The US Air Force is turning to artificial intelligence to enhance its wargame.
The Air Force wants a cloud-based, AI-powered “digital sandbox” as a hub to generate and run wargames at speeds 10,000 times faster than real-time, according to a recent request for information.
The WarMatrix system aims to improve some of the problems that have hampered defensive wargames for years, such as cumbersome and labor-intensive simulations.
“Currently, DAF [Department of the Air Force] According to an Air Force RFI posted Nov. 23, the company faces challenges such as relying on disconnected, outdated, and vendor-locked tools that prevent it from answering critical questions about capabilities, course of action (COA) analysis, or cost calculations.
“This effort marks a revolutionary transition from traditional analog methods to a fully digital scientific approach,” the Air Force declared.
related
This project is ambitious.
“WarMatrix is a toolkit and orchestration environment that enables rapid scenario creation and provides a common analysis workflow with direct human judgment,” an Air Force spokesperson told Defense News.
For example, the Air Force wants “technology that can run simulations at ultra-real-time speeds, ideally speeding up the simulation process by up to 10,000 times in real time,” the RFI states. For tasks such as experimenting with force design, logistics, and battle plans, WarMatrix can perform hundreds or thousands of iterations in the time it takes to run a single tabletop game.
Additionally, the system “must be scalable to hundreds of users and tens of thousands of entities, and built to operate across classification levels (unclassified, classified, TS/SCI/SAP),” the RFI states.
It must also be compatible with existing Air Force simulations such as AFSIM and those of other military services, with an interface that transforms “high barriers to entry” and makes the DoD's “simulation arsenal accessible to all Airmen.”
“This system is designed as a collaborative wargaming tool with the government serving as the lead integrator, rather than outsourcing the design and implementation,” an Air Force spokesperson said.
The Air Force relies on AI to solve many of the problems that have plagued defense wargames, such as weak computer opponents who have difficulty with mundane tasks like marching tanks down the road. Some computer simulations required relying on “Packsters,” human players who sit at computer terminals and control the enemy forces in the game, acting as a kind of human AI.
Matthew Caffrey, former head of the Air Force Research Laboratory's wargaming division, recalled problems with the large computer models used to review wargames.
“We've heard that it's too slow to adjudicate four moves a week, so an expert panel will adjudicate the moves and the big model will continue to run after the wargame for post-game analysis,” he told Defense News.
But the Air Force envisions using AI to create sophisticated computer players that can act competently and realistically. In WarMatrix, “every entity is represented as an autonomous agent that reacts to real-time events,” the RFI states. The simulation system will incorporate effects such as jamming and cyber warfare.
Particularly welcome to busy wargame designers and referees is WarMatrix's ability to handle much of the administrative burden of wargames. For example, the Air Force wants “an LLM that can facilitate real-time transcription and diary of qualitative data (such as commander discussions) to enable rapid analysis during and after an event.”
AI could also act as an advisor to help players craft a course of action by “using neurosymbolic processes to rank options, taking into account risk assessments and resource tradeoffs,” the RFI says.
The Pentagon's history with sophisticated computer simulations is checkered, but the Air Force says it will avoid making those mistakes.
“WarMatrix blends both computational precision and human insight to ensure transparency and strategic soundness in decision-making,” the spokesperson said. “Computing and tools, data integration, human-centered design, and auditable trails address the shortcomings of the ‘black box’ megasim by making assumptions and results testable and integrated.”
“This tool is made by wargamers, for wargamers,” the spokesperson said.
Michael Peck is a correspondent for Defense News and a columnist for the Center for European Policy Analysis. He holds a master's degree in political science from Rutgers University. Find him at @Mipeck1 on X. His email is mikedefense1@gmail.com.
