Pages

3 December 2023

How America's New Nuclear Bomber Is Using AI

Nick Mordowanec

The United States is at the forefront of developing and incorporating artificial intelligence as part of its military, though advancements must continue to keep up with adversaries, according to experts.

AI has become rampant in private and public sectors regarding software created for defense purposes and ubiquitous programs like ChatGPT. This intersection between technology and human ingenuity has drawn praise and concern for the rapid speed at which it is occurring, spurred by how the U.S., its allies and adversaries are incorporating such intelligence into people's everyday lives.

More than 50 percent of Americans recently polled as part of a National Defense Survey conducted by the Ronald Reagan Institute said China poses the greatest risk to the United States' future, exemplifying a bigger push toward subsidizing AI through defense spending to maintain a technological advantage.

The U.S. continues to modernize in other ways, including more funding for nuclear testing and revitalizing aircraft and naval vessels. The recent unveiling of the new B-21 Raider nuclear stealth bomber, developed by Northrop Grumman, for the U.S. Air Force could be considered the next step into a new military frontier due to the "invisible" aircraft's clandestine ability to penetrate air defenses and reach targets worldwide that approximately 90 percent of current U.S. bombers are incapable of doing.
B-21's AI-driven software ushers in a new era

The B-21 is a prime example for its physical build and ability to carry nuclear warheads and for how technology is speeding up testing and design, according to Tate Nurkin, a nonresident senior fellow with the Atlantic Council's Forward Defense and Indo-Pacific Security Initiative.

Nurkin told Newsweek that digital design and engineering have sped up the design and testing of the bomber, some of which takes place digitally using AI-enabled simulations before the first aircraft are built. He called it "key" to keeping the program on schedule and enabling more rapid upgrades while improving sustainability and supply chain management.

"Open architectures also will accelerate upgrades, especially software upgrades, including AI-driven software," said Nurkin, who has studied the topic extensively. "[For] mission flexibility, it can also serve as a stealthy sensor or communications/data fusion node in addition to serving as a stealthy bomber. The data fusion area is where AI is likely to have the most impact."

Tom Longstaff, chief technology officer of the Software Engineering Institute (SEI), told Newsweek that the Air Force has traditionally been ahead of what is occurring in the public domain with programs like ChatGPT, recognizing the necessity for more maturity in AI testing and evaluation.

SEI is a federally funded research and development center supported by the DOD and operated by Carnegie Mellon University that analyzes software, cybersecurity and AI for the federal government.

Longstaff co-chaired a Department of Defense-sponsored committee of industry, government and academic experts that met several times in 2022 and 2023. They published a report titled "Test and Evaluation Challenges in Artificial Intelligence-Enabled Systems for the Department of the Air Force," which was released in September by the National Academies. The report recommends that the U.S. Department of the Air Force (DAF) invest in AI development and prioritize AI testing and evaluation.

The report was about one year in the making. The National Academies tasked the committee to delve into three major questions regarding whether the Air Force can benchmark against the private sector with regard to the testing and evaluation of its systems.

"We did look at some specific use cases and specific areas in which AI would be enabled, especially in new airborne platforms, so the B-21 would certainly be in that category," he said. "We were looking a lot at autonomy as well, so looking at the test and evaluation of autonomous flight systems, things like Loyal Wingman (a military drone with an AI control system that can carry a large weapons load), and areas that would be in partnership with manned aircraft that would basically be used in various deployment situations."

While the report doesn't specifically refer to the B-21's systems, Longstaff said that AI—which can support platforms including autonomy for flight, object detection, target identification, and counter-defense identification—incorporated on other aircraft will lend itself to this new tech.

AI systems can recognize when either the platform needs to notify the pilot or react to protect the aircraft or whether they can enable the aircraft to accomplish its mission through target recognition or flight planning.

"There are lots of areas within AI that are going to augment the human pilots' ability in these flight systems," he said.
The B-21 Raider unveiled at Northrop Grumman's Air Force Plant 42 in Palmdale, California, on December 2, 2022. AI is becoming a staple of the U.S. Air Force and other military branches.FREDERIC J. BROWN/AFP VIA GETTY IMAGES

Years-long technological push by DOD

In 2018, the Department of Defense published a report about harnessing AI for security and domestic prosperity. Today, the DOD is funding 686 AI-centric programs across the defense establishment, according to the National Academy of Sciences, Engineering, and Medicine. In 2022, there were over 600 AI-related programs across the department.

