AI and machine learning are shaping the U.S. military’s decision-making processes.
Superman might have beaten bullets with his speed, but the U.S. Defense Department intends to do better. It has its sights set on developing cognitive technologies—computer vision, machine learning, natural language processing, for example—that are faster than the speed of human thought.
The military plans to tap machine learning and artificial intelligence (AI), in particular, to enhance decision making.
Both technologies figure prominently into the department’s third offset strategy, which pursues next-generation technologies to ensure U.S. military superiority over adversaries such as Russia and China while strengthening conventional deterrence. More grounded department goals for the technologies include shoring up networks against cyberthreats, unburdening the work force from performing tedious and repetitive tasks, and improving battlefield command.
The Army’s Communications-Electronics Research, Development and Engineering Center (CERDEC) has piggybacked on the offset’s momentum by producing the Automated Planning Framework (APF) prototype, which allows commanders and staff to analyze the military decision-making process. They can use it to evaluate maneuevers, logistics, fires, intelligence and other warfighting courses of action. “We see APF as a key technology in enabling commanders and staff to plan and issue orders faster than ever before,” says Maj. David Dilly, USA, Mission Command Battle Lab science and technology branch chief. “[It is] allowing the machine to do what it’s good at while giving humans time to concentrate on the art of command. APF is an important stepping stone for the eventual use of more general artificial intelligence where computers will help us understand, plan and fight in multidomain battle.”
AI and machine learning technologies will speed up commanders’ decision-making cycles, particularly during combat, by unloading some of their cognitive tasks, says James Hennig, acting associate director of CERDEC’s Technology, Plans and Programs Office. “We have a broad-spectrum approach to using artificial intelligence, machine learning and autonomy as tools in our tool chest, and we’re committed to growing and learning as an organization where those technologies might be applicable,” Hennig says.
One need falls within assured positioning, navigation and timing, which requires technologies for greater freedom of operations in GPS-denied environments, he says. “This is a particularly challenging problem. We look at the technology solution approach that includes machine learning and artificial intelligence, and we think that those may have the ability to help us,” Hennig says.
More broadly, researchers want AI and machine learning to aid the Army’s mission command construct, which integrates traditional warfighting functions, says Steve Mazza, a project leader in CERDEC’s Mission Command Capabilities Division. Overall, CERDEC seeks solutions for different classifications of problems. For now, the technologies produce relatively straightforward solutions. “You can come up with an algorithmic solution to these things, and every time you sort of turn the crank on the engine, you get the same answer. And every time, the answer is correct,” Mazza notes. But in the future, machine learning and AI could prove best at addressing problems for which there are no easy answers, he adds.
Machine learning, especially, requires large datasets from which machines extract useful information, Hennig says. Precisely how much data is needed to yield tangible results remains a question. “One of the things we’re trying to do is characterize how much training data you need for particular machine learning applications,” he suggests.
Applying a clear-cut set of metrics by which users objectively judge success using a feedback loop helps distinguish between machine learning and AI, Mazza explains. “When you’re using machine learning, you’re running some set of calculations in iteration, doing this many tens of thousands of times potentially, each time with an augmentation to that routine that’s provided by that feedback loop,” he says.
Researchers have divided AI into two camps: expert systems and neuronetworks, Hennig says. AI uses causal relationships to learn from behavior and training, while machine learning digests large datasets to learn how to assist humans with their jobs. CERDEC’s AI ambition is to push beyond the idea that computer systems only will perform tasks that require human intelligence, such as decision making.
Yet the thought of yielding decision-making charges to autonomous systems, especially for those in military command circles, causes some consternation. AI and robotics play an ever-increasing role in the world. But for skeptics, the technology is not to be wholly trusted. Developing and proving the technology is one battle; convincing certain users to embrace and trust that the algorithms will deliver the right answer is another, Mazza says. “This comes up in the soundtrack every single time I talk to warfighters. Automation and trust have to go hand in hand proportionally. If they don’t, then warfighters will simply refuse to use it, no matter how clever the solution. That trust is garnered through a number of avenues, not the least of which is establishing a pattern of repetitive success,” he explains.
Another way that trust is earned is through complete transparency, Hennig says. A study conducted by the Army Research Laboratory for its Robotics Collaborative Technology Alliance found that troops placed more trust in robotics when they were kept in the loop about a system’s entire decision-making process.
Trust issues aside, the overarching push to incorporate these technologies means that automating routine tasks promises to shorten by minutes, if not hours, a commander’s decision-making process while delivering the most advantageous options, says Lisa Heidelberg, chief of CERDEC’s Mission Command Capabilities Division. “We’re looking at how can we help the commander make decisions quicker,” she says. “That’s where autonomy and AI come into play. Right now, there is a lot of manual entering of data, crunching of data by various staff elements who do everything with spreadsheets and maps. Bringing in autonomy and bringing in AI right now are high priorities of the battle labs and Mission Command Center of Excellence to facilitate recommendations and predictions.”
A deluge of data has created a crippling cognitive burden, Mazza declares. “As information starts to proliferate on the battlefield, humans reach a point of saturation where they become overwhelmed by the amount of data that is available to process. It becomes increasingly difficult for people to sort out and prioritize that amount of data, making this an area that’s very ripe for the application of artificial intelligence and machine learning,” he says.
While mitigating the data dump could be reason enough to mature AI and machine learning, another factor is the automation that drives down costs. Automating simple and repetitive tasks could translate into a windfall, both in man-hours and dollars, according to an April report by Deloitte University Press. The authors estimate that automation could save up to 1.2 billion federal hours and $41.1 billion annually.
Saving money—and securing military prominence—also underpins the Defense Department’s overhaul of its lengthy acquisition process, offers Judson Walker, Brocade’s worldwide system engineering director, switching, routing and analytics. Change is needed to implement AI and machine learning and to realize an even bigger vision. “They want agility that might reach the speed of digital transformation,” he says.
In all likelihood, the department will have to embrace a crawl-walk-run scenario, beginning with a workflow automation scheme that requires a modicum of human intervention, Walker says. The Defense Department is using open source software from StackStorm, a startup that built software for automating data center operations and was bought by Brocade. Companies such as MasterCard, Netflix and Target already use the software, providing the department with an AI lessons-learned trail blazed by the commercial sector, Walker says. “In a lot of cases, it’s not a technical conversation anymore,” he says. “It now becomes a cultural conversation.”
AI and machine learning just might end up being real-life superheroes.