Over a two-week span in December, a heavily modified F-16D Fighting Falcon took to the skies no fewer than a dozen times with an empty cockpit and an artificial intelligence (AI) pilot at the stick. But while pilot-less aircraft aren’t all that uncommon in the skies over warzones today, this Fighting Falcon was packing something different in its memory banks: AI algorithms complex enough to allow it to dogfight all on its own.
The AI fighter pilots came from two different efforts, DARPA’s Air Combat Evolution (ACE) program and the Air Force Research Laboratory’s Autonomous Air Combat Operations, or AACO, program. These AI algorithms were fed into a highly modified F-16D and sent out on a variety of air combat operations, including close-quarters dogfights and beyond-visual-range engagements.
This successful series of tests marks a significant step toward the advanced AI teaming expected to emerge as a part of the next generation of fighters, with the U.S. and a number of foreign allies and competitors all aiming to couple their next stealth jets with scores of specially equipped drone wingmen.
And the Air Force seemingly just proved this concept is not only possible, but downright feasible by turning control of the most maneuverable F-16 in the world to an AI fighter pilot.
Related: Are the days of dogfights over? An in-depth air combat analysis
Hunted by winged Predators
Drones, a term that often encompasses everything from semi-autonomous data collectors to advanced remotely-piloted aircraft (RPAs), have been around almost as long as aircraft themselves. English inventor Dr. Archibald Low is often credited with fielding the world’s first powered drone, an aerial target that first took to the skies in 1917, just 14 years after the Wright Brothers’ first flight. While drones of various sorts have played a variety of roles in conflicts dating all the way back to the First World War, there’s an argument to be made that the era of drones began on July 1, 1995, when the General Atomics MQ-1 Predator entered operational service.
The Predator took the concept of seemingly ever-present eyes — and firepower — in the sky over conflict zones out of the realm of science fiction and straight into the hard-boiled reality of America’s multi-theater Global War on Terror. Equipped for 24-hour operations inside the uncontested airspace of Iraq and Afghanistan, the Predator’s Multi-Spectral Targeting System, made up of an infrared sensor, color/monochrome daylight TV camera, image-intensified TV camera, laser designator, and a laser illuminator, provided Predator crews with all the data they’d need to collect valuable intelligence and, when called for, deliver Holy Hell from on-high via two laser-guided AGM-114 Hellfire missiles.
The success of the MQ-1 Predator soon led to an upgraded iteration of the platform that added 10 feet to the fuselage and 17 to the wingspan, making it about the size of an A-10 Warthog.
This new remotely-piloted platform took all the capabilities that made the Predator so effective and turned them up to eleven, doubling the MQ-1’s top speed and service ceiling (120 knots to 240, and 25,000 feet to 50,000 feet, respectively) and, most impressive of all, expanding its payload capacity from a meager 300 pounds across two hardpoints to better than 3,000 pounds. While the Predator hunted with just two Hellfire missiles tucked beneath its wing, its replacement, dubbed the MQ-9 Reaper, can now carry eight.
Related: MQ-9B STOL: A new Reaper cousin could help Marines win the Pacific
Remotely piloted aircraft can do a lot…
But for all the incredible capability these aircraft bring with them into the fight, they remain reliant on direct human operation. Both Predators and Reapers have ground crews that directly manage takeoffs and landings via line-of-sight antennae, before transferring control over to two-person crews out of Whiteman Air Force Base in Johnson County, Missouri once on the prowl. These crews, made up of one licensed pilot and an enlisted sensor operator, take shifts operating their slow-moving RPAs, ensuring there are always fresh eyes on the screens and fresh hands on the stick.
This approach allows the United States to maintain a persistent presence over target areas, gathering intelligence or engaging targets on the ground as called upon by higher command. It also allows skilled crews to be anywhere in the world that they’re needed, with Reapers and support equipment arriving in theater via cargo aircraft like C-130s and control relayed back to Missouri, regardless of theater. Reaper crews could be flying over Syria on Monday, Iraq on Tuesday, and Afghanistan on Wednesday if operational requirements pressed the need.
