US Air Force successfully tests AI-controlled fighter jet in dogfight against human pilots

[midian182]

What just happened? In what has been hailed as a milestone moment for artificial intelligence's use in the military, the US Air Force has announced that it successfully tested a modified, AI-controlled fighter jet in a dogfight against human pilots last year.

The US Air Force Test Pilot School and the Defense Advanced Research Projects Agency (DARPA) first started testing a Lockheed Martin X-62A VISTA (Variable In-flight Simulation Test Aircraft) fitted with AI software back in December 2022, part of the Air Combat Evolution (ACE) program. Able to mimic the performance characteristics of other aircraft, the X-62A was flown for by the AI for over 17 hours.

DARPA revealed on Thursday that in September 2023, the X-62A carried out the first successful AI versus human within-visual-range engagements, also known as a dogfight.

The AI dogfights paired the X-62A VISTA against manned F-16 aircraft in the skies above Edwards Air Force Base in Kern County, California. After initial flight safety was built up using defensive maneuvers, the aircraft switched to offensive high-aspect nose-to-nose engagements where they approached within 2,000 feet of each other at 1,200 miles per hour. It isn't revealed which side won the simulated battle.

The X-62A has a backup pilot in the rear cockpit and a technician in the front who can disengage the AI and take over the controls should anything go wrong. DARPA says the pilots did not have to activate the safety switch at any point during the dogfights.

There have been 21 test flights conducted to date, with more continuing throughout 2024.

In addition to its autonomous flight capabilities, the X-62A VISTA, a modified F-16, also features a high-resolution camera, a compact size, lightweight construction, and is versatile enough to be used for a wide range of applications, including scientific research, surveillance, recon, environmental monitoring, and emergency response.

Could Air Force pilots be yet another profession that is eventually threatened by AI? The 2020 AlphaDogfight Trials, a three-day competition designed to demonstrate advanced algorithms capable of dogfighting (using VR simulations), saw an experienced F-16 Air Force pilot lose 5-0 to the AI agent. DARPA said the machine performed aggressive and precise maneuvers that the human pilot could not match.

DARPA says that the X-62A VISTA will continue to serve a variety of customers for research while providing key academic lessons for the next generation of test leaders.

AI-controlled weapons in the military has been a controversial area for years. Elon Musk was one of many experts who called for a ban on the technology in 2017. In response, the army said that the final decision to engage a target always rests with humans, not machines.

Permalink to story:

US Air Force successfully tests AI-controlled fighter jet in dogfight against human pilots

 

[The Talking Tech]

Army members will be the first ones to be replaced by Ai... but that's a good thing, we don't need anymore dead soldiers.

 

[Endymio]

Army members will be the first ones to be replaced by Ai

The jet pilots will be the first, as replacing them instantly means a platform that can now do 20g and 30g maneuvers without killing the man inside. But -- much more importantly -- the power source is already provided. We can't give AI infantry their own jet engine, and battery tech just isn't there yet.

 

[Jack77]

With AI piloting fighters, you can probably create other airplane designs. Because an AI does not have to worry about (especially negative) G forces that much.

 

[godrilla]

The test sample was small. If you increase the sample size naturally you will find an anomaly like Maverick that can beat the Ai. Although eventually the ai will learn from the outliers and improve as well. The job is going to transition from hands on in the cockpit to remotely via dispatcher remote technician hence the human has the final say to engage with the enemy remotely. Ai co-pilot is also another option.

 

[Uncle Al]

The real test will be in actual combat and there are plenty of opportunities for that around the world. Give a couple of them to Ukraine and let then test them. If successful they will close the skies to Russian fighters, which will give Ukraine a leg up in their war .... not to mention getting the real world combat data needed to determine if AI is all it claims to be.

 

[techstrike]

Working for AI-controlled fighter jets started before even the internet, now it just got matured

 

[Theinsanegamer]

I see the plot of Ace Combat 7 is coming along well.

 

[mbk34]

Do they still do dogfights? I thought they just bring anti aircraft missiles in range, press the button and then try and get out of there. I thought dogfighting just happened in the movies.

The test sample was small. If you increase the sample size naturally you will find an anomaly like Maverick that can beat the Ai.

All the advantages are with the AI. The plane doesn't need to carry the pilot and all the stuff that keeps him alive. It doesn't even need the armour that protects him. It can pull turns that would make a pilot black out. The difficult thing is getting past the legal hurdle of allowing AI to kill people. It doesn't even fear being shot down or even flying headlong into the opponent. it's also easy to train AI pilots.

Army members will be the first ones to be replaced by Ai

It's far more difficult to write code to navigate terrain filled with trees, wall, ditches, rivers mines etc. It's relatively straightforward to move through the open air and there are terrain maps to tell you where the mountains are etc. There's also less chance of shooting a civilian in contested airspace. Airplanes will be first, then ships, then land. Space would also be quite easy to work in for AI if we wanted to have satellites that deorbit opposing nations satellites etc.

 

[stroker]

So we're just gonna ignore the last 30 years of Sci Fi movies? Gotcha. Nothing could ever go wrong here. Moving on...

