DoD’s AI Plans Rest on Human Talent, Cloud

Military AI

Pentagon officials are certain that two of the biggest keys to the future of national security are two technologies that, for different reasons, are shrouded in a bit of uncertainty: artificial intelligence and cloud computing.

Department of Defense CIO Dana Deasy and Lisa Porter, deputy undersecretary for research and development, went to Capitol Hill recently to talk about the importance of AI to DoD’s plans and the need to invest in its development, and how the department’s push into cloud is critical to enabling AI and data analytics.

AI is “poised to change the character of the future battlefield and the pace of threats and capabilities we must face,” Deasy said in written remarks accompanying his testimony at a hearing of the House Armed Services Committee’s Emerging Threats and Capabilities Subcommittee. But the United States needs to pour ample resources into its development, he said, noting that countries such as China and Russia are investing heavily in the technology and “threaten to erode our technological and operational advantages and destabilize the free and open international order.”

DoD is looking to drive AI development by coordinating the efforts of its Joint Artificial Intelligence Center (JAIC), which was officially established in July, and the department’s research and engineering offices. JAIC, which brought DoD’s roughly 600 AI projects under one roof, has three principle themes that run through its work, Deasy said.

The first is getting AI technologies to warfighters quickly, by identifying mission needs and quickly demonstrating functional use cases. Second is scaling AI across the department, which involves developing a foundation that enables sharing of data, standards and libraries, development of reusable tools, “and AI cloud and edge services that helps jumpstart new projects,” Deasy said. Third is attracting a skilled workforce for JAIC, which is essential to making its projects work.

“It’s all about talent,” Deasy told the subcommittee. “And this will be representative across all the services and all components.” He said DoD is looking to establish small innovation centers in addition to the central JAIC, and possibly teaming service members with outside technologists.

One challenge for DoD is the same one that faces any enterprise laying plans for AI. At the moment, AI and its subsets such as machine learning, deep learning, and neural networks create systems that are narrowly focused, excelling in one area only, and require extensive amounts of training data. “AI systems that have been trained on one type of data typically do not perform well on data that are different from the training data,” Porter told lawmakers. And because of their specialization, AI tools aren’t easily shared. “AI systems that have been optimized for commercial applications may not yield effective outcomes in military applications,” she said.

DoD’s plans for AI-powered human-machine teaming rest on systems that are more adroit at learning from fewer examples and are able to apply that learning in unexpected circumstances.

There are plenty of examples of successful AI systems, and the technology shows a lot of promise for defense-specific applications, such as analyzing drone videos and satellite imagery, and detecting cyber threats. “However, there has also been a significant amount of hype and confusion about the current state of the art,” Porter said, adding that, “we must not abandon the tenets of scientific rigor and discipline as we pursue the opportunities that AI presents.”

Porter said JAIC will focus on scaling, integration, and new data curation techniques, while the Defense Advanced Research Projects Agency’s (DARPA) multiyear AI Next campaign works with industry on AI algorithms that learn more intuitively. DARPA recently committed $2 billion to the campaign, which aims to move AI systems from specialized tools to “partners in problem-solving.”

At the center of JAIC and AI development will be DoD’s enterprise cloud, which Deasy said will provide the foundation for the agency’s data, computing resources, and AI applications.  Central to DoD’s cloud plans is the 10-year, $10 billion Joint Enterprise Defense Initiative (JEDI) acquisition, currently planned as a single award.

DoD officials see AI and cloud as linked to the department’s future operations. “Let me underscore that point,” Deasy said in his written testimony, “our enterprise approach for AI and enterprise cloud adoption via the DoD-wide Cloud Strategy are mutually reinforcing, mutually dependent undertakings.”