Defense and intelligence agencies are working to ensure they have the networks in place to handle artificial intelligence (AI), because that’s the biggest impediment to deploying the emerging technology.
The National Geospatial-Intelligence Agency transports more data globally than any other U.S. agency, so its network needs to scale to keep pace with advances in computer vision, said Navy Vice Adm. Frank Whitworth, director of the NGA, during the 2024 DoDIIS Worldwide Conference.
The computer vision that Customs and Border Protection relies on for Global Entry benefits because the faces it processes make up 80% of the field of view, while anomalous behaviors and objects that the NGA’s AI flags make up about 0.00002% of the field of view, Whitworth said. Four combatant commands now use the NGA’s MAVEN AI for identifying targets, and the Department of Defense and its Chief Data and Artificial Intelligence Office want that number to grow for real-world operations.
Click the banner to learn more about optimizing your cloud connection.
“Our detections are way up,” Whitworth said, though he didn’t supply numbers.
The NGA employs a variety of AI models because they can be tailored to different commands’ needs, and in fact the bulk of its work is around issuing warnings with precision and accuracy as to the location of anomalous activity. Other AI models deal with navigation, helping boats stay ahead of Arctic ice floes and planes avoid obstructions.
To navigational ends, the NGA awarded a foundational research and development contract to Microsoft to create a digital twin of the world that combines imagery with textual data. The Office of Geography generated about 60 years’ worth of products in its past year of progress on the digital twin’s elevation layer, which aircraft and drones will be able to tap into.
AI Models Need to Be More Explainable
In keeping with the AI theme, another aspect of the technology that the DOD is working to address is explainability.
The Air Force wants to use AI in terrorism prediction and analysis by training models with historical data sets such as the Global Terrorism Database. But a model that determines who perpetrated an attack with 98% accuracy is still a black box if there’s no explanation for its degree of certainty, said Air Force Maj. Olusegun “SJ” Jegede, who is developing such a model.
“We have to improve transparency and trust in counterterrorism and AI models,” Jegede said.
Click the banner below to keep up with FedTech as an Insider.
This is done using techniques such as Shapley Additive Explanations, or SHAP, and Local Interpretable Model-Agnostic Explanations, or LIME, which show how different variables impact a model’s predictions. In Jegede’s models, variables include year, country, region, attack type, target type and weapon type.
Once the human running the model receives these explanations, he or she can make an informed decision about whether to accept or reject its predictions.
Jegede plans to use recurrent neural networks and long short-term memory to better capture the temporal dynamics of terrorist activity and eventually predict where attacks might occur.
A Trained Workforce Is a Happy, Useful Workforce
Emerging technologies such as AI will require the defense and intelligence communities to recruit and upskill talent with the know-how to use such tools. For that reason, agencies including the Defense Intelligence Agency have implemented a STEM pay supplement to boost hiring.
The DIA also expanded its online training portal at little cost to include more course offerings. As a result, training is up about 4,000%, and the DIA’s talent pool has increased its basic qualifications and the number of disciplines people are in, said DIA CIO Doug Cossa.
The agency is also partnering with industry to provide more opportunities to its workforce. The Education with Industry program embeds employees with industry partners for joint duty assignment credit.
“That’s been a big success, not only in terms of understanding business practices and how we can become more efficient in the way we operate but also the implementation of tools that we’ve already purchased,” Cossa said. “Oftentimes, we take advantage of only one small element or capability.”
Rising Data Volumes Can Be Addressed in Several Ways
The sheer volume of data agencies face is a challenge the defense and intelligence communities are tackling in several ways. Having accepted that they’ll never have the number of analysts they need for decision-making and incident response, agencies are turning to machine learning and AI to augment their workforces and operationalize cyber defense while implementing zero-trust security.
Data must not be squandered, which is why the Intelligence Community will release a data reference architecture for its 18 elements to connect once siloed information across their distributed ecosystem.
Many agencies have found they’ve fallen into a hybrid infrastructure model because it grants them mission flexibility they need, particularly when pushing data or applications to analysts and warfighters at the tactical edge.
To learn more about DoDIIS Worldwide, visit our conference page. You can also follow us on the social platform X at @FedTechMagazine to see behind-the-scenes moments.
