By Guy Snodgrass
Heron Systems, the winner of a two-year Defense Advanced Research Projects Agency (DARPA) competition, yesterday fielded an artificial intelligence system that managed to not only outgun a top current U.S. Air Force fighter pilot and weapons school graduate but to score a flawless victory against its human opponent, winning all five dogfighting engagements.
“It’s a giant leap,” according to Lt. Col. Justin “Glock” Mock, another weapons school graduate who also co-hosted the livestream.
The DARPA program, known as the AlphaDogfight Trials, was designed to “demonstrate the feasibility of developing effective, intelligent autonomous agents capable of defeating adversary aircraft in a dogfight.” In other words, if this were the movie Top Gun, Maverick and Goose would have just been smoked by an unmanned drone in head-to-head combat, five times in a row.
Perhaps most impressive is the timeline under which teams rapidly developed their systems.
In August 2019, eight AI companies were selected for the competition. The first trial was held three months later at Johns Hopkins University’s Applied Physics Laboratory, an exhibition match allowing teams to test the simulation environment. The second trial, held in January, was the first to rank the companies’ teams in head-to-head competition. Companies went from concept to reality in just over a year, demonstrating the ability to make tremendous gains with AI by rapidly designing, testing, and revising their algorithms.
But don’t pop the champagne just yet.
The AlphaDogfight Trials were a series of heavily scripted, simulated events. Heron Systems, along with the seven other companies who made it to the finals, had perfect knowledge of the system and computer environment, a significant advantage over the human pilot, and a departure from real-world flying. The pilot was forced to use VR goggles and non-standard equipment. In essence, the human pilot was forced to fight on the computer’s terms, not the other way around.
Cmdr. Vincent “Jell-O” Aiello, a former U.S. Navy fighter pilot, TOPGUN Instructor, and now the host of The Fighter Pilot Podcast, cautions that we shouldn’t read too much into the victory. “Humans have been proven to excel in one important area when facing off against AI—they know how to handle the type of uncertainty found in today’s combat engagements,” said Aiello. “Combat does not occur in sterile, static environments. It occurs in 3D, in real-time, where weather, your adversary, and a whole host of other factors come into play.”
Given the results from the AlphaDogfight Trials, it’s not hard to predict that the U.S. military will seek to push development of an unmanned aircraft capable of fully autonomous combat. The U.S. Navy is already experimenting with Boeing’s MQ-25 “Stingray,” an unmanned aircraft capable of launching and recovering aboard a nuclear-powered aircraft carrier to provide refueling service for manned aircraft. Plans are already underway to begin expanding that envelope to include persistent surveillance and possibly combat capabilities in the future.
Boeing is also producing Australia’s “Loyal Wingman” unmanned aircraft, part of an initial three-aircraft development package designed to explore aspects of human-machine teaming for aerial combat scenarios. The U.S. Air Force sees the development of AI-enabled aircraft as a leap forward in capability, “akin to the development of stealth aircraft and precision-guided munitions,” according to a published 2019 artificial intelligence strategy.
While there are years to go before the U.S. can expect to field a fully autonomous combat aircraft, DARPA’s AlphaDogfight Trials demonstrate the promise of rapid advancements in AI for aerial combat. That promise—and the interest it will generate—is perhaps DARPA’s greatest achievement from this series of trials.
According to recently retired Lt. Gen. Jack Shanahan, who recently served as the first director for the Pentagon’s Joint Artificial Intelligence Center, this “level of excitement/interest will by itself be enormously helpful in driving progress.”
Because one thing has been proven time and again: when it comes to AI, it’s not about succeeding on your first attempt. It’s about setting your sights on achieving a tangible, specific goal, and then iterating rapidly on a solution until you achieve it.
Credits |IHS, Forbes