Fighter jets need pilots to fly them, but no one knew that could change one day. Over the past 20 years, technology has shown that it can think faster than human beings and react faster in a precise, pristine environment. This week, a virtual Top Gun-style competition was organized by the Defense Advanced Research Project Agency – better known as DARPA – in which various artificial intelligence algorithms flew simulated jets in digital dogfights. There were no actual planes in the air, but the aim was to see which artificial intelligence expert could provide the most formidable fighter. The event started on Tuesday morning, with the most potent AI fighting a virtual F-16 flown by a real flesh-and-blood pilot on Thursday. The final competition was initially expected to occur at AFWERX in Las Vegas, featuring pilots from Air Force Weapons School based at Nellis Air Force Base. Still, the contest is now being conducted digitally due to the coronavirus pandemic.
- What is DARPA?
- About DARPA’s AI-powered Jet Event
- Ideology Behind The Event
- Future Goals
Let’s talk more about DARPA’s virtual AI-powered event.
What is DARPA?
Defense Advanced Research Project Agency (DARPA) has kept on to a single and lasting goal for sixty years: to make crucial investments in national security breakthrough technologies.
The origins of that mission and of DARPA itself dates back to Sputnik’s launch in 1957, and a determination from the United States to be the initiator and not the target of strategic technological surprises from that moment on. Working with innovators within and outside of government, DARPA has consistently delivered on that goal, translating innovative ideas into realistic capabilities, and even seemingly impossible ones. The ultimate outcomes included not only game-changing military capabilities such as precision weapons and stealth technology, but also icons of modern civil society such as the Internet, automated voice recognition and language translation, and recipients of the Global Positioning System small enough to fit into various consumer devices.
About DARPA’s AI-powered Jet Event
The first two trials took place in November 2019 and January 2020 at Johns Hopkins Applied Physics Laboratory (APL) in Laurel, Md. In the AlphaDogfight Trials, eight teams were initially selected to compete with the objective of “demonstrating advanced AI algorithms capable of performing simulated in-visual air combat maneuvers. This August’s event was the third stage in what is called the AlphaDogfight Trials. Last fall, the first trial in the series was many rookie algorithms trying to figure out the fundamentals of aviation. The event was scheduled for three days.
- The teams will deploy their flying algorithms on the first day of the event against “five adversaries AI algorithms developed by the APL.”
- Day two features a tournament in a round-robin style, where teams compete against each other.
- On the final day of the event, the four top-ranked teams will compete in a single-elimination bracket for the AlphaDogfight Trials Championship.
The eventual champion will then face off in an F-16 simulator against a human pilot to analyze “to test the Artificial Intelligence developer’s capabilities against a human.”
Ideology Behind The Event
Independent ventures like those are on their radars, but their emphasis is slightly different philosophically. Initiatives outside DARPA have tendered to fix on the Beyond Visual Range issue, which is not the initial thing we do with our pilots. Militaries may want to send an uncrewed fighter jet ahead, like a scout, and probably attack an enemy’s air defenses. But before something like this can happen, Javorsek argues that AI needs to prove it can perform a more simple task: the dogfight.
These experiments are intended to extend DARPA’s Air Combat Evolution (ACE) AI developer base program. The ACE program focuses on automating aerial combat and building “human trust in AI as a step towards improved teaming between humans and machines.”
Although this competition was virtual, companies in the real world are already working on the hardware for pilotless fighter jet-type drones. One such small uncrewed aircraft is named the Valkyrie, or XQ-58A, built by California-based Kratos Company. Another comes from Boeing, an uncrewed fighter jet with a rotating nose — it’s nicknamed the “Loyal Wingman.” The theory behind these machines is that they may be a kind of robot wingman, escorting a human-flown aircraft. Some advantages that come with this technology are
- They would be cost-efficient
- Wouldn’t have a pilot on the aircraft
- The loss of life would be minimal
Javorsek said he sees a difference between the types of tasks that AI and humans might perform looking into the future, with computers focused on physically flying an airplane. People are free to keep their eyes on the bigger picture. “The idea here is to have human brains at the proper spot,” he reflects. Artificial components should concentrate on the “low-level, maneuvering, tactical activities,” he says. In contrast, their flesh-and-blood counterparts can be “combat managers” who can read “adversary meaning, purpose, and feeling.”
Although this DARPA competition had the flying occurring in digital skies, real-life flight in an actual fighter jet is physically challenging on the aircraft onboard — something I had the opportunity to encounter in an F-16 first hand. Pulling hard turns or accelerating rapidly generates drastic G forces, and if the pilot and crew do not appropriately handle them, they may pass out. One day, certified artificial intelligence experts could fly a plane in combat. Still, if a pilot were to be on board hypothetically, they would want to remain conscious during the war. In other words, any stick-controlled algorithm would have to understand what humans can stand up to. Or, if the AI is in charge of an uncrewed drone, then there would be no need to worry about the impact of gravitational force on a human.