 

[umeng2002]

DARPA? So it's 20 years from being a viable platform. lol

 

[loki1944]

Lol; fake ai and goverment quality; no thanks

 

[Endymio]

So we're just gonna ignore the last 30 years of Sci Fi movies? Gotcha. Nothing could ever go wrong here.

The last 3,000 years of human history has taught us that societies that ignore technological progress are marginalized, if not outright eliminated entirely.

 

[nitebird]

There's a lot of risks to go full AI since you're creating a vulnerability in control disruption/hijacking that could even happen remotely. Anyone remember the drone incident? Maybe a squadron with a human pilot making decisions and AI dog fighting squad mates.

 

[ragreeen2646]

I guess we should just fire all humans and let AI do everything.
Don't you think the AI hype is getting insane.

 

[yannus]

I hope we can trust those who give orders to the AI.........................

 

[daffy duck]

With AI piloting fighters, you can probably create other airplane designs. Because an AI does not have to worry about (especially negative) G forces that much.

Remember that fighter the US built,with the forward swept wings that was ridiculously maneuverable but was unable to be flown by a human pilot it was so unstable, it was a solution to a problem ahead of it's time. Now an AI pilot could allow this design's potential to be fully realised

 

[Endymio]

I guess we should just fire all humans and let AI do everything.
Don't you think the AI hype is getting insane.

The USAF just demonstrated quite to the contrary that AI is not simply "hype".

 

[Watzupken]

I hope they understand that if the system gets hacked, your friendlies will rain destruction on themselves. There is a reason why you don’t want to remotely control something that is high risk.

 

[Endymio]

I hope they understand that if the system gets hacked, your friendlies will rain destruction on themselves. There is a reason why you don’t want to remotely control something that is high risk.

We've had remote controlled nuclear weapons for quite a while now. You can argue that there are multiple live people in the loop, but at the lowest level is an electronic circuit that, if somehow hijacked, would launch regardless.

It doesn't happen because systems are built with multiple, multiple layers of security. At the very least, an AI fighter would surely have a totally-independent hardwired IFF transponder that, regardless of what the AI brain wishes to do, won't allow a launch against a friendly target.

 

[nitebird]

At the very least, an AI fighter would surely have a totally-independent hardwired IFF transponder that, regardless of what the AI brain wishes to do, won't allow a launch against a friendly target.

That means that if an unfriendly country manages to spoof that IFF, they can shoot down our AI fighters during a real war.

 

[godrilla]

So we're just gonna ignore the last 30 years of Sci Fi movies? Gotcha. Nothing could ever go wrong here. Moving on...

The organics are immune to emp vulnerability and can with proper skill still use analog controls and safely glide the jet to safety.
If used in copilot .hybrid mode with a human where the human has the last say whether to engage or override the ai once being hacked with master override. The human can also use the ai as last resort to tap out to safety and have the ai engage the enemies where chances of organic survival are non existing.
Each scenario has its weakness.

We've had remote controlled nuclear weapons for quite a while now. You can argue that there are multiple live people in the loop, but at the lowest level is an electronic circuit that, if somehow hijacked, would launch regardless.

It doesn't happen because systems are built with multiple, multiple layers of security. At the very least, an AI fighter would surely have a totally-independent hardwired IFF transponder that, regardless of what the AI brain wishes to do, won't allow a launch against a friendly target.

True but if in the test the ai vs human, the allied human was made an enemy with a click of a button so it's not impossible. Also it's not implossible to reverse ingeneer this in the wrong hands. In the wrong hands this ai flying shells with rockets can be disastrous.

 

[Endymio]

That means that if an unfriendly country manages to spoof that IFF, they can shoot down our AI fighters during a real war.

Of course. But that applies AI or not. If an enemy spoofs our IFF systems, it'll affect human-piloted craft just as severely.

The organics are immune to emp vulnerability and can with proper skill still use analog controls and safely glide the jet to safety.

There are no analog controls on modern fighters: it's all fly-by-wire. If the electronics go out, the craft is dead.

True but if in the test the ai vs human, the allied human was made an enemy with a click of a button so it's not impossible

We intended that "allied human" to be identified as an enemy and it was. That's not comparable to an enemy hijacking the system. Your argument is like saying that, because a US President once ordered a nuclear strike, it proves that an enemy could spoof our system and cause us to nuke ourselves.

 

[godrilla]

Of course. But that applies AI or not. If an enemy spoofs our IFF systems, it'll affect human-piloted craft just as severely.


There are no analog controls on modern fighters: it's all fly-by-wire. If the electronics go out, the craft is dead.


We intended that "allied human" to be identified as an enemy and it was. That's not comparable to an enemy hijacking the system. Your argument is like saying that, because a US President once ordered a nuclear strike, it proves that an enemy could spoof our system and cause us to nuke ourselves.

I'll admit I probably watched to many action movies. It's official no one would reverse engineer or hack this ai and then blame it on us for starting ww3. Fingers crossed .

 

[loki1944]

The USAF just demonstrated quite to the contrary that AI is not simply "hype".

Lol; US airforce has not fought a near peer war since Vietnam. And that wasn't full bore either. Even Korean Warvwas nothijg near true airland batttle. You have to go to WW2 where strategies/tactics/CMs changed/evolved constantly. And don't forget it's fake ai made by government contract "quality ".