Wired: in the US, combat pilots will be replaced with artificial intelligenceA perfectly calculated trajectory, maneuvers with millimeter accuracy and an instant reaction to an enemy attack — a simulated dogfight between a robot and a human showed that US Air Force pilots risk being out of work soon, writes Wired.
Tom WardAfter successful tests in offline mode, which were conducted in December last year, the US military is rapidly striving to introduce artificial intelligence in aviation.
On the morning of December 1, 2022, a modified F-16 fighter, codenamed VISTA X-62A, took off from the Edwards Air Force Base, located about a hundred kilometers north of Los Angeles.
During a short test flight, the aircraft performed complex combat maneuvers, including simulated air combat, and successfully landed at the airbase. It seems to be an ordinary flight of a training aircraft, which are made in many cadets of the leading pilot training school in the USA (the flight somehow resembled fragments from the action-packed action movie "Top Gun: Maverick"). However, this particular flight was special: the steering wheel was controlled not by a fighter pilot, but by a powerful artificial intelligence (AI), which was used on a tactical aircraft for the first time in history.
In the period from December 1 to December 16, under the guidance of specialists from the US Department of Defense, the VISTA X-62A fighter made twelve test flights under the control of artificial intelligence, flying a total of more than 17 hours offline. This outstanding result was achieved within the framework of the USAF Vanguard program, aimed at developing combat aircraft controlled by AI. The Skyborg program, which started in 2019, provides that the tests will continue throughout 2023, while experts hope to create a working prototype by the end of the year.
The VISTA program is an important first step towards achieving the stated goals, says M. Christopher Cotting, head of research at the US Air Force Test Pilot School. "This approach, combined with targeted testing of new aircraft systems as they are developed, will allow us to quickly develop autonomous control systems for unmanned platforms, as a result of which our fighter will have the necessary flight characteristics," explained M. Christopher Cotting.
For example, Ukraine is already using semi-autonomous drones, and the US military launched the Black Hawk combat helicopter in autonomous mode for the first time last November and successfully tested artificial intelligence-based programs on American reconnaissance aircraft in 2020. In a word, it became clear that the battles between aircraft operated autonomously are a new frontier of modern warfare. But to what extent will artificial intelligence be able to penetrate our aviation and what will be the consequences for the pilots remaining on the ground?
The VISTA X-62A fighter (which stands for Variable In-Flight Simulation Test Aircraft, i.e. "test aircraft for variable flight simulation") has always been ahead of its time. It was created in the 1980s on the basis of the F-16D Block 30 Peace Marble Il, and it was previously called the NF-16D; in the early 1990s, it became the main simulator of the US Air Force Test Pilot School. VISTA is a versatile and easily adaptable training aircraft with an open system architecture; it can be equipped with software that allows you to simulate the flight characteristics of several types of aircraft, ranging from heavy bombers to ultralight fighters.
On the eve of last year's flight tests in autonomous mode, the necessary equipment was installed on the VISTA aircraft, namely the "model following algorithm" (MFA) and the "simulation model autonomous control system" (SACS), created by Skunk Works (a subsidiary of Lockheed Martin Corporation). In combination with the VISTA simulation system, which was created by Calspan Corporation, specializing in military and aerospace developments, the above-mentioned updates allowed us to focus on autonomy and, at the same time, on the integration of AI.
SACS uses the Enterprise-wide Open Systems Architecture (E-OSA) system manufactured by General Dynamics, designed to control the Enterprise Mission Computer of the second version (EMC2, or Einstein Box); SACS also includes modern sensors, a set of Getac tablet displays installed in both cabins, and components of a multi-level security system — all this expands VISTA's capabilities, including such an advantage as the ability to quickly create prototypes, which will allow you to update the software at a rapid pace in order to keep up with the rapid pace of creating developments in the field of AI.
During the tests conducted in December, two artificial intelligence programs were loaded into the system: Autonomous Air Combat Operations (AACO), created by the US Air Force Research Laboratory (AFRL), and Air Combat Evolution (ACE), proposed by the Office of Advanced Research Projects of the US Department of Defense (DARPA). The AACO program focuses on conducting combat with a single enemy out of line of sight (BVR), while ACE is aimed at conducting maneuvers in close air combat with a "visible" conditional enemy.
