The Defense Advanced Projects Research Agency’s August 2020 AlphaDogfight trials received a lot of attention, and for good reason: an AI agent created by a contractor beat a human operator in a five to nothing clean sweep during a virtual dogfight simulation.
But Col. Dan Javorsek, a DARPA program manager, told Nextgov on the latest episode of Critical Update that the Air Combat Evolution, or ACE, program, is much more than the man-versus-machine AlphaDogfight trials. The real goal is to create human-machine teams.
“The goal will be to have two contractors with their artificial intelligence algorithms competing against each other in live flight test to see who ends up coming out on top, and which contractor is able to produce the algorithm that the humans can rely and essentially, have a trusted enough level to be able to effectively win and negotiate this very dynamic within-visual range environment,” Javorsek said.
The key to accomplishing this vision has a lot more to do with psychology than might be apparent at first glance. A significant area of focus for the ACE program is trust: how to build trust between pilots and AI agents, and how to measure that trust.
For Javorsek, working with AI in fighter jets is a lot different from operating, say, a tank or a motorcycle. Instead, it’s more like working with a horse.
“When we talk about horsemanship or for anyone who’s kind of new to the process, we start with this discussion of a partnership that you have with your horse,” Javorsek said. “I think people oftentimes get into equestrian types of events thinking that their horse is basically just a biological motorcycle, and I can’t emphasize how different they are.”
That’s because working with a horse is working with another mind, Javorsek said. How to get warfighters comfortable with that concept is the question.
For more, listen to the full episode below or download from Apple Podcasts, Google Podcasts or your favorite platform.