Tech

An AI beat a human pilot in a dogfight simulation. Again.

It's happened before, but this is the first time an AI has done so in such a highly-controlled environment.

picture alliance/picture alliance/Getty Images

Back in 2016, researchers proved they could construct an artificial intelligence (AI) algorithm capable of beating some of the U.S.'s best fighter pilots in aerial combat, but apparently that wasn’t enough for the engineers of our future extinction. At DARPA’s AlphaDogfight Trial this week, an AI designed by Heron Systems went and did it again, beating a human F-16 fighter pilot 5-0 in a VR trial system. Uh-oh.

Not even a close competition — To earn its chance to stomp some of the most talented biological minds in warfare, Heron’s AI system first beat out Lockheed Martin’s own algorithm entry, along with six other defense contractors before going on to remind everyone just how screwed we are. What’s more, the AI did this with significant handicaps in place, like not being allowed to improve using information learned from its previous competition rounds, disabling its collision detection systems, and restricting weapons to only the aircraft's nose cannon.

Heron System’s AI system used what’s known as deep reinforcement learning, which allows algorithms to quickly try out problem-solving options in a virtual environment to approach something akin to understanding. We’re using the term “akin to understanding,” because really we’re too terrified to type out what it really is — that robots can pilot F-16’s now, and are more than capable at it than any of us.

So what? — Computers have been beating humans at all kinds of tasks — chess, games of go, big data analysis — for ages. Why should we care about this? Well, how about the fact that DARPA is apparently super excited about the new developments, because it thinks it’ll lead to a melding of (wo)man and machine that cruises 50,000 feet above our previously blissfully unaware heads.

Humans and robots, killing in harmony — “I think what we’re seeing today is the beginning of something I’m going to call human-machine symbiosis,” DARPA’s director of its Strategic Technology Office told Defense One. “Let’s think about the human sitting in the cockpit, being flown by one of these AI algorithms as truly being one weapon system, where the human is focusing on what the human does best [like higher-order strategic thinking] and the AI is doing what the AI does best.”

Between this and the ability to literally just 3D-print your own mini-Boston Dynamics quadruped Spot using a Raspberry Pi, we’re asking for a massive robot uprising at this point. Sure, organizations like Human Rights Watch strongly advise against casually amassing deadly robotic force, but we’re sure the Terminator armies will take that into consideration. Look, can’t we just pass a law restricting robot jobs to only cheerleading, and call it a day?