DARPA AI fighter defeats F-16 pilot in virtual aerial combat

[Ant-man]

https://www.thedrive.com/the-war-zone/35888/ai-claims-flawless-victory-going-undefeated-in-digital-dogfight-with-human-fighter-pilot
 

https://youtu.be/McH63bA9TDU

 

Have we learned nothing from Terminator!!!??

Edited August 20, 2020 by Ant-man

[ToHoldShort]

AI needs to go read the 214. All about those 18AA gunshots

[Sprkt69]

I’d like to see the AI try that in a real airplane

[Majestik Møøse]

Where was the AI getting the human PPLI from? The human was using his eyeballs, I suspect the AI had perfect data.

[Inertia17]

AI had perfect data. I have no idea how they would get this info airborne. I'd imagine that will be a long way off.

[Chuck17]

Just now, Inertia17 said:

AI had perfect data. I have no idea how they would get this info airborne. I'd imagine that will be a long way off.

 

[Inertia17]

1 minute ago, Chuck17 said:

 

Right. I wasn't particularly clear. What I meant was how they will assess the adversary airborne with enough fidelity for the algorithm.

[Chuck17]

Yep I’m with you, tweet provided for context only!

Cheers,

Chuck

[brickhistory]

As Lt Data and his companions come on board, many things will become more efficient for Big Blue.

Of course, finding snackos will be come harder and harder.

As will filling the ranks of human leadership.  Imagine the perfect world where only shiny pennies are brought in to command legions of AI systems.  

Cuts down on Tricare/retirement costs as well.

Win/win for Big Blue and Pentagon.

If it's just 'things' getting blown up, will anyone care anymore?  Will war become a more or less viable option?

[1:1:1]

4 minutes ago, brickhistory said:

Will war become a more or less viable option?

Judging by the number of countries our UAVs operate in, I don't think there's much question there.

[fox two]

[fox two]

7 hours ago, brickhistory said:

As Lt Data and his companions come on board, many things will become more efficient for Big Blue.

Of course, finding snackos will be come harder and harder.

As will filling the ranks of human leadership.  Imagine the perfect world where only shiny pennies are brought in to command legions of AI systems.  

Cuts down on Tricare/retirement costs as well.

Win/win for Big Blue and Pentagon.

If it's just 'things' getting blown up, will anyone care anymore?  Will war become a more or less viable option?

Nerd alert:

Star Trek's classic "A Taste of Armageddon" episode explores this. Two societies use computers to simulate the outcome of a war. The winner is declared without any fighting. But the losers are euthanized.

Edited August 21, 2020 by fox two

[brickhistory]

Well done.  I remembered that episode watched, probably repeatedly as a kid growing up on afternoon reruns.

 

And while it is an analogy, I wasn't necessarily going for the WMD type of unmanned weapons.

I inarticulately tried to bring up the "if it's all 'droid warfare - air/land/sea/space - would political leaders really hesitate as much as they might, in Western societies anyway, regarding meat-servo casualties?  Dunno.

Does the escalation ramp become "He destroyed so many of my droids that I have to go nuclear" instead of "Xi just killed a carrier and now I've got no choice?"

Either way, Lt Data will not be refilling the popcorn machine, I'm betting...

 

edited to add:  The summer drink of choice in the brick household this season is lime rum/coconut rum/fresh squeezed lime/some sparkling lime flavored water.

Repeat as necessary until it doesn't matter.

 

Like about now...

Edited August 21, 2020 by brickhistory

[busdriver]

Early in the video the Col talks about auto GCAS.  I suspect they're not thinking about making HAL right now, rather the thought process is more like the auto aiming setting in Halo.

[M2]

[Vimix22]

On 8/21/2020 at 12:14 PM, Inertia17 said:

AI had perfect data. I have no idea how they would get this info airborne. I'd imagine that will be a long way off.

In a visual engagement it could be a 360 camera with laser ranging? That would probably be enough data for the AI. Can you imagine having a BFM drone flying next to your 6th Gen fighter? Give it the Banzai command, get visual EID through your data link to the drone and watch it slay everything.

[FLEA]

19 minutes ago, Vimix22 said:

In a visual engagement it could be a 360 camera with laser ranging? That would probably be enough data for the AI. Can you imagine having a BFM drone flying next to your 6th Gen fighter? Give it the Banzai command, get visual EID through your data link to the drone and watch it slay everything.

