Air Force aiming to turbocharge wargaming with AI

The Air Force Futures directorate is looking for artificial intelligence tools that could radically speed up its wargaming by orders of magnitude.

The service is conducting market research to see what industry has to offer with regard to advanced software and other modeling and simulation technologies.

“The Air Force requires a transformed, future-ready system to achieve Decision Superiority and deliver an integrated Force Design,” officials wrote in an RFI released Sunday. “This initiative marks a revolutionary shift from traditional, analog methods to a fully digitized and scientific approach. The goal is to develop a system that provides a winning force design and helps close gaps in future campaigns. Currently, the DAF faces challenges including the inability to answer critical questions about capabilities, Courses of Action (COA) analysis, or costing, due to a reliance on disconnected, outdated, and vendor-locked tools.”

Old-school U.S. military wargames can be very time-consuming, expensive, challenging to replicate and analytically insufficient, among other shortfalls, observers have noted.

Now, officials seek tools that enable simulations to be conducted at “super real-time speeds” — potentially “up to 10,000x real-time,” per the RFI.

The Air Force’s wish list includes tech that harnesses reinforcement learning to build AI agents that can realistically simulate adversary behavior and square off against individual players, teams and other AI agents.

“Technologies must simulate war as a complex adaptive system. We seek event-driven agent-based simulation across all domains, where every entity is represented as an autonomous agent that reacts to real-time events,” officials wrote.

“The system must be scalable to hundreds of users and tens of thousands of entities, and must be built to operate across classification levels (Unclassified, Secret, TS/SCI/SAP),” they added.

To ensure realism, the platform should leverage “physics-based adjudication” to provide assessments of engagements and other effects, according to the RFI. So-called “secondary effects” must also be factored in, such as how electronic warfare or cyber attacks could create “friction” and undermine situational awareness, interfere with the transmission of orders, or gum up logistics networks.

The Air Force is interested in AI capabilities that can help war planners rapidly generate, evaluate and refine courses of action “utilizing a neuro-symbolic process to rank options with risk assessments and resource tradeoffs,” and support decision-making by “pausing simulations at key inflection points and presenting outcome fields, risk corridors, and visual clusters generated from millions of possible branches for commander review,” the document noted.

Officials also envision using large language models to generate tasking orders during wargames and provide qualitative analysis.

The service is keen on solutions that integrate LLMs to create integrated tasking orders by “extracting and formatting relevant data from selected [courses of action] according to joint standards,” and enable “real-time transcription and diarization of qualitative data (e.g., commander discussions) to enable rapid analysis during and after events,” according to the RFI.

The Air Force is contemplating a “WarEngine” concept that would provide an easy-to-use platform interface and make the service’s simulation platforms and technologies more accessible to troops.

“Solutions must function as a modular, cloud-based, multi-domain application that acts as an integration hub for digital wargaming. It must integrate with or function similarly to existing service-approved tools such as AFSIM, NGTS, and JICM, using modular APIs for transparency and repeatability,” officials wrote.

Responses to the RFI are due Jan. 9.