“Technology-enhanced wargaming gives the military the data it needs to move the region’s strategic and operational decisions ahead faster,” says Jeff Roth, Booz Allen’s director of experiential analytics. AI applied in agile frameworks is one of the key accelerators, helping stakeholders plan games strategically and capture faster insights. Here are three ways AI speeds answers to tough problems in an unpredictable region where a nuanced approach is critical:
» Collaborative game planning. “We remove the black box from around AI by building a game’s reward structure in partnership with the client,” says Vince Goldsmith, Booz Allen’s technologist for game design planning in the Indo-Pacific region. “We say, ‘What do you care about most? If it’s speed of mission, how important is casualty prevention and munition preservation?’ By tweaking different parameters within the reward structure, you can get completely different approaches.”
We also create visibility into the algorithm itself. This allows stakeholders and analysts to spot and correct any biased assumptions and double-check to ensure no critical element is left out.
» Flexibility and efficiency. “We use AI every day in our personal lives—for map routing and so on—but those commercial applications aren’t cleared for military use,” says Jeff. “But enhanced technologies are increasingly available. Our team keeps a watch out for the latest tools that are accredited for a military environment.”
He points out that even basic AI features like automatic transcripts can save team members hours in notetaking and enable them to identify recurring topics, such as gray zone warfare, that represent key issues.
Our modular, open-architecture systems approach allows defense teams to proceed at the best pace for their situation, whether rapidly transforming their analytical capabilities or adding scalable tools incrementally. “We help them focus with clarity on their decision making and not get bogged down in the minutiae of creating documents or taking notes or running events,” Jeff says.
» Reinforcement learning. Unlike traditional machine learning, which relies on training algorithms with specific examples, our AI plays the game and learns by trial and error. We achieve this by applying reinforcement learning, which introduces a neural network algorithm that acts as an actual player or agent to respond to a set of situations and learn by trial and error.
“We give the algorithm a set of actions and integrate it within a specific environment. Then based on the reward structure we and the client built together, it’s going to achieve the highest reward to optimize the output,” says Vince.
“Truly embedding these reward structures actually allows our clients to create a bench of AI agents—not just one, but multiple agents specialized for different use cases,” he continues. Organizations can then apply them to show the same scenario from multiple viewpoints—that of a commander and a logistician, for example.
» Accelerated insights. “You play a wargame, maybe it's one game each day and that's it. The AI plays fast, and keeps playing overnight. And so at the end of its day, it says, ‘Well, I've done this 10,000 times and look what I found out,’” Vince explains.
This allows stakeholders to discover the most optimal choices and set them as a baseline. Then they can refine the operational plans, have the team run the algorithm again, and enhance those strategies further. Teams can also change the game between rounds to reflect additional considerations.
“After a game runs, we can say to the client, “'As an operational planner, is this what you would expect?' And maybe they’ll say, 'You know what, it didn’t take X, Y, and Z into account, so let’s build that into the reward structure and environment and let it run again,'" says Vince.