In a recent duel in California between two American fighter jets, only one of the pilots was human. The other wasn’t. The highest-ranking civilian in the Air Force was traveling in the front seat of the second jet, which was being driven by artificial intelligence. It was the pinnacle of how far the Air Force had progressed in creating a technology that originated in the 1950s. However, it’s merely a taste of the technology to come.
In the race to keep ahead of China in the application of AI to weaponry, the US is competing. Concerns about future warfare being waged by machines that choose and attack targets without direct human involvement have been raised by the attention being paid to artificial intelligence. According to officials, this won’t occur, at least not from the US perspective. However, there are concerns about what a possible opponent would permit, and the military believes that there is no other choice except to quickly field American capabilities.
The vice chairman of the Joint Chiefs of Staff, Adm. Christopher Grady, stated, “Whether you want to call it a race or not, it certainly is.” We both understand that this will be a crucial component of the battlefield in the future. China is working as hard on it as we are.
An examination of the evolution of artificial intelligence in the military, upcoming technologies, and methods for controlling themActually, AI’s military origins are a cross between autonomy and machine learning. When a computer analyzes data and rule sets to conclude, machine learning takes place. When those findings are used to perform without additional human input, autonomy arises.
The Navy’s Aegis missile defense system was developed in the 1960s and 1970s, marking the beginning of this. Aegis was trained to recognize and intercept incoming missiles independently and faster than a human could using a sequence of if/then rule sets that were designed by humans.
However, the Aegis system’s responses were constrained by the rules it was given and were not intended to learn from its mistakes.
“If a system uses ‘if/then,’ it is probably not machine learning,” said Air Force Lt. Col. Christopher Berardi, who is assigned at the Massachusetts Institute of Technology to support the Air Force’s AI development.
Machine learning is the branch of AI that deals with building systems that learn from data.
When large data and powerful computing power came together in 2012, AI saw a significant advancement as computers were able to analyze data and create rule sets on their own. It’s what specialists in AI have dubbed the “big bang” of AI.
Artificial intelligence is new data produced by a machine that is writing the rules. AI-enabled autonomy is the ability to program systems to make decisions based on machine-written rules and then behave accordingly.
This month, Air Force Secretary Frank Kendall saw a taste of this cutting-edge warfighting when he participated in a dogfighting exercise over California’s Edwards Air Force Base on Vista, the first F-16 fighter plane controlled by artificial intelligence.
Though there are hundreds of active AI programs throughout the Pentagon, that jet is the most obvious example of the work that is being done in this area. To build a data set from the deluge of messages sent back and forth between crews and air operations centers during flights, service members at MIT labored to sift through thousands of hours of recorded pilot conversations. This was done to help the AI distinguish between important messages, such as a runway being closed, and routine chatter in the cockpit. The intention was to train the AI to identify which messages should be elevated for controllers to receive them sooner.
The military is developing an AI substitute for GPS satellite-dependent navigation in another important project. High-value GPS satellites would probably be struck or interfered with in a future conflict. The loss of GPS might render American banking, communication, and navigation systems blind, and hinder the ability of the country’s fleet of warships and planes to coordinate a response.
To develop a different approach that makes use of Earth’s magnetic fields, the Air Force piloted an artificial intelligence application last year that was installed on a laptop that was fastened to the floor of a C-17 military cargo aircraft.
Although it has long been known that airplanes might navigate by following the magnetic fields of Earth, this hasn’t proven to be feasible since every aircraft produces so much electromagnetic noise of its own that it hasn’t been possible to effectively filter out signals from the planet alone.
Col. Garry Floyd, director of the Department of Air Force-MIT Artificial Intelligence Accelerator program, stated that magnetometers are extremely sensitive. “We would see it if you turned on the strobe lights on a C-17.”The outcomes “were very, very impressive,” according to Floyd, as the AI learned which signals to follow and which to ignore from the flights and mountains of data. “We are discussing tactical airdrop caliber.”We believe that should we find ourselves functioning in an area where GPS reception is unavailable, we may have added an arrow to our quiver of options. which we’ll do,” Floyd declared.
Only the C-17 has been used to test the AI thus far. Tests will also be conducted with other aircraft; if successful, this might provide the military with a backup plan if GPS fails.
As the Air Force trains Vista, the AI-controlled F-16, it has a lot of safety measures in place. The still-learning AI is prevented from performing actions that could endanger the plane by mechanical constraints. Additionally, there is a safety pilot who, at the touch of a button, can take over from the AI.
Since the algorithm cannot learn while in flight, it starts with the data and rule sets it has already constructed from earlier flights. The algorithm is reloaded onto a simulator after a new flight has concluded, where it is fed fresh in-flight data to learn from, generate new rule sets, and enhance its performance.
But AI is picking things up quickly. AI has already surpassed certain human pilots in dogfighting exercises due to its lightning-fast computational speed when analyzing data and flying new rule sets in the simulator. Its rapidity in figuring out the best method to fly and maneuver has also helped it surpass other pilots.
However, officials stressed that controlling the data reinserted into the simulator for the AI to learn from is the most crucial approach to ensure safety, which is still a top priority. In the instance of the jet, it’s ensuring that the information represents a safe flight. In the end, the Air Force envisions a version of the AI being built acting as the brains for a fleet of 1,000 unmanned jets that General Atomics and Anduril are working on developing.
The service members deployed to MIT edited the tapes in the experiment to eliminate classified material and the pilots’ occasionally sour language to train AI on pilot communication.
Understanding pilot communication is “a reflection of pilot thought processes, of command and control.” If the machines are to become sophisticated, they must comprehend this as well, according to Joint Chiefs Vice Chairman Grady. “They don’t have to pick up swearing.”
