A new collaboration between the U.S. Navy and academia could help bring artificial intelligence to the fore in national defense.
Indiana University and the Navy recently kicked off a new, multiyear commitment. The university's School of Informatics and Computing will team with researchers at the Naval Surface Warfare Center's Crane, Indiana, division to incorporate machine learning and artificial intelligence into military sensor technology.
"We know that our potential adversaries are developing new technologies, using state of the art UAVs in order to conduct surveillance and drop small bombs," said Robert Cruise, chief scientist for the Special Warfare and Expeditionary Systems Department at NSWC Crane. "In order to counter this, we have to make our operators more intelligent, and the way to do that is to augment them with artificial intelligence."
In the initial phases of the collaboration, the Crane team will be generating realistic, military-type sensor data from a range of inputs in order to give the university researchers fodder for their testing.
"We will come up with data to train their learning machines," Cruise said. "We will take electro-optical sensors, radar sensors and passive [radio-frequency] sensors, and we will feed that multi-sensor data to their artificial intelligence. We will tell the AI what it is looking at, and the AI will learn how to associate the sensor input with the ground truth — what is actually out there generating this data."
Through machine learning, researchers will look to develop systems that can read sensor inputs more rapidly and more accurately. This could drive a range of battlefield outcomes.
"First, we would want that intelligence to be able to detect and characterize specific objects that might be present in the battlespace," Cruise said. The ability to correctly identify and categorize combatant elements in the field, without the need for human intervention, could be a big step forward.
"Then we would want that AI to characterize an entire battlespace situation, to communicate with the operator about the potential cooperative relationships that might exist between objects," Cruise said. This second phase of interpretation would potentially take battlefield intelligence to a new level. Properly trained, computers could synthesize multiple sources of information and correlate current and potential events far more effectively than any single human operator.
Finally, AI could in theory project that tactical landscape forward, performing light-speed calculations to determine the likely trajectory of events based on present readings. "We'd like the artificial intelligence to project into the future, to understand what threat is posed by this current situation," Cruise said.
Military planners say they have good reason to turn to computers in the quest for tactical intelligence. Volume is one issue: Modern battlefield sensors are producing more data than human analysts can practically ingest.
Speed also is a consideration: The hope is that with a solid grounding in AI, machines will be able to analyze information as a faster clip than is presently possible.
Finally, military leaders are thinking about persistence, the ability to take in sensor readings and consistently deliver meaningful reports, regardless of the surrounding circumstance. "The AI doesn't get tired," Cruise said. "The human operator is always subject to fatigue and may miss something that the AI will be able to pick up on."
University leaders say the Navy is a natural partner in their efforts. "Artificial intelligence, machine learning and human-computer interaction are three areas of interest to the researchers at Crane, and also areas of great strength at our school," the school's assistant dean for research and director of research collaborations, Martina Barnas, said in a news release. "We're ideally positioned to assist their efforts in this important arena."
As they look to potential applications of their work, the research partners have an eye on the sky.
"One implementation that is foremost in our minds is the threat posed by rogue unmanned aerial systems," Cruise said. "To be able to detect and respond to threat UASs, whether they are here domestically or coming out to our ships in a foreign port, that is going to require a lot of sensors, it is going to require a lot of analytic capabilities."
In the unmanned aerial war space, "one of our operators would have several sensors and perhaps several of our own unmanned platforms that could be used to counter a AUS threat — but that will require the application of AI," Cruise said. "We would have essentially a swarm of our own UASs, but there needs to be speed and comprehensiveness to the way in which we digest information if we are going to perform in that kind of scenario."
Cruise foresees larger-scale uses for AI as well, in the growing ability to think on a global, strategic scale.
As the Navy maneuvers in an increasingly complex geopolitical space, "we have to make absolutely sure each of our ships is extremely capable of defending itself. But we also have to ensure that multiple ships can work in concert, can work with each other and reinforce each other in responding to potential threats. AI could help improve that coordination," Cruise said.