governed conflict from grey-zone and terrorist scenarios—and argues that differentiated, granular risk mitigation can build international consensus and need not be zero-sum. - **Late (~50%–80%)**: Shifts to liability, accountability of individuals and states, and human control within human-AI military teams, alongside policy questions such as multilateral security negotiations. (Excerpts do not cover these sections in detail.) - **Ending (~80%–100%)**: Focuses on "autonomy" and "meaningful human control" in weapons systems, closing the loop from design principles back to governance and international norms. (Excerpts do not cover this section in detail.) 【Key Takeaways】 - **Responsibility must be designed in, not bolted on** (Opening): The SOTEF methodology makes ethical, legal, and societal values explicit, commensurable, and auditable through iterative, multi-stakeholder feedback loops. - **Risk mitigation should be differentiated, not uniform** (Middle): The Risk Hierarchy maps military AI applications to distinct risk levels based on IHL compliance and battlefield reliability, avoiding one-size-fits-all measures. - **Human control is a design requirement, not a slogan** (Early): Meaningful human control demands responsiveness to human moral considerations and traceability of outcomes back to a responsible person. - **Cognitive and volitive machines serve different human roles** (Early): Cognitive machines fuse sensor data into situation pictures; volitive machines execute deliberate human decisions—keeping humans superior in information and options. - **Accountability requires a distinct "somebody"** (Early): Even with fully automated effectors, system design must guarantee that a specific responsible agent can always be identified. - **Rules of Engagement must shape technical design** (Middle): Synthetic legal advisors and RoE assistants can mechanize simple rules while humans retain complex judgments, building operator trust over time. - **Self-regulation can be rational, not just concessional** (Middle): States and militaries have compelling reasons to adopt self-regulatory measures because risk mitigation in AI-enabled systems is not always zero-sum. - **Ethical competence is a soft skill to be built systematically** (Early): Personality development and training should cultivate moral judgment alongside technical excellence in defense AI programs. 【Reading Tips】 - **Deep-read Sections I and II** for the practical frameworks (SOTEF, Risk Hierarchy, E-AID); these are the most concrete and immediately applicable chapters. - **Skim the philosophical and legal background** in early chapters if you already know the Responsible AI landscape—focus instead on how the authors translate principles into engineering tasks. - **Watch for the FCAS and Rapid Defense scenarios** as running case studies; they anchor abstract arguments in concrete design decisions and are worth tracing across chapters. - **Treat the five-section structure as a progression**: start with implementation models, then liability, then human control, then policy, then weapons autonomy—each builds on the prior. - **Note the two-day author workshop** mentioned in the preface; it explains the unusually coherent cross-referencing between chapters, so read them as a conversation, not isolated papers. 【Coverage Limits】 The excerpts cover Sections I and II (implementation models, risk evaluation, engineering ethics) in reasonable depth, but Sections III–V (liability, human-AI teaming, policy negotiations, weapons autonomy) are only sketched. Specific chapter titles, author names beyond the editor, and detailed legal or policy arguments are not available in the source material.