The AI-controlled virtual fighter jet has trounced a human pilot in a dogfight. The feat occurred during the finale of the US military’s AlphaDogfight challenge.
This was put on by the controversial Defense Advanced Research Project Agency (DARPA).
This involved some of the largest military contractors including Lockheed Martin participating in the competition.
Peter W. Singer, a strategist and Senior Fellow at New America believes this exercise marks a major milestone in the new technology.l
He tweeted: “A John Henry-versus-the-steam-engine moment.”
READ MORE: Coronavirus news: Schools report cases in Berlin – second wave fears
“Each and every data point makes that system more intelligent, more capable.
“Win or lose, the human pilot that shows up for round 2 and then 3 will be effectively the same 1.0 version, while the technology will be at 2.0, 3.0 on so on.
“Moreover, AI is also a ‘black box’, so its creators soon won’t even know how it improved itself. Just that it did.
“But don’t expect wholesale move to this kind of system anytime soon.”
The military strategist then outlined three reasons why it will take time.
He tweeted: “The decision to automate is not just about “could” but ‘should’.
“There are a host of legal and ethical concerns that would have to be navigated through first.
“Second, the military is, for very good reasons, a conservative organisation, seeking to go into battle with what is known and trusted.
“Any new tech, let alone one of such change, has to go over that hurdle.
“Third, there is massive investment in existing programs of record, with literally hundreds of billions of dollars and tens of thousands of jobs at stake.
“Anything that alters those plans is a threat that will be fought bureaucratically.”
The teams had to start from the ground up to teach their AIs how to fly a fighter jet.
Lee Ritholtz, director and chief architect of AI, from Lockheed Martin, said: “You don’t have to teach a human [that] it shouldn’t crash into the ground.
“They have basic instincts that the algorithm doesn’t have.
“That means dying a lot. Hitting the ground, a lot.”
For Lockheed Martin, it took several servers running trial-and-error dogfights around the clock to come up with its final AI, a piece of software capable of being run on a single graphics card.
The winning team’s AI had been through more than 4 billion simulations.
This equates to approximately 12 years of flight time, as Ben Bell, senior machine learning engineer at Heron Systems, told Defense One.