Promoting Faith in AI for National Security

Image

The Pentagon’s four-year-old Office of Chief Digital and Artificial Intelligence is working past its early pilots and limited experiments to make artificial intelligence (AI)—and now agentic AI—reach a wider body of users at a larger scale, including the intelligence community, said Andrew Mapes, acting principal deputy chief digital and artificial intelligence officer (CDAO).

“Our focus has evolved over time,” Mapes noted. “It continues to be getting the engineers as close as possible to the warfighter, but it also continues to be expanding our enterprise capability for the department.”

The CDAO’s efforts are now led by Cameron Stanley, the former national security transformation head for Amazon Web Services, and before that, the chief data officer for the Undersecretary of Defense for Intelligence and Security. Secretary Pete Hegseth’s AI Strategy for the Department of War, released in January, underpins the CDAO’s activities—in addition to executive AI policies—all of which are providing a supportive climate for AI, Mapes explained.

“Look at the AI strategy,” Mapes said. “Look at what is really driving the White House and the National AI Action Plan. And if there was a circumstance or a blocker that you understood to be in the way at one point in time that has prevented action, do not necessarily assume that is still the case. A year or two ago, the circumstances were not right. Either we did not have the capability, or the environment was not necessarily conducive for us to pursue a different AI solution or opportunity. That is not the case now.”

For instance, in the last year, the department has “very rapidly expanded” the use of the Maven Smart System, including across key warfighting networks—on NIPR, the Non-Classified Internet Protocol Router Network; SIPR, the Secret Internet Protocol Router Network; JWICS, the Joint Worldwide Intelligence Communications System; releasable environments with allies and partners; and special access programs, he stated. 

In addition, a new role in the CDAO is signaling the Pentagon’s desire to more closely align AI efforts with the intelligence community, reported the CDAO’s Robert Malpass, deputy chief digital and artificial intelligence officer for the Intelligence Mission Area. 

Malpass and Mapes both spoke at the Intelligence and National Security Alliance’s Spring Symposium in April during a panel moderated by Suzanne Heckenberg, the alliance’s president.

“My role was created with the signing of the department’s AI acceleration strategy to make sure that they are effectively managing and leveraging AI—and its transformative capability to change how humans interact with data,” Malpass said. “[It is] specifically about the relationship between the military and our partners in intelligence, across the intelligence community, who are such a valuable and rich source of data to drive some of those decisions and drive some of those mission outcomes.”

As such, the CDAO can now become a more effective consumer of data from the intelligence community. The office is also “having some important conversations” about what the relationship should be between intelligence producer and intelligence consumer, Malpass said.

“One of the biggest areas is around how we as the department react and adapt to the journey of AI that is happening both inside and outside the department, but also very importantly, within the intelligence community itself,” Malpass offered. “It is a two-way street, where investments in AI, automated processing, and automated dissemination on the intelligence community side need to be matched by investments on our side to support that, and the combatant commands’ J-2s to support the intelligence squadrons, and everybody that is on the receiving end of this information.”

Another area to consider, Malpass suggested, is that as the use of AI increases, so will the pace of decision-making. That, in turn, will require faster builds of operational plans and management of operational cycles across the entire community. “That is only going to get faster from here,” Malpass noted.

He also recommended that leaders take a stance of figuring out how to make AI and data operationalization work, as opposed to a common past starting point of, “No, this is not possible.”

“In some ways, the best way to do that is to work backwards and start with considering what the core decisions are that are being made, and the source data that goes into those decisions,” Malpass shared. “How does that data need to look, and what is the right relationship between the human decision-makers that are responsible for some of these decisions, and the information that they need to make decisions effectively.”

Malpass also promoted the success of the CDAO’s AI agent builder tool, which is available to the intelligence community and has created 100,000 agent tools.

“Agent builder—which for those of you who have not played around with it—is the ability to chain together multiple agents, multiple specific personas, for example, document summarization or generation, into a cohesive workflow that is completely low code or no code,” he clarified. “Anybody across the department can start to build out and work with advanced AI in their own context, within their own workflow.”

The CDAO’s Test and Evaluation team has been “working tirelessly” on how to evaluate the safety, trust and reliability of workflows for the users incorporating AI agents. 

“The level of trust that you should have in an AI output may look very different in a war-fighting context than an administrative context,” Malpass reasoned. “It may look very different in a situation where the decisions have to be brought up to a human, where there’s that human review, versus a workflow that may be entirely automated.”

End users across the department are clamoring for AI education and training, Malpass said, and the CDAO’s office has had to pivot to meet the surging demands, especially with generative AI. 

“It was actually an area where we had to greatly ramp up the number of classes, the frequency of the courses, and the throughput to train people at how to use these tools,” he stated. “Because the demand for not just the tools themselves, but training and advice on how to use them effectively exceeded what our initial expectations for that were, in a really positive way. It is encouraging.”

Additionally, Mapes highlighted that the Advana Platform—which started in 2021 as a bespoke financial audit tool—is now the department’s enterprise data and analytics environment, which includes a platform space for intelligence and security analytics. The CDAO is also organizing the platform into three layers, including a so-called war data platform, adding war data platform application services, and breaking out the Advana for the financial management piece.

“Around the same time as the release of the AI Acceleration Strategy from the Secretary, we re-designated Advana as the war data platform,” Mapes explained. “We are refocusing it on three areas, the tools and apps layer, putting the comptroller in the driver’s seat to really drive financial management. And then the rest of the tools and apps layer, for CDAO to rationalize what the enterprise tools and capabilities are that we’re going to continue to offer as part of that environment. Under that is the data layer that we are looking to modernize.” 

And as an attorney, Mapes stressed that federal laws and policies would guide the ethical use of AI in the department—with an emphasis on keeping the ‘human-in-the-loop’—and that AI would reshape the military to an extent. 

“AI is not going to replace people, but people that use AI are going to replace those that don’t use AI,” Mapes warned. “I think that is really kind of the mindset and approach [to have], where we are looking to enable individuals to move faster, be more efficient, and use these capabilities in ways that are beneficial.”

Lastly, the present leadership has been key to adding more AI solutions throughout the department, Mapes said. “We have had great success in being able to move at the pace that the secretary [Pete Hegseth] and Emil Michael, the chief technology officer, has set out for us,” he noted. 

“But we really are continuing to be plugged into the combatant commands to understand what their priorities are, what do they need in the field. And then also relooking at doctrine, policy across the entire DOTMLPF [doctrine, organization, training, materiel, leadership, education, personnel and facilities] spectrum,” Mapes said.

Malpass added that AI, and especially agentic AI, “offer the ability to democratize AI capabilities for individual people and small teams to process and make sense of vast quantities of data, and make sense of that data in the context of their mission and in the specific way that they need that information be processed, displayed and built out into an operational workflow.

“It is a very exciting time,” he said.