The USAWC is on a multi-year endeavor to develop a wargame that reinforces our curriculum and provides students with repetitions building campaigns in crisis and conflict\u2014education that includes\u00a0practice and performance<\/a>. Pedagogical wargames are typically rigid in nature, adhering to a strict ruleset, order of events, and timeline to simplify a complex combat environment within a compressed period. This inevitably presents two challenges: one, the game abstracts concepts to such a level that the scenario becomes detached from reality; and two, the \u201coverhead\u201d of the game and its rules become the focus, overwhelming the learning opportunities for both students and instructors.<\/p>\n In February 2024, Derek Martin and John Nagl\u00a0described<\/a>\u00a0early efforts to use wargaming in the War College\u2019s\u00a0China Integrated Course<\/a>\u00a0(CIC). These rigid wargames with strict rules and onerous adjudication processes produced mixed results. During Academic Year 2026, a few seminars experimented with a simpler, tabletop-style exercise using large language models (LLMs) to speed adjudication. While looking back to the early 19th<\/sup>-century Prussian free-adjudication-style wargames, the team also looked forward by using artificial intelligence (AI) to help build detailed responses to student inputs. The result was a game that allowed students to focus on building operational plans and achieving learning outcomes rather than mastering artificial game rules. While not without its challenges, the results were universally positive, and the participating faculty deemed the game a success!<\/p>\n This problem of balancing reality with simplicity has historical precedent. Almost a century before the\u00a0Scientific American<\/em>\u00a0visit to the USAWC, Lieutenant von Reisswitz of the Prussian Guard\u00a0presented<\/a>\u00a0the Chief of the General Staff, General von Muffling, a game called\u00a0Kriegsspiel<\/em>. Though this wargame was popular in professional military settings, it met with resistance because it relied on a burdensome, rigid rule system. Several leaders in the field next developed \u201cFree Kriegsspiel,\u201dwhich simplified the system by removing much of the computational overhead. The new system asked game directors to adjudicate the outcomes of moves without depending on charts and tables, but instead using their expertise. This brought its own challenges, with some\u00a0complaining<\/a>\u00a0that \u201cthere is no one here who understands how to conduct it rightly!\u201d<\/p>\n Free-form adjudication has been widely applied, especially at the political and strategic levels. It relieves instructors and students of rigid rulesets, freeing them to focus on discussion of complex topics and learning outcomes. Yet the burden on faculty to \u201cconduct it rightly\u201d remains. LLMs can help reduce this burden, thus freeing instructors to focus on student learning.<\/p>\n Preparation and Conduct of the Game<\/strong><\/p>\n Faculty deliberately prepared students for the Pacific Strategy exercise by incorporating multiple campaign planning iterations beforehand. During the Military Strategy and Campaigning (MSC) course, students conducted a day-long design exercise developing strategies to counter aggressive Chinese posturing in the Indo-Pacific. They also wrote a paper presenting military options for US Indo-Pacific command to respond to a Chinese attack on a Philippine Navy frigate and a U.S. aircraft carrier in the Philippine Sea. Students were then required to adopt the Chinese perspective when developing options for seizing Taiwan during the China Integrated Course\u2019s lesson on the People\u2019s Liberation Army\u2019s (PLA) military capabilities.<\/p>\n Pacific Strategy progressively increases the tempo of the campaign\u2014a Chinese invasion following a Taiwanese declaration of independence\u2014over the wargame\u2019s three days. Instructors control the event through a simple digital map that displays to students only their forces and whatever opposing forces they would have realistically identified. The first day simulates a six-month crisis in which both sides attempt to improve their posture for war. The second day has two turns, each simulating a month of activity with the Chinese invasion and the United States response. The third day simulates just 72 hours of fighting in two intensive turns before concluding with a discussion and review of the previous days\u2019 events.<\/p>\n Each turn begins with the faculty issuing an intelligence report and orders from higher headquarters. The respective teams then have one hour to submit an intelligence, surveillance, and reconnaissance plan, which the faculty then use to reveal additional adversary units through an LLM-facilitated intelligence summary. Turns conclude with a 30-minute student briefing to the faculty, accompanied by a written concept of operations (usually 5-10 pages in length) and a map depicting the desired movement of units (see graphic below). Students also used an LLM to help them rapidly produce their plans, though most of the work remained manual.<\/p>\n Aided by an LLM, the faculty adjudicated the moves, updated the digital map, and gave oral and written feedback to the teams. One faculty member provided an immediate briefing of the major results while the rest of the team prepared a detailed report delivered to the students thirty to sixty minutes later. The faculty also provided guidance from higher before transitioning to the next turn.<\/p>\n Using AI to Support Adjudication<\/strong><\/p>\n The adjudication process relied on faculty discussions fleshed out by an LLM. After receiving the students\u2019 30-minute briefing of their plan, the adjudicators met to discuss major results and run the student plans through the LLM with faculty-determined parameters to generate an operations and intelligence (O&I) summary.