But for all their persistence, firepower, and advanced technology these aircraft offer, their reliance on human operators, and the inherent lag associated with transferring data over vast distances, create a significant limitation. In complex air combat operations, the difference between mission accomplishment and being blown out of the sky can come down to fractions of a second… and that spells doom for remotely-piloted aircraft like the Reaper.
Related: Russia is testing its own ‘Loyal Wingman’ for the Su-57
But they can’t dogfight
Back in 2021, I interviewed an MQ-9 Reaper pilot for Popular Mechanics. He walked me through the aircraft’s Multi-Spectral Targeting System that draws in data from a full suite of onboard infrared and video sensors, target designators, and more — all of which can be displayed as separate feeds or fused into a single high-definition display.
“[As] opposed to a multi-targeting display that’s maybe six inches across by a pilot’s knee in most aircraft, we have unparalleled ability to see and locate targets,” he told me.
But all this high-definition data comes at a high price: lag. The pilot told me that it takes approximately 1.2 seconds for the Reaper’s signal to reach him. Based on what he sees, he makes a decision and inputs a response, followed by another 1.2 seconds of transit time as the command is relayed back. As a result, it’s not unusual for there to be a three-or-more seconds lag between a threat presenting itself to the Reaper’s sensors and the aircraft actually responding to it.
In a dogfight, three seconds to react is about two and a half more than you usually get. While there are a number of efforts underway to offset this technical hurdle, the most promising comes in the form of artificial intelligence. In other words, to make dogfights survivable for drone aircraft, it’s time to let the drones do some of the thinking for themselves.
Related: More than missing guns: Why America lost dogfights over Vietnam
The X-62A: DARPA and the AFRL’s AI-enabled, thrust-vectoring F-16
On Monday, the Defense Department revealed a series of 12 advanced fighter maneuver (or dogfighting) test flights conducted between December 1 and 16, 2022, in the X-62A Vista — a heavily modified Block 30 F-16D Fighting Falcon that had been converted into something much more capable by Lockheed Martin’s Skunk Works. The X-62A offers superior aerobatic performance than any F-16 in the sky today, but maneuverability is just part of the package.
What made these dogfight exercises special wasn’t the unusual aircraft flying them, but rather its pilots: two different AI algorithms sourced from the Air Force Research Laboratory and the Defense Advanced Research Projects Agency. In effect, what they produced was a multi-access thrust-vectoring F-16 piloted by artificial intelligence.
The X-62A’s vectoring nozzle and adaptive control fly-by-wire system allow it to mimic the flight characteristics of a wide variety of aircraft — anything from a C-17 Globemaster II to the F-22 Raptor. But in 2021, its existing Vista Simulation System (VSS) was upgraded to the System for Autonomous Control of Simulation (SACS) — making it not only capable of mimicking other airframes in flight, but able to do so under the direct control of what the Air Force calls an AI agent.
Related: F-16XL: Why America didn’t get the best F-16
AI enters the chat… and the dogfight
Over the course of about two weeks, the X-62A flew against a variety of simulated opponents in a series of one-on-one engagements held both within and beyond visual range. The enemy fighters existed only within the digital realm, but the AI agent and its fighter flew against them as though they were just as real as any other.
“We conducted multiple sorties [takeoffs and landings] with numerous test points performed on each sortie to test the algorithms under varying starting conditions, against various simulated adversaries, and with simulated weapons capabilities,” said Air Force Lt. Col. Ryan “Hal” Hefron, the DARPA program manager for ACE.
It flew these dogfights with two different AI agents onboard — one, dubbed Autonomous Air Combat Operations (AACO) from the Air Force Research Laboratory (AFRL), and the other, Air Combat Evolution (ACE) from the Defense Advanced Research Projects Agency (DARPA). Incredibly, the Air Force says it was able to swap AI fighter pilots in just a matter of minutes, allowing the same aircraft to fly operations with each AI pilot over the span of just hours.
