In nearly every military conflict, there’s a cat-and-mouse game that plays out between opponents, where one opponent implements a new tactic or technology and the other finds a way to counter it.
When I was a commander in Iraq, improvised explosive devices (IED) emerged as a threat to our troops. Initially crude and consisting of an artillery round buried in the ground with a long wire running to a car battery, these IEDs taught us that a “trigger man” had to be somewhere near the end of a wire. In response, we quickly created tactics to pin down and find the enemies before they could retreat. We were effective, so the enemy developed more sophisticated wireless IEDs using things like cell phones and garage-door openers. We eventually fielded jamming systems for these wireless detonators, and the insurgents went back to hard-wired IEDs, using slightly different tactics. And on and on it went.
The nation's military can expect this same type of thing to occur on future battlefields, especially around the application of effective new technologies such as artificial intelligence (AI) and machine learning (ML). However, this future cat-and-mouse game will play out much faster—at machine speed—and the military needs to be prepared.
The U.S. military is spending millions of dollars preparing for the next conflict by training algorithms that can accomplish various tasks much faster than humans can. One such task is automatic target recognition (ATR), which locates and tags enemy vehicles using feeds from cameras or other sensors on the battlefield. These ATR algorithms can be incredibly effective, but they’re only as good as the data on which they’re trained—data which often consists of footage of enemy vehicles as they currently look today, with telltale camouflage patterns, geometries, and thermal signatures.
A key maxim in war: The enemy always has a vote. Our adversaries will absolutely find ways to counter the nation's AI, just as the U.S. will find ways to counter theirs. In the ATR example above, an enterprising enemy could thwart the nation's multimillion-dollar algorithms by simply changing the visual and thermal signatures of their vehicles from those on which the algorithms were trained to something unrecognizable to a computer. That could be something easy and inexpensive to do; a few thermal tape strips, a bucket of blue paint, and some glitter ordered from Alibaba might be all it takes to render a tank invisible to the military's ATR systems.
Since the enemy will invariably find ways to counter the algorithms the U.S. is using in the field, the military must invest now in building an open-architected digital network that gives maximum flexibility to rapidly train and deploy new algorithms from the enterprise to the edge during combat. The U.S. must also work to implement ML operations (MLOps), which will help manage the lifecycle of ML engineering and deployment. Read more about how MLOps brings AI to the tactical edge.
In a future, yet inevitable, algorithmic warfare game of cat-and-mouse, the nation's decision superiority will rely upon an agile, adaptable digital architecture that allows the U.S. military to rapidly update its systems with effective new algorithms so that it can react faster than its adversaries.