The Military Transformations Programme at RSIS organised a series of three webinars examining emerging issues in military AI during October 2024. The webinars aimed to raise awareness of and discuss recent research on the challenges and opportunities concerning the use of AI in the military domain.
Webinar 1: Using Open-Source AI Tools to Assess Military AI Capabilities
In the first webinar of the series, held on 3 October 2024, Ritwik Gupta from the University of California, Berkeley, discussed the challenges of assessing military AI capabilities using open-source AI tools. His talk highlighted an example of how AI models could be trained on open datasets to assess their military value, and the associated risks and benefits involved.
During the Q&A session, some of the issues discussed included the difficulty in obtaining datasets with actual value in an operational setting, the feasibility of simulating military AI capabilities in war games, and the need for greater collaboration between AI researchers and policymakers.
Webinar 2: Accountability for Military AI in International LawAnna Rosalie Greipl from the Geneva Academy of International Humanitarian Law and Human Rights delivered the second webinar, held on 9 October, which focused on the underexplored issue of how to attribute responsibility for use of military AI under international law. Her talk examined the challenges of establishing the legal basis to punish war crimes committed using AI systems and evaluated the effectiveness of existing solutions.
In the Q&A segment, the discussion extended beyond lethal autonomous weapon systems to include AI-based decision support systems, particularly the associated risks of reduced levels of human judgement in legally-relevant military decision-making. Uncertainty persists over how existing international legal frameworks can address those risks, including international humanitarian law and international criminal law.
Webinar 3: Employing AI in Nuclear Command, Control, and Communication (C3) SystemsIn the final webinar of the series, which took place on 16 October 2024, Benjamin Zala from the Australian National University, examined the implications of integrating AI into nuclear command, control, and communication (C3) systems. Drawing on various examples of past nuclear near misses, his talk highlighted both the benefits and risks associated with employing AI in nuclear C3 systems and emphasised the importance of human oversight in mitigating risks while recognising the potential advantages of AI integration.
The discussion in the Q&A segment focused on the ability of human decision-makers to critically engage with and utilise AI-enabled nuclear C3 systems given that individuals often over-rely on machine outputs – a phenomenon known as automation bias. Another issue raised was the feasibility of translating nuanced human judgement into training data for AI models, highlighting the complexities involved in integrating human intuition with machine-informed human decision-making processes.
