Pentagon Struggling to Take Advantage of Artificial Intelligence
The Pentagon sees artificial intelligence and related technologies as key enablers of future military operations. But the Defense Department faces challenges as it seeks to acquire them.
U.S. military officials envision a wide variety of potential applications for AI including intelligence analysis and exploitation, targeting, cyber warfare, missile defense and autonomous platforms.
“All of the services are actually quite engaged in a campaign to understand where advanced artificial intelligence and autonomy can be inserted … to help defeat adversaries across the spectrum of potential conflicts that we might find ourselves in,” Vice Chairman of the Joint Chiefs of Staff Gen. Paul Selva said at a recent Senate Armed Services Committee hearing.
“It is very compelling, when one looks at the capabilities that artificial intelligence can bring to the speed and accuracy of command and control and the capabilities that advanced robotics might bring to a complex battle space,” he added.
In April, then-Deputy Secretary of Defense Robert Work established an algorithmic warfare cross-functional team at the Pentagon to advance these efforts.
“Although we have taken steps to explore the potential of artificial intelligence, big data and deep learning, I remain convinced that we need to do much more, and move much faster, across DoD to take advantage of recent and future advances in these critical areas,” he said in a memo outlining the initiative.
The Defense Department is still exploring how best to acquire artificial intelligence tools, Marine Col. Drew Cukor, the chief of the algorithmic warfare team, said at a recent military technology conference in Washington, D.C.
“I wish we could buy AI like we buy lettuce at Safeway where we can walk in and swipe a credit card and we can walk out with our head of lettuce,” he said. “This is not easy to do. There is no black box that delivers you the AI system that the government needs, at least not now. Maybe in a few years we will be there but there are key elements that have to be put together.”
The U.S. government has issued a broad agency announcement for algorithm development contracts, he noted. Vendors will be selected through a competitive selection process.
The technology must then be integrated and fielded, and once an algorithm is put on a platform it must be optimized over its lifecycle, Cukor said.
“You don’t buy AI like you buy ammunition,” he said. “There’s a very deliberative workflow process.”
However, Cukor’s team has been given rapid acquisition authorities to look at other ways of procuring the technology. It is “an opportunity for about 36 months to really explore … what are the best ways to engage industry so that we come out of this advantaging the taxpayer and advantaging the warfighter,” he said.
While much attention has been paid to advances that potential adversaries such as Russia and China are making in the field of artificial intelligence, the Pentagon is also facing stiff competition here at home as it searches for computer science talent.
“We are in an AI arms race frankly, and it’s happening in industry,” Cukor said. “The big five internet companies really are pursuing this heavily.”
He noted that Eric Schmidt, chairman of Google’s parent company Alphabet, is now referring to Google as an AI company, not a data company.
“Everyone is in this space and there’s just a ton of money” being invested in it, Cukor said.
Commercial tech giants spent about $20 billion to $30 billion last year on these technologies, according to a report by the McKinsey Global Institute titled, “Artificial Intelligence: The Next Digital Frontier?”
Companies in Silicon Valley and other technology hubs are paying top dollar to attract talented workers.
“What I notice from the government perspective as I go around is that these young software engineers are essentially making NFL salaries,” Cukor said. “The computer science departments of our major universities are wiped out because they’re all on sabbatical working for” commercial tech firms.
There are a limited number of people who have the right skill sets to work on artificial intelligence projects. That means the Defense Department is forced to compete with the high salary-paying private sector when it comes to recruiting.
“The difficulty for government is how do we entice these engineers to do government work?” Cukor asked.
Rather than engaging in an unwinnable bidding war for their services, the Pentagon should appeal to AI experts’ patriotism and their desire to work on challenging projects that are unique to the military, he said.
Meanwhile, Cukor’s team is looking to procure cutting-edge technology from the commercial industry.
That could prove challenging, said Paul Scharre, director of the technology and national security program at the Center for a New American Security, a Washington, D.C.-based think tank.
In many cases, “they are not companies that are really familiar with working with DoD, and there are some that really want nothing to do with DoD,” he said.
The Pentagon acquisition system is too slow, and the profit margins too low, for many commercial firms to want to do business with the department, he said. Those that do engage are sometimes disappointed, he added.
“In the robotics space I’ve talked to a lot of companies that have kind of a foot in the military and non-military spaces, and there’s just immense frustration in terms of working with DoD,” he said.
To be successful, Pentagon officials looking to buy AI technologies are going to have to go around the traditional acquisition system, said Scharre, who is leading the new artificial intelligence and global security initiative at CNAS.
Organizations such as the Defense Innovation Unit Experimental, known as DIUx, could help in this effort, he said. The outfit was created in 2015 by former Secretary of Defense Ash Carter. The objective was to try to bridge the divide between the Pentagon and Silicon Valley, and accelerate the procurement of new technology using rapid acquisition authorities. Additional offices have since been established in other tech hubs such as Boston and Austin, Texas.
“DIUx and other entities that could maybe work around the system and acquire technologies or adapt them in innovative ways are the right way to go” when it comes to acquiring artificial intelligence products, Scharre said. “We just need to make sure that they continue to have life and interest and bureaucratic importance under Secretary [Jim] Mattis and that we keep the pedal to the floor on finding ways to be more adaptive and embracing some of these new technologies.”
Google and other firms in Silicon Valley aren’t the only companies pursuing AI technology.
“We’re actually beginning to see more traditional defense industrial partners innovating in that space as well,” Selva said at a recent Mitchell Institute for Aerospace Studies event in Washington, D.C.
Companies such as Northrop Grumman, Lockheed Martin and Raytheon are working on AI projects and thinking about how to incorporate the technology into military platforms.
“I have more of what I would call frontline, big industrial contractors coming to talk about the value of artificial intelligence and what it might mean for weapon systems in the future and how it might be used to make current weapon systems more effective,” Selva said. “That’s really the destination we need to get to.”
But the traditional defense industry is facing some of the same challenges as the Pentagon when it comes to competing with the commercial industry, said Tom Jones, who recently served as Northrop Grumman’s vice president of advanced concepts and technologies.
Commercial companies are “assigning literally thousands of engineers to go work on incorporating machine learning and artificial intelligence into their product lines or … [making large] investments in research facilities for autonomous cars,” he said. “The amount of money that’s going in there is something we simply can’t keep pace with in the defense industry.”
Traditional military contractors are behind the curve when it comes to adopting the technology, he said.
“That’s a hard problem to solve because … a lot of the best minds are moving to these large commercial ventures,” he said. “You can’t even find them in the universities anymore. They’re being stripped out of universities to go work and develop algorithms” in places like Silicon Valley.
As it creates products for the Pentagon, the defense industry needs to find a way to leverage the large investments that commercial companies are making in areas like machine learning, he added.
Scharre said we might see startups with dual-use AI technologies partner with defense companies or be bought out by defense companies who want to acquire their technology and expertise.
Cukor sees a role for both traditional and non-traditional companies as the Pentagon pursues artificial intelligence.
“There should be no concern that we’re going to blow up the entire industrial complex here,” he said.
There will continue to be a demand for large-scale systems that defense contractors provide, he noted. “We still need that code, that baseline hardware/software infrastructure” that algorithms can be inserted into, he said.
“Algorithms are now going to be an important element of our weapon systems,” he said. “We’re going to be updating these algorithms continuously. So I see an ecosystem of vendors … [where] we still have large defense industry vendors out there that are providing the mainstay of the weapon system, and a whole other ecosystem of very fast software companies … who can bring these algorithms to our platforms.”
Dr. Hans C. Mumm