<\/p>\n After the student teams submit their respective turn sheets, the adjudicators upload each into the LLM that examines the student plans against preloaded material from the core courses at the USAWC. This includes the\u00a0Pacific Strategy\u00a0<\/em>scenario book and each team\u2019s privileged instructions. Using an \u201cOpen then Closed<\/a>\u201d approach, the adjudicators prompt the LLM to prioritize the course and scenario material, minimizing calls to internet data.<\/p>\n When the teams\u2019 respective moves produced large-scale combat, the faculty team determined the outcome and the LLM produced battle damage assessments.<\/p><\/blockquote>\n Next, the adjudicators use the LLM to analyze and find points of intersection or tension between the two plans and recommend an outcome. In the spirit of experimentation, during this first iteration adjudicators did not give the LLM guidance on criteria to determine the outcome, allowing it to determine for itself. It weighed the turn sheets on three tensions: (1) risk v. decisiveness, (2) multi-domain integration, and (3) alliance and international engagement. Although we did not set these parameters, they generally mirror the curriculum, and future faculty teams could create and upload spreadsheets identifying what the LLM should look for in the turn sheets, better clarifying how it should regulate its decision.The adjudicators next use the LLM\u2019s\u00a0discrete event simulation<\/em><\/a>\u00a0<\/em>capability to create a realistic 6-month timeline of events resulting from the interaction of the two plans.<\/p>\n In addition to the timeline, the faculty uses the LLM to generate an O&I report for student teams that includes key events, enemy activity and other intelligence, conditions of the operational environment (weather, civilian population, etc.), and orders from national leaders. When the teams\u2019 respective moves produced large-scale combat, the faculty team determined the outcome and the LLM produced battle damage assessments. The LLM cross-referenced the order of battle in the\u00a0Pacific Strategy\u00a0<\/em>handbook and generated approximate losses for major units. For example, after the PLA team launched its assault on Taiwan, the LLM produced the following report:<\/p>\n It is important to note that the adjudicators had to \u201ctrain\u201d the LLM to produce a suitable O&I. Human oversight was necessary to adjust outcomes and force students to manage specific situations to meet training objectives\u2014like dedicating resources to recover a disabled ship. AI hallucinations occasionally occurred requiring instructor editing. The process became more efficient each turn as faculty refined the input and parameters, but human review remained a critical final step in producing the O&I.<\/p>\n Way forward<\/strong><\/p>\n Information security was a concern throughout this exercise. For the pilot, teams started using a government-provided version of\u00a0Ask Sage<\/a>, which consistently froze and lost information. As such, all groups moved to commercial systems, including\u00a0ChatGPT<\/a>,\u00a0Claude<\/a>, and\u00a0Grok<\/a>, often relying on individual instructors\u2019 paid subscriptions (i.e. ChatGPT o1). Despite being fictional scenarios, the wargames used real-world information that, while unclassified, could reveal sensitive patterns in U.S. cognitive approaches. To mitigate this risk, the military requires a robust LLM operating on a closed network. Exercise planners will benefit from building diverse faculty teams and robust capability guides to alleviate the age-old accusation that \u201cthere is no one here who can conduct it rightly!\u201d in free-adjudication games. Questions like \u201cHow far can that aircraft carrier travel in 72 hours?\u201d and \u201cHow many rounds of ammunition exist on Guam?\u201d continually surfaced. Experienced faculty can quickly find this information, but having agreed-upon parameters available to all participants will speed up the adjudication process and prevent highly competitive students from calling foul!<\/p>\n Managing moves on the map was also a challenge. Faculty experimented with several methods to geographically depict the situation to students using a \u201chome-built\u201d wargaming toolkit application. Ultimately, we identified it as an area to improve, but for which there are viable solutions with modern user interfaces. Just as advances in precision cartography\u00a0improved wargaming<\/a>\u00a0in the 19th<\/sup>\u00a0and 20th<\/sup>\u00a0centuries, so should our contemporary map provide a manageable, immersive environment.<\/p>\n While imperfect, Pacific Strategy was a step forward in the USAWC\u2019s effort to develop a wargame simple enough to work for 400 students in 24 different seminars but realistic enough to \u201cto get results and experience, to profit by the very mistakes made.\u201d Simplifying the game allowed students to focus on planning, while employing AI in adjudication empowered faculty to \u201cconduct it rightly\u201d without sacrificing learning outcomes. As in all professional military education wargames, building diverse faculty teams that practice playing before introducing the game to students remains a critical task.<\/p>\n Thomas W. Spahr<\/a><\/strong>\u00a0is the \u00a0DeSerio Chair of Theater and Strategic Intelligence at the U.S. Army War College. He is a retired colonel in the U.S. Army and holds a Ph.D. in History from The Ohio State University. He teaches courses at the Army War College on Military Campaigning and Intelligence.<\/em><\/p>\n Layton Mandle<\/a><\/strong>\u00a0is a Postdoctoral Fellow in Emerging Technology and Artificial Intelligence at the United States Army War College.<\/em><\/p>\n![\"\"](\"https:\/\/www.fie.undef.edu.ar\/ceptm\/wp-content\/uploads\/2025\/09\/image.jpeg\")
![\"\"](\"https:\/\/www.fie.undef.edu.ar\/ceptm\/wp-content\/uploads\/2025\/09\/image-1.jpeg\")
\n