The British Navy Bought Artificial Intelligence to Make Its Sailors Better Shots

​We live in an age of semi-autonomous weapons. Nearly every major military power uses computer-controlled systems to guide some of their heaviest artillery on land and at sea. But a new artificial intelligence system by Canadian company Deep Vision could make robot guns more accurate than ever before—and they're starting by making humans better shots.

Deep Vision's system, called Dynamic Fall-of-Shot Feedback (DFOSF), is designed to be implemented on navy ships. Computer vision algorithms track where rapidly fired bullets hit the water in relation to a designated target. The system tracks both small arms fire and the intended recipient, and provides live feedback to the gunner about how accurate their shots are.

For now, the system is being developed as a training tool for the United Kingdom's Royal Navy using funding from the Center for Defence Enterprise (CDE). Small arms fire at sea is relatively common, but traditional training exercises require physical retrieval of a dummy target to gauge how well the sailors-in-training did.

But Deep Vision believes DFOSF could make training more efficient, and the system has been used on pre-recorded footage of live fire training exercises to demonstrate.

It's difficult enough for a human observer to pick out where bullets hit the water in rough seas, much less a computer program. To accomplish this, DFOSF uses a technique called data abstraction, which reduces computational complexity by compressing a ton of pixel data into a simpler, more manageable representation.

"You can think of it like each object type having its own word in a language that the machine understands," Deep Vision Chief Science Officer Michael Outhouse wrote in an email. "A car could have its own representation, and likewise, so would the impact from a bullet on the water surface."

But training could just be the beginning for DFOSF, according to Outhouse. The technology could also one day be used to help autonomous or semi-autonomous weapons hit their target.

"As assistance to fire control systems, the tracking provides guidance for firing," Outhouse explained. "The detection of where shots landed also has merit, and can potentially be used to rapidly and automatically correct for atmospheric disturbances between the firing vessel and the target."

Current generation autonomous weapons systems, such as the US Navy's Phalanx defense system for ships and the United Kingdom's Taranis drone all make numerous "decisions" on their own in terms of target detection and destruction before a human operator approves their actions. It's possible that, if implemented in these systems, DFOSF could supplement these abilities too.

There's no immediate indication that DFOSF would replace any humans in the control loop of such systems, but it could be seen as yet another step toward what anti-killer robot lobbying groups like the International Committee for Robot Arms Control (ICRAC) call "lethal autonomous weapons systems"—robots that kill people on their own, without human guidance or intervention.

The technology is still in its nascent stages, and Deep Vision's ambitions are certainly grand considering that the system hasn't even made it into training yet. But investment by the United Kingdom's army technology incubator, CDE, indicates a willingness to try the technology in the field one day soon.