Commanders rarely have complete information when they make decisions, and each decision can shape how the situation evolves. The role of military education is to prepare cadets for this dynamic. Case studies are the standard tool and, while these are valuable, they usually present cadets with a fixed situation rather than one that evolves in response to what they decide. Generative AI offers a way to make that training environment adaptive. We tested what changes when the scenario responds to a cadet’s decisions instead of staying fixed on the page.
In a study at Rabdan Academy, 67 cadets completed both an AI-driven simulation and a written case study built around structurally similar command problems. Cadets rated the simulation exercise almost a full point higher on a five-point scale for situational interest, psychological presence and perceived learning. These are self-reported measures of cadet engagement and learning, and whether they translate into better decisions in the field has not yet been tested.
The Limits of the Fixed Scenario
We designed two exercises around the same kind of command problem. Each puts a junior officer in charge of a small team on a time-critical mission. Each has limited vehicles, an unreliable route and a team member who may not be fit to continue. The settings differ so that cadets cannot simply carry answers from one exercise into the other.
In the written case, cadets command an emergency response team after flash flooding. Survivors are reported at two sites: civilians, including children, at a village school, and a worker trapped under debris further away. The team has two vehicles, each better suited to one of the rescues, and flooding has lengthened the route to the school. The medical officer is fatigued, a local civilian offers an unverified route shortcut, and air support has been delayed. Cadets set out a single response plan, identify the missing information that would most change the plan, and state the main risk they are accepting, then compare their answers with a suggested answer guide.
In the AI-driven simulation, cadets lead a small team delivering medical supplies across the desert before sunset. One truck has an engine fault, the fast road is reportedly blocked, the alternative route is long and very hot, and one soldier feels unwell. The situation then evolves in response to each decision, and cadets choose from three new options each time. After five decisions, they receive an after-action review of their choices.
The simulation ran on Claude, a commercially available large language model developed by Anthropic. We developed a structured prompt to set the scenario, the number of decisions, and the rules for how the situation could change. After each choice, the AI rewrote the situation so that the next decision carried the consequences of the last one. Generative AI worked here as a scenario engine by changing the decision-making environment for the cadets in real time.
This draws on established educational theory. David Kolb’s (1984) account of experiential learning and Donald Schön’s (1983) account of reflective practice both describe judgement as something that develops through acting and then reflecting on the outcome. Analysing someone else’s decision from the outside is a different exercise. What generative AI adds is a practical way to make the scenario itself responsive – instead of scripting every possible branch in advance, the learning environment can adapt in response to a cadet’s decision.
When the Scenario Responds
The simulation changes what the cadet is being asked to do. Whereas a written case asks the cadet to plan once against a situation that stays fixed, an adaptive simulation asks, “What will you do now?” It then changes the situation in response and asks again.
Real command decisions work the same way. A commander’s first decision shapes the conditions for the second, while the information environment shifts, other actors react, and some options close as new ones open. In the simulation, the cadet’s earlier choice therefore becomes part of the next problem they must solve.
What We Found at Rabdan Academy
Sixty-seven cadets completed both exercises, and the order varied across the group so each cadet could compare their responses across formats. After each exercise, they rated it on a five-point scale. The simulation scored higher on every measure, suggesting increased engagement and learning.
Situational interest averaged 3.54 in the first scenario, rising to 4.45 for the simulation. Psychological presence, the sense of being inside the scenario rather than reading about it, increased from an average of 3.34 to 4.17. Ratings of the learning experience rose from 3.53 to 4.43. All three differences were statistically significant. It was the same pattern for individuals as well as the whole group: 55 of the 67 cadets found the simulation more interesting, and 54 rated it the better learning experience.
These are measures of engagement and self-reported learning, and, with a single cohort at one institution, this is a first test. Within those limits, the result is clear. The same cadets responded differently to two matched problems once the situation in front of them began reacting to their choices.
What This Means for Military Education
Case studies remain useful for examining historical decisions in depth and for running structured discussion across a class, but they perform a different function in the classroom from AI-generated scenarios. A case is well suited to helping cadets examine a decision closely, while a simulation is better suited to giving them repeated chances to make decisions and live with the consequences.
For military education in the UAE and the wider Gulf, AI simulations are most useful as a complement to case-based teaching, particularly where educators want cadets to experience changing information and consequences before reflecting on their decisions. They may be especially useful for exercises involving ambiguous information, competing priorities and situations in which an initial decision changes the options available later. Their use still requires controlled scenarios, clear learning objectives and instructor-led debriefing rather than unrestricted access to generative AI.
Allowing a cadet’s own decisions to shape the next problem may be where generative AI has the most to offer military decision training. The next step is to test whether those differences in the learning experience translate into better judgement over time.
