AI vs. National Security: Plugging Hardware, Data and Software Leaks
The United States should approach its competition with China holistically, widening the definition of national security, concluded executives, officials and military personnel from the United States and allied countries.
During a corporate event in Washington, D.C., on Tuesday, SIGNAL Media was allowed exclusive access to closed-door discussions, with the condition that quotes and insights be unattributed, to allow decision-makers to speak candidly.
A U.S. government official explained how crippling technology sanctions against the People’s Republic of China (PRC) were researched, implemented and will continue to expand, as the United States and its allies restrict trade in semiconductors that may be used in advanced artificial intelligence (AI) models that could upgrade military operations.
“The reality is the PRC has good engineers, they have a lot of money—a lot, a lot, a lot of money—that they're throwing into this sector,” the official told attendees.
Discussions recognized the PRC’s economic and technological capability to catch up but stressed that restrictions would create a pause to allow the United States and allies a defense AI lead that could translate as a sustainable battlefield advantage.
The U.S. armed forces, as well as allies and partners, possess data that should be shared to train AI models. For example, one military expert shared that Ukraine has quality visual data on Russian targets that could be useful for friendly warfighters. Still, given the various levels of classification, how this information and these models are shared, could have deep impacts in various scenarios.
The flip side of this is that Russians also have a similar ability to recognize Western platforms for their AI-enabled systems.
Regarding both countries’ data, an analyst said that they train smart weapons with large volumes of drone video labeled by humans. A secondary consideration for these weapons are the two types currently in use by Ukraine: autonomous drones that cover the final mile toward their target and systems traveling longer distances to reach their mission objectives with little to no supervision in GPS-denied environments, flying and hitting their targets using computer vision.
In the medium term, this analyst warned of the challenges of training warfighters for a combat space where unmanned vehicles will be numerous, smart and coordinated.
Technological cohesion and innovation at speed are key attributes to deter adversaries seeking gaps in the AI and political fields, according to a Europe-based IBM executive.
During the event, several officials from the Department of Defense stressed how siloed and classified data stood in the way of AI innovation, as well as proper formats and labels.
According to a top official, the U.S. Army will present to the defense community a tool that would tackle this issue successfully. This novel tool is a first step toward creating an operational picture that integrates various sources in real time, in line with recent public announcements from the Pentagon on data use advancements. This groundbreaking tool will be ready in mid to late 2024, the official estimated.
While advanced AI-enabling technologies are placed beyond the PRC’s reach, lower-grade—or legacy—semiconductors are available to this country through indigenous manufacturing and international trade. These chips are known to find their way into Russian weapons systems through China and other adversarial actors.
Another avenue for rival actor capability enhancement is the use of cloud computing to train complex AI models, like large language models. In this case, the government requires large processing capacity suppliers to inform authorities if they receive a request compatible with such demand.
SIGNAL Media parent AFCEA International was the thought leadership partner for the 20th edition of Spade, a yearly global event organized by IBM Corporation.
