If you are one of those people concerned with machines taking over the planet and enslaving us, then here’s another reason to fret: DARPA, the agency behind all things right in the world, is challenging AIs to battle it out in simulated air combat scenarios.
In recent years, DARPA has been really pushing the pedal to come up with some type of autonomous combat system, for now just as a means to prove such a concept could work.
At the moment, research is stuck at an algorithm level, but once this is sorted out the leap to a fully functional, physical system should be very quick. But for the algorithms to be work, they need to be tested first.
That’s why DARPA has programs like the AlphaDogfight Trials, aimed at teaching AI enough skills so that at one point air-to-air combat could become automated.
Essentially, DARPA says the Trials are just a step forward for an improved human-machine teaming system, which can be translated “don’t worry, a human will always be in charge.” Relaxed, now?
In the first three runs of the virtual competition, which will be held from November 2019 to March 2020, eight teams will compete against one another, demonstrating the capabilities of their respective AIs.
And if you think for a second this is just DARPA playing around, here are some of the names competing in the Trials: Aurora Flight Sciences (which is Boeing, only spelled with a lot more letters), Lockheed Martin, and the Georgia Tech Research Institute. Just to name a few (full list in the press release section below).
For the tests, the teams will use an open-source flight dynamics simulator for the environment and another one for the visual system. The fighter jet the AI will be controlling is an unclassified FlightGear version of an F-15C.
The algorithms will go up against enemies called Red AIs, and the two top performers will engage each other at the Nellis Air Force base in front of a live audience. The winner of this duel will then have to do a boss fight against a Weapons School fighter pilot in a simulated dogfight.
DARPA says there is no money prize for the winner and the tests are just meant to allow the teams to “benefit from demonstrating their technologies to potential Defense Department end users.”
Talon, anyone?
At the moment, research is stuck at an algorithm level, but once this is sorted out the leap to a fully functional, physical system should be very quick. But for the algorithms to be work, they need to be tested first.
That’s why DARPA has programs like the AlphaDogfight Trials, aimed at teaching AI enough skills so that at one point air-to-air combat could become automated.
Essentially, DARPA says the Trials are just a step forward for an improved human-machine teaming system, which can be translated “don’t worry, a human will always be in charge.” Relaxed, now?
In the first three runs of the virtual competition, which will be held from November 2019 to March 2020, eight teams will compete against one another, demonstrating the capabilities of their respective AIs.
And if you think for a second this is just DARPA playing around, here are some of the names competing in the Trials: Aurora Flight Sciences (which is Boeing, only spelled with a lot more letters), Lockheed Martin, and the Georgia Tech Research Institute. Just to name a few (full list in the press release section below).
For the tests, the teams will use an open-source flight dynamics simulator for the environment and another one for the visual system. The fighter jet the AI will be controlling is an unclassified FlightGear version of an F-15C.
The algorithms will go up against enemies called Red AIs, and the two top performers will engage each other at the Nellis Air Force base in front of a live audience. The winner of this duel will then have to do a boss fight against a Weapons School fighter pilot in a simulated dogfight.
DARPA says there is no money prize for the winner and the tests are just meant to allow the teams to “benefit from demonstrating their technologies to potential Defense Department end users.”
Talon, anyone?