Pentagon AI office wants to match software projects with best cloud environment

The Pentagon’s top artificial intelligence office is developing a plan to weave together the department’s cloud-based software development platforms to connect AI projects with cloud environments that fit each venture’s data.

Across the Defense Department, services and components use disparate cloud and software development platforms that could be useful to other components. The Joint Artificial Intelligence Center, tasked with accelerating AI adoption at the DoD, wants to link those environments so people across the Pentagon working on AI projects can find the best cloud for their workloads.

The JAIC’s emerging idea would help connect stovepiped organizations through what the office is calling a joint common fabric of development environments.

The JAIC wants to “become the sort of player that pulls everyone together to enable more of a federation-based approach or a fabric-based approach so that workloads … can find a home in the right place,” Nand Mulchandani, chief technology officer of the JAIC, said in an interview. “Or workloads can actually migrate from one cloud to another as needed, and that people have a common identity across these different clouds so that it’s not a clunky confederation of stuff.”

The problem now, he said, is that components have potential AI workloads that they don’t have the capacity to handle, and there’s no department mechanism to connect those workloads with other clouds.

“Currently, the answer is no vacancy, I can’t accommodate you,” Mulchandani said. “Well, the fabric idea all of a sudden now says, ‘Hey, you know what? I will now recommend to you the best cloud for the workload and type you have.’”

The JAIC, established just over two years ago, understands the value of creating an enterprise cloud fabric based on its own experience developing artificial intelligence uses. Mulchandani highlighted a business operations AI project that used the DoD Comptroller’s AVANA cloud, while the JAIC’s Smart Sensor project to detect objects at the tactical edge uses a different cloud called SUNet.

The difference between the two — a business systems cloud and a war-fighting cloud —underscores how various programs have differing cloud needs, and how the JAIC can play a central role in connecting projects with proper clouds. The projects, Mulchandani said, have “different security needs, different data types, different labeling requirements, different GPU [graphics processing unit] compute requirements.”

The JAIC team working on the Joint Common Foundation, an enterprise platform of AI development tools, has shouldered part of the development of the enterprise cloud fabric.

“So as we build our Joint Common Foundation, we’re thinking about this joint common fabric,” JAIC Director Lt. Gen. Michael Groen said during a recent webinar. “So how do you stitch these development and operational environments together so that you can actually share data readily, you know, from an Army sensor into an Air Force system? So we’re expanding our approach as we start to flesh out our DevSecOps environment across the department.”

DevSecOps, an incremental software development technique, differs from other delivery options in which software is delivered all at once. The approach saves money because developers to fix issues as they build and rollout software incrementally, rather than making costly fixes after a finished product is delivered.

Not only would directing users toward the right cloud development environment ensure that data is stored in fitting environments, it also would trim costs by eliminating duplicative development efforts.

Mulchandani described the fabric concept as an “emerging concept over pizza and beer.” The AI hub is discussing the idea with components across the department. In those conversations, the center has found that there is “incredible, strong willingness” to work together, he said.

“What everyone realizes is that they specialize in certain things that they do well, but they don’t do everything. Nobody has the budget or the capability to cater to every developer type,” Mulchandani said.

The next step to connect the cloud software environments is outlining the basic processes that the JAIC can standardize to make moving projects and information between clouds easier. That includes standardizing security and identity requirements, Mulchandani said.

One of those ongoing efforts is work with the Air Force on Cloud One and Platform One, the service’s enterprise cloud and its DevSecOps software development platform, respectively. The JAIC and the Air Force are co-funding work to use software containers, which package together everything an application needs to run so it works reliably in different environments. That standardization would reduce costs, Mulchandani said.

Leave a Reply

Your email address will not be published. Required fields are marked *