van Diggelen Jurriaan, Burken Christine Boshuijzen-van, Vince Julian
TNO Defence, Safety and Security, Soesterberg, Netherlands.
UNSW Canberra, Kensington, NSW, Australia.
Front Psychol. 2025 Jul 16;16:1536667. doi: 10.3389/fpsyg.2025.1536667. eCollection 2025.
This study explores the design space of meaningful human control (MHC) for military AI systems using the repertory grid technique (RGT) to capture expert perspectives across disciplines. By interviewing twelve experts from fields such as military, engineering, philosophy, and human factors, we identified key constructs related to autonomy, moral sensitivity, destructiveness, and human-machine interaction. The findings reveal both consensus and variation in experts' understanding of MHC, highlighting challenges in interdisciplinary alignment. The study identifies key variables for MHC and uses them to create a design map that guides system designers in integrating MHC concepts into AI applications. By establishing a shared vocabulary and improved elicitation methods, we aim at facilitating future discussions and research aimed at establishing and maintaining MHC.
本研究利用 repertory 网格技术(RGT)探索军事人工智能系统中有意义的人类控制(MHC)的设计空间,以获取跨学科专家的观点。通过采访来自军事、工程、哲学和人因等领域的 12 位专家,我们确定了与自主性、道德敏感性、破坏性和人机交互相关的关键构念。研究结果揭示了专家对 MHC 理解中的共识与差异,凸显了跨学科协调中的挑战。该研究确定了 MHC 的关键变量,并利用这些变量创建了一个设计地图,以指导系统设计师将 MHC 概念集成到人工智能应用中。通过建立共享词汇表和改进启发方法,我们旨在促进未来关于建立和维护 MHC 的讨论和研究。