NEWS AI versus F-16 Pilot

[nittany03]

I have. Sounds like yours are just pessimists or have been too long in government.

It sounds like you're stretching the ragged edge of your credibility on the subject. Any doubt I had was flushed when you started talking about rearward-firing guns like a 7th grader at an airshow. This thread was started by someone who has actual experience in fleet BFM. I see other usernames in it who also do. You . . . don't.

I say again . . . "Donny, you're out of your element." Listen and learn.

 

[Treetop Flyer]

I have. Sounds like yours are just pessimists or have been too long in government.

Human-in-the-loop or human-on-the-loop

Can you show me an example of an aircraft that is high speed, long range, high altitude, low RCS, but also small and cheap? Those things are mutually exclusive

 

[Brett327]

Sounds like yours are just pessimists or have been too long in government.

The list of mutually exclusive features you provided is preposterous. That's not pessimism... that's physics.

 

[IwannabeaPHROGdvr69]

 

[Hair Warrior]

The list of mutually exclusive features you provided is preposterous. That's not pessimism... that's physics.

How does a falcon or eagle soar at altitude at slow speeds with little effort, high above its prey, then accelerate rapidly down to strike its prey accurately at crazy speeds?

 

[Brett327]

How does a falcon or eagle soar at altitude at slow speeds with little effort, high above its prey, then accelerate rapidly down to strike its prey accurately at crazy speeds?

Really?

 

[nittany03]

How does a falcon or eagle soar at altitude at slow speeds with little effort, high above its prey, then accelerate rapidly down to strike its prey accurately at crazy speeds?

Uhh . . . by trading altitude for airspeed? A peregrine falcon doesn't come equipped with a freaking afterburner. I can't believe this question is being asked, let alone that I'm answering it.

I once worked with a tactics instructor who had an acronym for situations like this: an SLOC . . . Sudden Loss Of Credibility.

 

[Farva01]

Overall I was impressed with the study.
We do a fair job on training to BFM in the Navy. The argument that the dogfight is dead is for another debate, however, BFM is a great trainer to understand your energy state in relation to the enemy and to exercise your situational awareness in a 3D space. You have to aggressively and quickly cue your weapons, assess the quality of your cueing, and determine if you should pull the trigger to employ that weapon, all while at the same time preventing the enemy from doing the same. I have yet to meet a pilot who is above average at BFM, but isn't good at all other mission sets.
I think that sometimes we lose the forest through the trees for what we are trying to actually accomplish when we hit the merge: to get the first shot off. Quite often you complete an engagement, and your opponent is proud of the shots they took on the deck. They don't realize that they were shot multiple times before reaching there. Its pretty hard to fight your best 1v1 when you are missing a wing or an engine.
But you cannot just put your lift vector on the other jet and pull and hope to win the fight. That will solve about 75% of your problems, but the truly great BFM pilots are thinking two to three merges ahead. You have to be smoothly aggressive. Yank on the G's when required, but knowing how to preserve energy smartly and regain energy efficiently is the key. In a tight fight, gaining a small amount of angles at each merge is how you end offensive on the deck.
So what did the AI really well? Forward quarter gun shots. That amount of closure requires intense concentration and refinement of pipper placement, while also trying to avoid hitting the other aircraft. Forward quarter gun employment is not allowed in training because of the high midair potential. AI isn't worried about morting itself, and can accurately and quickly acquire a forward quarter gun shot. The fight was only guns, but I have no doubt that AI could employ a valid AIM-9X shot quicker than I can.
Energy preservation. AI will always be able to fly the appropriate G or Alpha required to maintain the perfect energy package. A human has to dedicate a huge amount of brain space to make sure they are flying the aircraft correctly. Especially a fly-by-wire aircraft where there isn't as much "feel" in how the aircraft is performing. I have to constantly check to make sure I am flying the correct alpha in a slow fight.
High performance turning. During one of the fights, the tow aircraft are in a prolonged 9G 2-circle fight on the deck. Humans cannot sustain that. AI doesn't have to focus on breathing and squeezing.
So yes, the human was fighting on the computer's terms. The computer had perfect information that it wouldn't necessarily have without a robust sensor package. ROE isn't a factor, blah blah. But this shows some potential. As an example, having the AI supplement my weapons employment could be a powerful combo.

 

[Hair Warrior]

Uhh . . . by trading altitude for airspeed? A peregrine falcon doesn't come equipped with a freaking afterburner. I can't believe this question is being asked, let alone that I'm answering it.

