Artificial intelligence and wargaming

The Defense Department has shown interest in artificial intelligence (AI) for decades. ¹ This interest appears to have accelerated over the past several years due to its anticipated potential to transform operations and enhance capabilities across various warfighting domains. ² AI can analyze vast amounts of data, tantalizingly promising commanders a decisive “decision advantage” over their potential opponents. AI and autonomous systems are also seen as holding the promise of reduced human casualties in war, even as adversaries are seen as eroding U.S. conventional capabilities. ³

Simultaneous to this most recent renewed defense interest in AI has been the reinvigoration of United States and allied wargaming over the past decade. ⁴ Prussian Kriegsspiel, the forerunner to modern wargaming, just celebrated its 200th anniversary at the North Atlantic Treaty Organization (NATO) Wargaming Initiative 2024 in Hamburg, Germany. ⁵ This was the first international wargaming conference hosted by the German Armed Forces that anyone present could remember.

Increased attention to wargaming and AI is then the next logical step. Here, too, is an area of longstanding interest: analysts have published on the promise of adding “advanced computing” to wargaming for generations.^6,7 This current wave of interest in the intersection of wargaming and AI is simply the last manifestation of the defense community’s long-running commitment to both wargaming and developments in computational tools. In particular, two major lines of thinking and research are on (1) how to use the latest developments in AI to improve and benefit wargaming and (2) what the latest wave of AI development around the world means for military capabilities and how those AI-enabled capabilities should appear in wargames.

This special edition of Journal of Defense Modeling and Simulation (JDMS) brings together five articles that focus on wargaming and AI. It originally grew from the 2020 Connections Wargaming conference, virtually hosted by the Center for Naval Analyses (CNA). One of the special edition editors, Yuna Wong, ran the working group on wargaming and AI at that conference. These five articles were individually published online by JDMS in 2022 and focus mostly on machine learning (ML) when discussing AI. The breakout of ChatGPT and other large language models that came in 2023 means that “AI” currently is synonymous with generative AI in the popular vernacular. However, the many excellent points that the authors explore in their articles are still highly relevant and still apply as wargamers and analysts now have additional AI tools at their disposal.

Aaron Frank’s article considers the challenge of developing wargames about AI while the technology and its potential applications evolve so rapidly. To deal with this uncertainty, wargamers will need to “game AI without AI” by finding ways to mimic real or potential AI capabilities within the budget and resources available to a typical wargaming project. Frank discusses how gamers can leverage established wargaming techniques to overcome these obstacles. He also raises issues around how wargames could ultimately help human players better understand the ways AI may affect the decisions that remain under human control.

James Ryseff and Michael Bond explore the challenges facing wargamers who hope to develop AI algorithms for their games. They show how misalignments between wargaming and AI have hampered the development of effective AI algorithms within games in the past and why gamers are often tempted by the allure of “going big” to create hyper-capable, all-encompassing AI applications. They then discusses why developing smaller, cheaper, more focused AI models might prove to be a superior approach and how these types of models could be practically implemented.

The article by Paul Davis and Paul Bracken considers how advances in AI could be applied to political–military modeling and simulation applications. The authors discuss the historical challenges and hurdles that prior efforts to apply AI to these problems have faced. They then illustrate how wargamers could design an architecture to overcome these problems and develop more advanced decision aids for incorporation into wargames.

Maarten Schadd, Anne Merel Sternheim, and Olaf Visker demonstrate the challenges of designing an AI model which understands a commander’s intent. This is particularly necessary for wargaming components which hope to simulate a decision support tool within the context of a wargame. They demonstrate an approach to using formalize intent grammar to translate a commander’s written intent into a structure that AI algorithms can understand and discuss how this capability could be incorporated into future wargames.

Lastly, the team led by Danielle Tarraf experimented with implementing autonomous combat capabilities in a wargame. Her team adjusted the rules and engagement statistics for a commercial tabletop wargame to reflect remotely operated and fully autonomous vehicles, as well as vehicles with AI/ML-enabled situational awareness. The experiment demonstrated how AI/ML capabilities could be incorporated into an existing, tactical ground wargame.