And I would immagine what the drone lacks in sensory/data collection can be partially made up for in greatly enhanced performance characteristics. From what I gathered from the video the AI was flying another F-16? But a purpose designed drone without life support, pink bodies, ejection seats, glass canopy, displays, etc..... Opens up some advantages in airframe design for sure. 

[Smokin]

It was a computer program running a computer program vs a pilot running a computer program.  The results are so predictable a 5 year old could have told you who would win.  Simulators are not real airplanes.  The AI would not have the data in an airplane it had in the sim; perfect performance modeling of both jets, perfect data on adv airspeed, alt, g, distance, heading, AA, etc.  Probably even knew exactly what flight control inputs the pilot was making before they would have been apparent visually.  A laser and a camera doesn't get you that data.

The entire thing was done with the stated purpose to give confidence in a drone wingman for future fighter pilots.  A cagematch in a sim does not do that, so the stated objective was not possible to meet.  The entire thing smacked of purely a publicity stunt.

[FLEA]

2 hours ago, Smokin said:

It was a computer program running a computer program vs a pilot running a computer program.  The results are so predictable a 5 year old could have told you who would win.  Simulators are not real airplanes.  The AI would not have the data in an airplane it had in the sim; perfect performance modeling of both jets, perfect data on adv airspeed, alt, g, distance, heading, AA, etc.  Probably even knew exactly what flight control inputs the pilot was making before they would have been apparent visually.  A laser and a camera doesn't get you that data.

The entire thing was done with the stated purpose to give confidence in a drone wingman for future fighter pilots.  A cagematch in a sim does not do that, so the stated objective was not possible to meet.  The entire thing smacked of purely a publicity stunt.

I think we are all aware of that but the accomplishment itself is a huge step forward. This is a lot farther than teaching a computer to beat a chess champion. A computer learning to navigate a 3 dimensional space is a pretty significant step forward. 

[Ant-man]

My father-in-law is working on machine learning for AFRL in regards to future aircraft development. As stated above, this experiment was more about demonstrating the capability of an AI to make real-time decisions based on human inputs. The Air Force’s scope of interest in AI extends beyond just the ACE (DARPA’s Air Combat Evolution) program, and I think there are some interesting albeit scary future capabilities being discussed. Of particular note is the R2-D2 program, which is just like it sounds: an AI navigator/FE/copilot. I’m interested to see if any of this plays into the B-21’s proposed capabilities.  

[Majestik Møøse]

6 hours ago, FLEA said:

And I would immagine what the drone lacks in sensory/data collection can be partially made up for in greatly enhanced performance characteristics. From what I gathered from the video the AI was flying another F-16? But a purpose designed drone without life support, pink bodies, ejection seats, glass canopy, displays, etc..... Opens up some advantages in airframe design for sure. 

Just imagine if the drone was rocket-powered instead of air breathing! Like Mach 4 speed. And I bet we could make it pull 30g. And of course, there’s no human in there so we could give it a one-way mission. So we’d better make them cheap - like less than $1m each - and light so one fighter could control 8 or even as many as 20 of them. They can even have little radars or IR sensors in their noses, but they wouldn’t need them until close in because the controlling fighter (the mothership!) is guiding them most of the way to their target.

[SurelySerious]

5 hours ago, Majestik Møøse said:

Just imagine if the drone was rocket-powered instead of air breathing! Like Mach 4 speed. And I bet we could make it pull 30g. And of course, there’s no human in there so we could give it a one-way mission. So we’d better make them cheap - like less than $1m each - and light so one fighter could control 8 or even as many as 20 of them. They can even have little radars or IR sensors in their noses, but they wouldn’t need them until close in because the controlling fighter (the mothership!) is guiding them most of the way to their target.

Dude, what if the fighter that launched that drone didn’t have to be the one controlling that drone? Like via some kind of network, information from another airplane could be passed to the drone? Nah, won’t happen in our lifetime. 

[HuggyU2]

The reality is that the AI jet will win the fight... only to mort on the RTB when its sense-and-avoid system doesn't see the MQ-9.  

[GKinnear]

8 hours ago, HuggyU2 said:

The reality is that the AI jet will win the fight... only to mort on the RTB when its sense-and-avoid system doesn't see the MQ-9.  

To be fair...the MQ-9 won't see it either

[Breckey]

https://www.militarytimes.com/news/your-military/2020/08/18/two-us-drones-collide-and-crash-over-idlib-syria-defense-official-says/