For Pentagon’s AI programs, It’s Time for Boots on the Ground
The Department of Defense (DoD) has spent years carefully studying, researching and experimenting with artificial intelligence (AI), alongside some operational successes like Project Maven that leveraged AI for warfighting. But the time is now for more boots on the ground. To accelerate the adoption of AI, the department needs to start slogging through the mud to scale the implementation of additional high-priority use cases.
At the end of last year, Radha Plumb, the Pentagon’s chief digital and AI officer, announced the establishment of the AI Rapid Capabilities Cell (AI RCC) to accelerate the adoption of frontier and advanced AI capabilities across the department. By targeting a specific set of use cases and targeted initiatives aimed at putting advanced AI in the hands of warfighters, the AI RCC wants to allow the department to move at greater speed to capitalize on emerging technologies, like generative AI, and invest more deeply to scale foundational technologies supporting AI adoption across the DoD.
Deploying advanced technologies on the battlefield before its adversaries has provided the United States with a strategic advantage on the global stage for decades, Plumb said. But now that superiority, and the advantage it provides, is being challenged. “AI adoption by adversaries like China, Russia, Iran and North Korea is accelerating and poses a significant national security risk,” she has said.
For its part, China is aiming to duplicate U.S. innovation and export it at prices subsidized by the Chinese government that U.S. companies can’t match. “The Chinese wisely recognize that if a country standardizes on China’s AI platform, it likely will continue to rely on that platform in the future,” Microsoft Vice Chair and President Brad Smith wrote in a recent blog post, laying out the high stakes in this global competition between different national approaches to AI.
To stay ahead of competitors, and maintain U.S. global superiority, the DoD needs to ethically incorporate AI into critical path use cases without any delay, Derek Strausbaugh, senior director of the National Security Group at Microsoft, told SIGNAL Media.
“We are living in the age of AI. You can see the investments being made all over the world, by industry and both by our allies and our geopolitical competitors. We simply need to move faster with clear intent.”
Decision Advantage From the Front Lines to the Back Office
“The AI RCC lays out a good plan but of course the benefits come through execution. It is good to see these efforts brought in under a single umbrella with the CDAO and Defense Innovation Unit collaboration,” Strausbaugh said. He noted that the use cases outlined for AI RCC include warfighting scenarios and what is called enterprise management—back office areas where lessons learned from commercial use of AI can be very quickly and directly applied: Logistics, financial management, software development and others.
“There’s a great opportunity to move out on many of these very quickly,” he observed.
“Planning is one use case DoD has highlighted for the use of generative AI capabilities—incorporating both large language models (LLMs) as well as agentic orchestration of AI workflows,” Strausbaugh said. Generative AI systems can rapidly process vast sets of external data—global news sources, weather, social media, scientific papers—to create foundational models. Grounding these foundational models in domain-specific data from internal communications and military systems used to orchestrate and track logistics, readiness and intelligence, etc. can then better fine tune the model to assist analysts in comprehensively framing the planning process within the military information domain.
“If you’re talking about fully developing a complex planning or battle management scenario with all the pertinent information at our disposal and then trying to map out all the possible responses and countermeasures, no human can reasonably be expected to do that in an expeditious fashion. A geopolitical competitor with more agile decision-making based on a superior information flow built upon the principles of speed and accuracy will consistently outmaneuver us.”
A grounded AI can provide human augmentation by generating draft plans for consideration by planners who can use their creativity and experience in assessment and ultimately build a logically sound, doctrinally sound set of plans for consideration by commanders in a fraction of the time it takes to conduct joint planning with legacy technology, said Strausbaugh.
By combining generative AI with agentic orchestration, the AI planning system can become more responsive to change
and account for changing battlespace conditions. Agents to independently assess new or changing data inputs and publish and collect potential orders could be used to move beyond planning to distributed battle management where every human planner and commander at every echelon is operating off of the most up to date plans and battlespace understanding.
It’s important to note, Strausbaugh added, that the AI system is there to augment, not replace, human decision-making. “The planner makes the decisions, and in doing so, they can also provide feedback on the AI inputs that can be used to further improve the model in training and fine tuning,” he said.
There are other generalizable enterprise or business system functions, like software development or financial management that the AI RCC is also moving ahead with.
Modern Infrastructure for AI
Some might see this shift to deployment as overdue but the long runway, all those studies and experiments, has given DoD customers a sophisticated understanding of the kind of infrastructure they need for AI operations, and it’s driving a new, highly sophisticated wave of cloud migration, according to Strausbaugh.
“Cloud 1.0 was basically [DoD customers] asking how do I reduce costs? How do I gain operational efficiency? … How do I use platform as a service offerings to avoid having to own and maintain hardware?” he said.
But more recently, the drive to cloud in the DoD has shifted from a push move (customers seeking to move to maximize savings) to a pull one (customers seeking to move to maximize value), he said.
In this second wave, DoD customers view cloud migration as a way of realizing the extraordinary potential value and utility of their data to create decision superiority and maximize warfighter effectiveness. In cloud 2.0, they are moving portfolios of data and applications into the cloud, because by doing so, they can leverage enormous advantage—using specific data sets relevant to particular DoD missions (e.g., logistics) to enhance foundational LLMs and build generative AI systems that can really increase productivity and speed up decision-making.
AI programs require extensive cloud resources, but it’s not just a matter of size or quantity, Strausbaugh said. “Customers come with sophisticated asks based very much on experimentation and some experience. How do we best organize our data to ensure it retains its value—particularly timely relevance? How do we best perform operations on that data so that it’s AI ready? And ultimately, do you have the right hardware for us to build and retrain models and run performant AI applications on?” he said.
Strausbaugh concluded that the Department of Defense is at a critical juncture with AI: “The general population’s familiarity with ChatGPT and other similar implementations has created a curiosity and willingness in the department to understand and engage in a very healthy dialogue about how to best capitalize upon these emerging technologies. I think there’s a shared consensus between industry and government that we should both automate the mundane in an effort to reduce the cognitive load on our warfighters and at the same time keep the human-in-the-loop and on-the-loop where generative and agentic AI systems can be an assistant to improving the speed, quality and accuracy of the decisions being made by our nation’s biggest asset, our people. There is an incredible opportunity and responsibility laid out before us.”
