Creating and Playing Tactical Decision Games using AI

One of the recurring questions in my work -- whether I'm writing history, fiction, or building training tools -- is how people make decisions in chaos. When the information is incomplete, the stakes are high, and time is short, what does good judgment look like? And how do we teach it?

Tactical Decision Games (TDGs) have long been a common tool for militaries to use to train this kind of thinking. They're short battlefield dilemmas designed to put you in the hot seat: Here's the situation. Here's what you must accomplish. What do you do?

The challenge with TDGs is that good ones can take a lot of time to set up, and once they’re used, they’re finished. Most instructors recycle scenarios from old archives or back issues of magzines like Armor because building fresh ones takes more effort than they have time for. That bottleneck is what sparked this experiment: could a GPT create new TDGs on demand, play them out, and then provide useful feedback, without me scripting every detail ahead of time?

The Experiment I built a Custom GPT with a simple task: generate tactical dilemmas and challenge the player's decisions. I didn't feed it prewritten scenarios; I wanted to see if it could improvise while staying true to the principles of war. I did give it a lot of background material on an enemy (the Arianan’s from NATOs DATE scenario), as well as information on company and platoon level tactics. The process is straightforward -- it gives you the situation, you tell it what you'd do, and it reacts dynamically. The question wasn't whether it could mimic old TDGs, but whether AI could act as a force multiplier: making decision-training more accessible and scalable without losing the core stressors that make TDGs effective.

What Surprised Me

I expected the GPT to stumble without detailed prompts, but it produced plausible dilemmas and feedback rooted in core tactical principles. It even demonstrated flexibility I hadn't anticipated -- it could apply those same principles in unfamiliar contexts, without needing explicit instruction to do so.

The biggest concern, though, is its bias toward pleasing the participant. Rather than pointing out flaws or challenging bad calls, it often defaults to encouragement. In real TDG sessions, the friction -- the critique -- is where the learning happens. Without that, the exercise risks becoming affirmation rather than training.

Why It Matters

If AI can generate and facilitate TDGs effectively, it could break open a closed loop. Small units, study groups, even individuals could run endless decision drills without relying on curated archives or expert facilitators. That's potentially transformative -- but only if the feedback is sharp enough to build judgment rather than flatter it.

This raises bigger questions for me: Can an AI's critique ever replicate the depth of a human-led discussion? What biases does it introduce, and how do we correct for them? Does democratizing access outweigh the loss of nuance? These are the tensions I'm exploring, and they tie directly into my broader search for clarity amid chaos.

Try It

I've made the GPT public as part of this ongoing experiment. It's rough around the edges -- intentionally so -- but I'm curious how others experience it. Does it challenge you? Does it miss the mark? Where does it surprise you?

You can try it here.

If you do, I'd love to hear what you find.

A Hidden Surprise

As I kept pushing the AI in a series of games to make the challenge harder for me, a very straneg thing happened that I think is related to AI hallucinations. Instead of the typical mdoern combat scenarios that the GPT is supposed to run, the game turned into something else: my modern infantry company was sent through a “rift” to some other place, where they faced attacks from three historical forces also pulled in through rifts. We fought Mongol hordes, Greek hoplites, Napoleonic infantry brigades, and eventually even soldiers from some sci-fi future. The model kept evaluating my actions based on the principles of war, and the game was fun – but how it arrived at this solution to “making it harder” I don’t know. So don’t be afraid to push the model in weird directions!

The post Creating and Playing Tactical Decision Games using AI appeared first on Phil Halton.