Pentagon to test how generative AI would perform in fight with China

In the next 90 days, U.S. Indo-Pacific Command, working with corporate partners and the Pentagon, will take a hard look at how generative AI tools similar to ChatGPT could help commanders make battlefield decisions more quickly against high-tech adversaries like China. 

“Our goal is to test in the INDOPACOM [area of responsibility] with some specific Navy use cases over the next 90 days, a prototype between Anduril, Palantir AI solutions to try…drive down the time down and increase the decision space for commanders,” Radha Plumb, the head of the Defense Department’s Chief Digital and Artificial Intelligence Office, or CDAO, told Defense One on Tuesday. 

A senior industry official familiar with the CDAO effort described the new project as not a single new product or tool, but a framework between a consortium of next-generation technology companies and commanders to better understand how they might use generative AI in operations—and then either building what they need or refining tools that already exist. 

“Ultimately, it's about solving whatever data, AI interoperability, or interoperability problems that exist. So it's like: ‘Here's this [existing tool] right now that we could plug and play to get forward deployed to whoever the end user is. Or, if you have a specific, unique thing that you need, we will help you build.’ There will be a whole menu of options,” the official said.

The effort marks a new chapter in CDAO’s continuing exploration of how to use generative AI for defense purposes, and builds on pilot projects to test the use of large language models in military healthcare, logistics, and writing contracts. But this new effort is also the  first use of cutting-edge generative AI tools to rapidly accelerate the time it takes for commanders to understand the battlefield and issue commands during a major conflict. 

Wednesday, outgoing White House National Security Advisor Jake Sullivan told a small group of reporters that one of the key enduring issues the United States will face is integrating AI into basically all defense operations—a point he said he has reinforced to the incoming Trump national security team. 

“I think they need to move very rapidly in the integration of artificial intelligence capabilities into not just weapons systems, but everything: the back office, you know, logistics and supply systems. All of it, basically. And I think DOD is working that, but we have to go a lot further, a lot faster.”

One of the challenges that effort will face is constraints on the amount of computing power needed to incorporate generative AI into practices and operations, whether on the battlefield or the back office. The most recent National Defense Authorization Act attempts to tackle that through a mandate to expand high-performance computing infrastructure (section 1532.)

The issue is bigger than just writing a check for more enterprise cloud hours, Plumb said. “Traditionally, the government has done compute in a couple of different ways: we buy commercial compute baked into our cloud for just general storage and analysis and analytics, not for these advanced AI applications. And we buy it more often at the unclassified level than in the classified settings.” 

But generative AI requires a lot of computing power on top of the Defense Department’s already huge needs. 

“The better models require a lot of compute resources in order to do the kinds of inference we want them to do. That's because they're pulling across a wide range of data sources. So they need to pull all those data sources in, especially if we want to tweak them via things like fine tuning or augmentation onto DOD-specific contexts,” she said. It’s a need the Defense Department is trying to adapt to meet. 

Growing collaboration among the newer breed of defense tech firms, the ones most focused on AI, will offer opportunities to do that faster—but also introduce new risks.

Will AI giants wield too much power over the Pentagon?

A secondary concern for national security leaders looking to leverage next-generation AI tools is how the Defense Department can get the most out of new offerings while also retaining power over the increasingly influential companies providing them. This is particularly challenging  as those companies converge in a new consortium, aligning themselves to present a unified front to the Defense Department when it comes to requirements, budgets, policies, and more.

Sullivan said the rise of the consortium has a number of benefits for disrupting the defense contractor community. “Part of the reason that they are aligning in the way that they are is they found trying to come in one by one individually has not stimulated change,” he said. 

The entrance of the new defense consortium will be a “net positive,” Sullivan said, so long as the White House and the Defense Department remain focused on their budgeting and oversight duties.

“If the process of acquisition and procurement, relative to the consortium of these companies, is undisciplined, that's a risk,” he said. “So having a louder voice from non-traditional parts of the defense industrial base, I think, is a net positive, so long as we don't substitute one form of dominance of sort of DOD procurement policy [under traditional defense prime contractors] for another form of dominance.”

Plumb said the way the CDAO has structured its Open Data and Applications Government-owned Interoperable Repositories program, or DAGIR, should allow the Pentagon to pick and choose among the best products and companies and even switch between them without forcing new entrants or companies to redesign products or tools from scratch. 

She described the approach as, “Let's buy the pieces we need to buy, but let's make sure the contracts and the agreement are such that we can replace any layer as needed, and also that those layers fit together so I don't have to put all of my apps on Palantir if I think an app might work better on Anduril, or might work better on some new company that I haven't even heard of.” 

But one senior Defense Department official highlighted the potential risks if the department or the White House allows some tech players to gain too much influence.

“The landscape here is not great for the government in terms of maintaining leverage,” the official said. “You've got a hugely consolidated market in the digital environment.” 

That consolidation, if it’s not handled well by the government, could create a situation where new defense contractors—perhaps unintentionally—emphasize maintaining their position and profitability over innovation. 

“Without government leverage, what you end up [with] is industry driving the solutions to warfighters, which may not be the solutions warfighters need. You get profitable contracts for industry without warfighters getting the tools and technology they need.”