I once worked with a tactics instructor who had an acronym for situations like this: an SLOC . . . Sudden Loss Of Credibility.

Right... it’s a one way descent. Does it need to go back up in altitude if you have three more behind it? Ten more? So, “physics.”

My larger point, which I am poorly communicating, is don’t discount the possibility of masses of cheap, expendable, unmanned vehicles of all shapes and sizes showing up in the near future battlespace. When you take the human and human-supporting features out of the product design, you can radically change the design and make something pretty damn lethal in the aggregate
. Just because the US military isn’t doing it today doesn’t mean adversaries won’t. If a manned aircraft is facing an AI-piloted air superiority fighter in 2040, the unmanned airframe won’t look like a QF-16. It’s going to be a lot lighter weight, longer endurance, fewer/no countermeasures (ECM, chaff, flares). Less redundancy of wiring/lines built into its internals. No martin baker. Heck, maybe no landing gear (ScanEagle doesn’t have any either). You can get a lot of lethality when you rethink an air superiority drone from the ground up. And yeah, you can certainly trade altitude for airspeed fifty times in a row if you have fifty drones up there and they each do it once. Those fifty drones may not get many kills against a pointy nose manned aircraft but maybe the adversary only needs them to go after our AWACS, tankers, P-8s, RC-130s, AC-130s, and other big wings.

 

[Hair Warrior]

Where you are getting pushback is the stuff you described literally goes against existing principles of physics and engineering in getting high performance out of low cost. Those are mutually exclusive.
To make it simple: taking a person out of a car, even redesigning it from the ground up, even removing airbags/crumple zones/HVAC/etc still limits you to the physical constraints of engine technology and fuel or battery consumption.

Roger. I realize what I am describing doesn’t exist today and seems impossible to engineer.

Your car analogy is apt. Have you ever ridden in a Tesla when it goes 0-60 on ludicrous mode?

And, that thing drives itself autonomously, doesn’t burn gas, and is comparable in price to other cars.

 

[RedFive]

Right... it’s a one way descent. Does it need to go back up in altitude if you have three more behind it? Ten more? So, “physics.”

My larger point, which I am poorly communicating, is don’t discount the possibility of masses of cheap, expendable, unmanned vehicles of all shapes and sizes showing up in the near future battlespace. When you take the human and human-supporting features out of the product design, you can radically change the design and make something pretty damn lethal in the aggregate
. Just because the US military isn’t doing it today doesn’t mean adversaries won’t. If a manned aircraft is facing an AI-piloted air superiority fighter in 2040, the unmanned airframe won’t look like a QF-16. It’s going to be a lot lighter weight, longer endurance, fewer/no countermeasures (ECM, chaff, flares). Less redundancy of wiring/lines built into its internals. No martin baker. Heck, maybe no landing gear (ScanEagle doesn’t have any either). You can get a lot of lethality when you rethink an air superiority drone from the ground up. And yeah, you can certainly trade altitude for airspeed fifty times in a row if you have fifty drones up there and they each do it once. Those fifty drones may not get many kills against a pointy nose manned aircraft but maybe the adversary only needs them to go after our AWACS, tankers, P-8s, RC-130s, AC-130s, and other big wings.

Dude. We're pilots. We know.

 

[Treetop Flyer]

Roger. I realize what I am describing doesn’t exist today and seems impossible to engineer.

Your car analogy is apt. Have you ever ridden in a Tesla when it goes 0-60 on ludicrous mode?

And, that thing drives itself autonomously, doesn’t burn gas, and is comparable in price to other cars.

And yet it’s just a car. It’s not small, it’s fast off the line but not as much at speed, it doesn’t have longer range than comparable cars, it’s not cheap, it’s not low RCS, it has to warm up to do its 0-60 party trick while other comparable cars can just stomp the gas any time.

So yeah. Not a great example, and there won’t be one because what you described is fucking fantasy.

