WASHINGTON — As the Pentagon prepares to release its fiscal 2025 budget proposal, experts in artificial intelligence are hopeful that planned investments in the technology will be safeguarded from concerns that could cut into other defense accounts.
The department’s spending on AI in fiscal 2024 has been hamstrung by the lack of full-year appropriations so far, leaving money requested for such operations unapproved and inaccessible. Officials at the Pentagon’s central hub for AI have had to “cannibalize some things in order to be able to keep other things alive,” chief officer Craig Martell told reporters last month.
And while the current spending difficulties look to be rectified soon, DOD is set to release its fiscal 2025 budget request on March 11 under Defense financing caps set by the bipartisan debt limit deal, which holds down spending across the department.
The reason why AI might escape cuts is twofold: There’s a growing prioritization of AI within the department, and lawmakers who are convinced of the technology’s potential are championing the Defense Department’s expenditures in this area, even as they look to trim spending elsewhere.
“Our adversaries understand that it is one area in which they will attack us,” Sen. Mike Rounds, R-S.D., a member of the Armed Services Committee and a leading Senate voice on AI issues, said last month in an interview.
“We have to be able to not only respond to those attacks, but we’ve got to stay ahead of all of our adversaries with regard to the deployment of AI,” Rounds said. “The defense of our country is number one; it is more important than any of the other things we do, and AI is a critical part of it.”
‘Our friends’ on the Hill
The Pentagon’s request for $1.8 billion in fiscal 2024 to support the delivery and adoption of AI-enabled capabilities, workforce development and data management efforts is a fraction of the department’s $842 billion budget request. But it reflects consistent growth, from $1.1 billion for AI-related efforts in fiscal 2023 and $874 million the prior year.
According to a spring 2021 Government Accountability Office tally, the department maintained 685 unclassified military AI projects funded through its research and development and procurement accounts.
“It’s completely appropriate that the AI budget is growing fast,” Gregory Allen, the director of the Wadhwani Center for AI and Advanced Technologies at the Center for Strategic and International Studies, said in an interview. “If anything, it should perhaps be growing faster, constrained only by what is the pace at which they can actually put the money to good use.”
Defense analyst Byron Callan of Capital Alpha Partners said in a note to industry Friday that it’s “likely” that DOD will seek to preserve military personnel and readiness in its spending plan, meaning “reductions from the plan could fall squarely on investment programs.” It’s unclear how research and development dollars, where AI funding has also been allocated, will fare.
Asked about the difficulties of navigating the current funding environment, Martell, the head of the department’s 2-year-old Chief Digital and AI Office, or CDAO, pointed to lawmakers’ past financial and policy-related backing for the military’s AI endeavors.
That support, he said, has extended to the office itself, which is leading the department on establishing infrastructure, putting processes in place, enabling the creation of AI answers for the Pentagon and ensuring that the military “has the tools to be successful.”
Formerly head of machine learning at rideshare company Lyft, Martell was brought in to lead the Pentagon’s consolidation of four separate AI and data-related organizations into the office’s single structure.
“Congress has been our friends,” he told reporters during a DOD-sponsored AI symposium last month. “They have been very supportive, not just in funding, but just in the laws that they have been passing, the [National Defense Authorization Act] about what CDAO can do, so making the case about who we are and what we’re providing isn’t hard.”
Josh Wallin, a fellow in the Center for a New American Security’s defense program, predicted that DOD’s planned AI expenditures would be “relatively safe” in future budgets — especially compared with major acquisition programs, for example, where there’s a tendency to lower production quantities of certain systems to free up money.
“When we look at AI investments, and especially when we’re talking about things like research and development, that’s kind of an all-or-nothing thing,” Wallin said in an interview. “You can’t necessarily decide to pause if you’re anticipating that you’re going to have some future developments in these systems.”
Potential policy changes
Lawmakers could pursue policy changes in the national security space that could impact the Pentagon’s access and ability to leverage the technology.
While the Senate has spent months convening forums on AI that Majority Leader Charles E. Schumer, D-N.Y., has said would lay the groundwork for legislation this year, the House in recent weeks launched its own bipartisan task force. The group will examine how lawmakers can help boost innovation while potentially implementing guardrails “to safeguard the nation against current and emerging threats,” according to a February leadership press release.
Sen. Todd Young, R-Ind., who has co-led bipartisan efforts on AI in the chamber, said he and Rep. Jay Obernolte, R-Calif., who chairs the task force, have pledged to stay in touch as the Senate gets closer to considering legislation.
Young said short-term changes could involve DOD personnel tweaks that would give the military and other agencies “more flexibility when it comes to hiring and allowing some government employees to spin off into the private sector and [come] back and forth.”
Congress could also act to remove any hurdles that may be preventing DOD from identifying or procuring AI solutions, Young told reporters Feb. 29, though he didn’t provide specific examples.
California Democratic Rep. Sara Jacobs, a member of the House task force and the Armed Services Committee, said in an interview that she is interested in tackling a wide range of issues related to national security. For example, she suggested that lawmakers consider setting guardrails on the data that officials use to train their AI-enabled systems, including making decisions on how to treat servicemembers’ personally identifiable information or insights from intelligence agencies that aren’t in the public domain.
Congress also must help dictate when autonomy or AI is allowed to be used on the battlefield — and when it’s not, such as in the case of a nuclear weapon launch, she said. Jacobs said she’s looking into how involved humans need to be in AI-enabled systems’ decision-making for the fiscal 2025 defense policy bill.
“Humans are fallible, and at least we have accountability processes in place,” she said. “But it’s hard to have the same kind of accountability if a decision is made purely by an algorithm.”