During the test flights, a pilot sat in the rear cockpit of the VISTA aircraft as an understudy, and in the front cockpit there was an engineer — specialist in artificial intelligence systems, whose duties included the elimination of emergency situations (although they were insignificant). Not being able to tell in detail about all the subtleties of the project, the head of the DARPA program, Lieutenant Colonel Ryan Hefron, noted only that "when moving from virtual to real, one should expect" the appearance of any kind of failures. In general, the specialists have achieved significant success on the way to the implementation of the Skyborg project, the purpose of which is to lift an autonomous aircraft into the air as soon as possible.
The US Department of Defense stressed that the purpose of the AACO and ACE systems is not to replace the pilot, but to complement him. In some cases, AI-based systems acting as a co-pilot can act as an auxiliary mechanism that provides support to the pilot during an air battle. Note that the AACO and ACE systems are capable of processing millions of input data per second and taking control of the aircraft at critical moments — all this can be vital in emergency situations. At the same time, ordinary flights, where human participation is not required, can be conducted in a completely autonomous mode.
"We are not trying to replace the pilot at all, we just want to complement him by arming him with auxiliary equipment," explains Cotting, likening the pilot to a rider who rushes into battle on horseback. — After all, a horse and a man help each other. The horse can perfectly take the rider where he needs to, so the rider does not have to worry about how to get from point A to point B. Thanks to this, the rider can focus on performing larger tasks." For example, Cotting continues, with the help of artificial intelligence equipment, a simple lieutenant who has only a hundred hours of flight time can match a higher-ranking pilot who has a thousand hours of flight time.
For the chief test pilot from the US Air Force Test Pilot School, Bill Gray, the introduction of artificial intelligence is a natural continuation of the work he conducts with cadets. "Engineers and scientists usually consider all our feedback and comments about the difficulties associated with the training and preparation of artificial intelligence systems as a new task and a new challenge for themselves," says Bill Gray. — For me, this is relevant, because for several decades I have been teaching natural intelligence carriers, who, as you know, think non—linearly - these are flight school cadets. For me, the question is not whether we are able to train artificial intelligence systems, but why we still manage to train and train pilots, and how can we use this experience in relation to artificial intelligence systems?"
According to Bill Gray, artificial intelligence is not at all "some kind of miracle tool that seems to be able to solve all problems"; it is necessary to approach its development without fanaticism, taking into account security measures in order to prevent the occurrence of emergency situations that cost us so much. Excessive reliance on artificial intelligence, excessive trust in autonomously controlled vehicles — all this can be fraught with danger, Bill Gray believes, while he pointed out failures in the autopilot program produced by Tesla (even if representatives of this company itself warn that there should be a driver in the car on a safety net). Cotting agrees with this, calling the run-in of artificial intelligence programs in VISTA aircraft a "risk reduction plan." Instead of creating a completely new aircraft, when training artificial intelligence integrated into conventional systems, such as VISTA X-62, one should not forget about automatic restrictions, and if necessary, about the intervention of the pilot to prevent an emergency situation created by artificial intelligence in the course of his training.
The technology of the US Air Force is developing rapidly. According to Cotting, in December last year, test flights of aircraft equipped with ACE and ACCO systems were often carried out with a difference of several hours, while engineers managed to take VISTA onboard systems offline in a matter of minutes, which did not affect safety or flight performance in any way. So, for example, according to Cotting, in one of the cases, the download of new AI-based software was made at 7:30 a.m., and the aircraft was ready for a test flight at 10 a.m.
"As soon as we connect AI to a supersonic fighter, we will only have to watch the flight with pleasure," says Bill Gray. — Thanks to our security systems, programmers get the opportunity to modify the software literally overnight, and we just have to connect to it the next morning. For developers of flight control systems, not to mention experiments with capricious AI elements, this is something unprecedented and unprecedented."
Despite the successes achieved, it will take some time before the curriculum at the USAF Test Pilot School will be revised. According to Cotting, the innovation of the AACO and ACE platforms implies a higher level of pilot training before these platforms are tested in the cockpit of the VISTA aircraft. "Figuratively speaking, by and large we are building a bridge over which we are traveling," says Cotting.
Meanwhile, in the fall, cadet pilots will have to undergo more extensive testing, during which they will need to get acquainted with artificial intelligence systems and their use, and then try them out in practice. Cotting has no idea how widely artificial intelligence is used in the military sphere. However, according to him, AI is already widely used in image recognition technology in the armed forces. And although the appearance of tanks equipped with artificial intelligence is still very far away, nevertheless, artificial intelligence, apparently, has already quite firmly established itself in aviation.