“The flights demonstrated that AI agents can control a full-scale fighter jet and provided invaluable live-flight data,” DARPA wrote in its press release.
In all, the Air Force racked up 17 hours of AI-piloted fighter operations over Edwards Air Force Base in California during these exercises.
Related: US SpecOps are using AI to look for an edge
DARPA’s ACE has already squared off against human pilots and won
One of these two AI agent fighter pilots is no stranger to garnering dogfight-related headlines. Back in August of 2020, DARPA held its AlphaDogfight trials, pitting eight teams against one another until only the most capable guns-only AI fighter pilot remained. The victor, an AI agent developed by Heron Systems, then took on a real human F-16 pilot in a virtual showdown… and the AI mopped the floor with its fleshy opponent, winning five times in a row without the human pilot ever scoring a single hit.
Heron Systems credited their victory to their AI fighter pilot’s superior nose control, making the F-16’s simulated M61A1 Vulcan 20mm cannon as deadly as possible.
“It’s got to keep that opponent in that one degree cone to win the game,” Ben Bell, Heron’s Senior Machine Learning Engineer, told Sandboxx News at the time.
“You saw that a lot with Lockheed, we’re both nose on, we’re both creating damage, but when their nose is off by that one degree, that’s where we were able to win a lot of these engagements.”
But it wasn’t just the AI’s precision that made it such a dangerous competitor — it was also its lack of self-preservation. As Bell explained to me at the time, they designed their algorithm to place equal value on damaging the enemy as it placed on minimizing risk to itself. As such, it was willing to do things most human pilots wouldn’t (or shouldn’t).
“If the agent sees a 51% chance of scoring a kill as it heads into a neutral merge, it’s going to take it,” Bell explained.
This speaks to one of the real values inherent to turning over control of our most advanced fighters to artificial intelligence pilots: Without a human operator in harm’s way, the envelope of allowable risk becomes much wider. Some operations that may seem too dangerous to send a human pilot into could be seen as viable for a pilot-less aircraft. And with advanced, low observable, but very budget-conscious aircraft in development like the XQ-58 Kratos Valkyrie, which costs just a bit more per airframe than a Tomahawk cruise missile, taking risks may not even come with a massive financial downside, either.
Related: AI wins flawless victory against human fighter pilot in DARPA dogfight
Just how good are AI fighter pilots?
While this is undoubtedly a significant step toward turning advanced fighter operations over to AI, there remains a significant gap between the controlled testing environments of these exercises and the complex reality of air combat.
“It’s important to realize that a BFM (Basic Fighter Maneuvers) engagement can occur in any direction and any altitude. We’ll often begin with a basic starting parameter to develop a site picture to reference, but a real engagement doesn’t have those cuffs,” Justin “Hasard” Lee, an F-35 Pilot instructor and former F-16 pilot, told Sandboxx News.
“The enemy always has a vote, meaning they always reserve the right to do something you’re not expecting. When that occurs you have to find a creative solution to counter the unexpected problem.
Real-world limitations, like an emphasis on the preservation of expensive equipment like a highly modified, thrust-vectoring F-16, require a more complex approach to decision-making than must be leveraged in a simulated environment.
During the AlphaDogFight trials, Heron’s AI pilot was widely described as “aggressive” by DARPA staff and the Air Force pilots on hand. Under the control of Heron’s AI, the virtual F-16 would practically play chicken with its opposition — something the human pilots were quick to point out would be a violation of training regulations in a real simulated dogfight exercise. Of course, in an actual dogfight, there are no such limitations… but Heron’s aggression may still have been turned up just a bit too high to be realistic.
In the real world. it seems, the AI agent fighter pilots weren’t quite as successful as they were in the digital realm, but as Hefron points out, that’s to be expected.
“We didn’t run into any major issues but did encounter some differences compared to simulation-based results, which is to be expected when transitioning from virtual to live,” Hefron said.
Related: Can the F-35 dogfight? The truth behind the infamous 2015 report
What does this mean for the future of fighter pilots?