 

[Swanee]

Overall I was impressed with the study.
We do a fair job on training to BFM in the Navy. The argument that the dogfight is dead is for another debate, however, BFM is a great trainer to understand your energy state in relation to the enemy and to exercise your situational awareness in a 3D space. You have to aggressively and quickly cue your weapons, assess the quality of your cueing, and determine if you should pull the trigger to employ that weapon, all while at the same time preventing the enemy from doing the same. I have yet to meet a pilot who is above average at BFM, but isn't good at all other mission sets.
I think that sometimes we lose the forest through the trees for what we are trying to actually accomplish when we hit the merge: to get the first shot off. Quite often you complete an engagement, and your opponent is proud of the shots they took on the deck. They don't realize that they were shot multiple times before reaching there. Its pretty hard to fight your best 1v1 when you are missing a wing or an engine.
But you cannot just put your lift vector on the other jet and pull and hope to win the fight. That will solve about 75% of your problems, but the truly great BFM pilots are thinking two to three merges ahead. You have to be smoothly aggressive. Yank on the G's when required, but knowing how to preserve energy smartly and regain energy efficiently is the key. In a tight fight, gaining a small amount of angles at each merge is how you end offensive on the deck.
So what did the AI really well? Forward quarter gun shots. That amount of closure requires intense concentration and refinement of pipper placement, while also trying to avoid hitting the other aircraft. Forward quarter gun employment is not allowed in training because of the high midair potential. AI isn't worried about morting itself, and can accurately and quickly acquire a forward quarter gun shot. The fight was only guns, but I have no doubt that AI could employ a valid AIM-9X shot quicker than I can.
Energy preservation. AI will always be able to fly the appropriate G or Alpha required to maintain the perfect energy package. A human has to dedicate a huge amount of brain space to make sure they are flying the aircraft correctly. Especially a fly-by-wire aircraft where there isn't as much "feel" in how the aircraft is performing. I have to constantly check to make sure I am flying the correct alpha in a slow fight.
High performance turning. During one of the fights, the tow aircraft are in a prolonged 9G 2-circle fight on the deck. Humans cannot sustain that. AI doesn't have to focus on breathing and squeezing.
So yes, the human was fighting on the computer's terms. The computer had perfect information that it wouldn't necessarily have without a robust sensor package. ROE isn't a factor, blah blah. But this shows some potential. As an example, having the AI supplement my weapons employment could be a powerful combo.

AI is really good at Injesting sensor data and making decisions based on a matrix, as well as flying the numbers to a T. (It does what it's told, at times to a fault.)

It struggles with multiple outcome decisions where something is ambiguous. But we do too. Using AI to enhance our capabilities is the step. If you could have AI that worked a sensor and flew the airplane better than any wingman, and did what you told them to, and only made decisions that you told it that it could make, and bring the rest to you, you'd have a pretty lethal AI flown airplane.

It's in the works. Albeit it with UAVs, (hey robot, go here, look for this, if you see X, tell me immediately, otherwise talk to me in 10 minutes. And don't shoot anything without asking me first).

And AI won't fuck up and shoot a bogey going into the merge.

 

[nittany03]

Right... it’s a one way descent. Does it need to go back up in altitude if you have three more behind it? Ten more? So, “physics.”

My larger point, which I am poorly communicating, is don’t discount the possibility of masses of cheap, expendable, unmanned vehicles of all shapes and sizes showing up in the near future battlespace. When you take the human and human-supporting features out of the product design, you can radically change the design and make something pretty damn lethal in the aggregate.

These already exist. They’re called “missiles.” Heck, you could argue that Tomahawks are in actuality some type of UAV. "Aircraft" to "UAS" to "missile" is and has always been a continuum, and one that's always worth reassessing as technology improves. But the key word here is "evolutionary," not "revolutionary." What @Farva01 and @Swanee are talking about is a credible use of AI/ML from folks who have the professional experience to explain the art of the possible in 2020-2030.

 

[whitesoxnation]

AI is really good at Injesting sensor data and making decisions based on a matrix, as well as flying the numbers to a T. (It does what it's told, at times to a fault.)

It struggles with multiple outcome decisions where something is ambiguous. But we do too. Using AI to enhance our capabilities is the step. If you could have AI that worked a sensor and flew the airplane better than any wingman, and did what you told them to, and only made decisions that you told it that it could make, and bring the rest to you, you'd have a pretty lethal AI flown airplane.

It's in the works. Albeit it with UAVs, (hey robot, go here, look for this, if you see X, tell me immediately, otherwise talk to me in 10 minutes. And don't shoot anything without asking me first).

And AI won't fuck up and shoot a bogey going into the merge.

Just imagine if the logic was hostile = shoot and somehow the all friendlies get dec’d hostile in a shooting platform. Interpretation of sensors is not always right, be it eye balls or electronic means.