Longstaff's committee looked into whether AI preparation was in place for the Air Force to utilize, and another was how the military branch's procedures compared to the private sector. The third point of interest was specific to adversarial AI and whether or not the Air Force was prepared to test systems against adversarial AI approaches if required.

"Our findings were that the Air Force, as part of the DOD enterprise, was truly leading in a lot of the research with regard to both artificial intelligence and test evaluation with artificial intelligence," Longstaff said, alluding to positive investment among the Federally Funded Research and Development Center (FFRDC) community, but also academia, government labs, and even private sector investment.

Through various interviews and site visits, the committee identified different methods and risks for adversarial AI. The main secondary finding they discovered was that operationalizing this requires significant leadership within the Air Force, encouraging a need to advance senior leadership's AI knowledge.

"And that needs to happen pretty quickly because they are tied to platforms that are already being released and already coming out, this significant AI footprint in a lot of Department of Defense systems that are being released," Longstaff said.

Nurkin said that AI is freeing up humans to carry out more valuable tasks, such as managing missions, making decisions, and helping focus human attention on key threats, challenges, dynamics, developments and targets.

He said that the U.S., along with China, is likely leading the world in terms of the scale of defense investment in AI and the development of AI-related capabilities—from mundane back-office tasks to autonomous systems to logistics. The DOD has positioned itself for more effective development.

"To be sure, most modern militaries are investing in a range of military AI capabilities, though with the exception of China, I would say the U.S. research and development efforts are more extensive," Nurkin said. "China is very active in this area, and they would be the main competitor in terms of scale of AI investment and development, but generally most militaries are looking for ways to achieve the advantages in speed, efficiency, scale and precision that AI enables.

"However, the story is a little more nuanced than just the scale of DOD investment in research and development of AI capabilities because DOD has, to date, struggled at times to adopt these capabilities at scale."

One reason why scaling up such capabilities could be attributed to the stubbornness of the DOD itself, he said, referencing the department's culture of bureaucracy "which tends to want to move slow and is not always embracing the sort of changes in concept, culture and process required to more fully embrace AI and HMT (human-machine teaming) at scale."

"The instincts and incentives of the bureaucracy also affect how DOD buys stuff," he said. "The acquisition process for software is very slow, which not only means that it is difficult to integrate and adopt cutting-edge technologies but also creates real challenges for the commercial industry and startups that are doing really interesting stuff in AI and software development.

"Engagement and speeding up the processes of buying software and other emerging technologies from both the traditional defense industry and commercial industry is absolutely critical to improving DOD adoption of AI-enabled capabilities."
Fending off cyber attacks

The effects of AI have already been witnessed, Nurkin noted.

He referenced how in September 2021, Secretary of the Air Force Frank Kendall revealed that AI-enabled targeting was used in a "real world kill-chain operation," meaning that algorithms were used to help discern targets for military strikes.

"Currently, AI is being used to detect targets, say in satellite imagery or videos, which would take a lot longer for humans to detect," Nurkin said. "So, it is speeding up the process, but they are not autonomously selecting targets. They are recommending them and courses of action to humans who then are making the decision."

There is also Joint All-Domain Command and Control (JADC2), an ambitious DOD program developed to fuse data and unify all branches of the U.S. Armed Forces (land, air, sea, undersea, space, cyber) and then determine possible targets and recommend the best weapons—both kinetic and non-kinetic, and courses of action to engage those targets.

"The big takeaway is that AI is increasingly being tested, experimented with, and deployed across several mission sets and tasks, but there are still challenges to the incorporation at scale across DOD and the [Armed] Services," Nurkin added.

Longstaff said that the primary challenge that both the Air Force and other branches should continue to focus on are traditional cyber-attacks which can potentially alter the training of such systems before they're deployed, to maybe potentially disrupt what these systems might do once they're live in the supply chain.

"AI is not simply turning over our defense to a computer," Longstaff said. "It's not turning over the ability to defend our country and do everything to a computer that might turn against us. That just isn't the way technology works.

"We don't have the kinds of technologies that people are most afraid of in that way, and nor are we developing them in DOD systems in a way that we will be turning over that locus of control.

"So, for the time being, I think what people should realize is that machine learning is really helping us do our job better in pretty much every way that you can think about that. But it's not turning over control."

No comments:

Post a Comment