It’s impossible to discuss AI agents running advanced fighter ops without addressing the fighter-pilot-shaped elephant in the room. For decades now, we’ve consistently heard that fighter pilots will be a thing of the past just as soon as drones can think and act quickly enough to replace them… but even with the recent success of these AI fighter pilot programs, that eventuality still seems a long way off.
In fact, getting human pilots out of the equation isn’t even really the long-term goal of efforts like DARPA’s ACE or the AFRL’s AACO. Instead, these programs, and a number of others like them, are all about teaming artificial intelligence with real fighter pilots to create what amounts to a collaboration between the two.
Programs like the Air Force’s Next Generation Air Dominance, the Navy F/A-XX, and a number of foreign efforts have all stated plainly that they envision the next generation of fighters as more than a single jet, but rather a “system of systems” that includes the crewed fighter and a constellation of drones flying in support. The idea is to leave the high-level decisions up to human decision-makers in nearby fighters or even AWACS, then using artificial intelligence to allow the drone to execute its orders without direct support.
The Air Force Research Laboratory calls this approach autonomous aircraft teaming architecture, and it can allow a relatively small number of high-cost, crewed airframes to manage whole teams of lower-cost drone wingmen, potentially even entire swarms of them, to accomplish a variety of air combat operations through a combination of the low observable technology that dominated the latter half of the 20th century, and the high-volume saturation attacks that defined the bombing raids of World War II.
“Embedded within the teamed aircraft, complex algorithms and cutting-edge sensors enable the autonomy to make decisions based on established rules of engagement set by manned teammates,” the Air Force Research Lab websites states about its Skyborg program.
But AI won’t just be flying the drone wingmen, DARPA also envisions it riding along in the cockpit of crewed fighters as well. Pilots have long contended with the complexity of flying a hundred million dollars worth of state secrets through contested airspace and still keeping their minds on the mission at hand. Newer fighters like the F-35 go a long way toward reducing the cognitive load on pilots, automating as much as possible to free up the operators’ attention for the fight at hand. In the future, AI could take this even further, managing many of the basic functions of the jet to allow the human part of the system to do what it does best — make good combat decisions.
The fact of the matter is, these successful tests will potentially have massive ramifications for the future of air combat, but this isn’t the final chapter in the book of human fighter pilot history.
But it just may be the first chapter in a new book about collaborative AI and human-driven air combat.
Read more from Sandboxx News
- What it’s really like to push 9Gs in a dogfight flying the F-16
- AI wins flawless victory against human fighter pilot in DARPA dogfight
- The story behind this all-gold F-16 – and 5 more new combat plane paint jobs you need to see
- That time a Russian fighter shot down one of its own in a mock dogfight turned real
- What really happened when F-22s squared off against the Eurofighter Typhoon?
Joshua Campbell says
The use of artificial intelligence in military aviation is an exciting development that has the potential to transform the way we conduct air combat. The recent demonstration of an AI pilot flying a real F-16 in simulated dogfights by the US Air Force is a significant milestone in this field, as it shows that AI can not only handle complex tasks but also respond to unexpected situations in real-time. This technology has the potential to revolutionize the way we train pilots and conduct air operations, leading to increased efficiency, safety, and effectiveness in combat. However, as with any emerging technology, there are also concerns about its ethical implications, and it will be important for the military to carefully consider these issues as they continue to develop and deploy AI in aviation.
Thank for your articles
Roxanne Moore says
The negative items on my credit report had deprived me of so many opportunities. I had 19 inquiries and other derogatory items which were about 15. These made my credit score very low 541(Transunion), 529(Equifax) and 518(Experian). My cousin told me about METRONET CREDIT SOLUTION a credit specialist who had helped him restore his credit. I reached out to them immediately and right now I’m glad I did. I pulled my credit after 9 days as I was instructed and to my surprise my score is now 798(Transunion), 788(Equifax) and 780(Experian) and all the negative items are gone. I’m so happy, and so I’m recommending them to any person in need of credit repair.