It's happening: AI vs Human

New and old developments in aviation technology.
  • Author
  • Message
Offline

nutshell

Senior member

Senior member

  • Posts: 484
  • Joined: 04 May 2016, 13:37

Unread post27 Jul 2020, 20:49

I didn't find anything about this issue, so i'm going to post it.

https://www.airforcemag.com/air-force-t ... man-pilot/

Not a last minute news but still quite fresh.

In a year from now, the US are going to make Elon Musk's military wet dream to come true: fighter drones.

I don't really know what to think. By listening to the many fighter pilots podcasts, we're not really pushing the human body so hard it's a limiting factor, maybe is it going to be a nice way to save money?

Smaller aircraft, better fuel efficiency, expendable, saves time spent on pilot training.

Combat wise, i can't see anything else but a downgrade.

Thoughts?
Offline

hornetfinn

Elite 3K

Elite 3K

  • Posts: 3183
  • Joined: 13 Mar 2013, 08:31
  • Location: Finland

Unread post28 Jul 2020, 12:42

I don't think real fighter drones will become a real option to manned fighter aircraft within several decades. IMO, first there will be manned fighter aircraft with increasing capabilities through AI and associated technologies. I see loyal wingman drones becoming more widespread but they will not replace manned fighter aircraft but rather complement their capabilities. I think most promising use of these drones would be SEAD/DEAD, EW and CAS missions. Drones could do the most dangerous things like getting close to enemy, act as a decoy or emit signals.

Elon Musk is right that fully autonomous fighter aircraft could be made a lot more maneuverable than manned fighter aircraft. However that alone would be of little use if those drones can't get close enough to use that advantage. It will be tough and long road to make them capable of doing all kinds of missions reliably and in large numbers and co-operating with each other effectively.

Of course someday we will definitely see fully autonomous fighter aircraft controlled by very advanced AI which can beat human beings in pretty much everything. That will take a lot of time and money and IMO we are still taking first real steps towards that direction. I doubt that will happen within the next 20 years at least and might take longer than that.
Offline

nutshell

Senior member

Senior member

  • Posts: 484
  • Joined: 04 May 2016, 13:37

Unread post28 Jul 2020, 15:54

hornetfinn wrote:I don't think real fighter drones will become a real option to manned fighter aircraft within several decades. IMO, first there will be manned fighter aircraft with increasing capabilities through AI and associated technologies. I see loyal wingman drones becoming more widespread but they will not replace manned fighter aircraft but rather complement their capabilities. I think most promising use of these drones would be SEAD/DEAD, EW and CAS missions. Drones could do the most dangerous things like getting close to enemy, act as a decoy or emit signals.

Elon Musk is right that fully autonomous fighter aircraft could be made a lot more maneuverable than manned fighter aircraft. However that alone would be of little use if those drones can't get close enough to use that advantage. It will be tough and long road to make them capable of doing all kinds of missions reliably and in large numbers and co-operating with each other effectively.

Of course someday we will definitely see fully autonomous fighter aircraft controlled by very advanced AI which can beat human beings in pretty much everything. That will take a lot of time and money and IMO we are still taking first real steps towards that direction. I doubt that will happen within the next 20 years at least and might take longer than that.


The ACM argument got debunked by many fighter pilots (there's an interview about that on Mover's YT channel). A fighter pilot is far away from physical limits when that happens and that's the worst case scenario, ACMs.

I still think the biggest point on favor of combat drones is that you can make smaller aircraft (no cockpit/canopy), carrying more fuel and with a tougher superstructure. If anything, those drones should last longer than a typical manned fighter.
Another one is that being intrinsically lighter. it can also lead to better fuel economy (which i guess is one of the main concern in year 2020).

OTOH a combat drone needs special attention to jamming, increase reliancy to a lower orbit, satellite constellation (which is going to be really hard to defend) and forces the "IT Guys" to fight the ever ending war of bandwidth constraints.
The last part is crucial, since, as i said, a high performing network is going to be exposed to the attacks from the enemy.

Probably 60 years too early.
Offline

marauder2048

Elite 1K

Elite 1K

  • Posts: 1288
  • Joined: 14 Mar 2012, 06:46

Unread post28 Jul 2020, 18:09

If you look at typical weight due to the pilot plus his support equipment it's around 1000 pounds.
And for typical cockpit volumes, assuming the UCAV has a flush design, the reclaimed volume is
not large say: 200 gallons. If you could convert all of that to JP-8 that's around 1300 pounds of fuel.

So you are really talking about single digit percentage improvements in empty weight and
internal fuel. It's nothing to sneeze at but's not the sort of gain it's often made out to be.
Offline

sprstdlyscottsmn

Elite 4K

Elite 4K

  • Posts: 4878
  • Joined: 10 Mar 2006, 01:24
  • Location: Phoenix, Az, USA

Unread post28 Jul 2020, 18:33

I think the real advantage is the ability to monitor air data, positional data, and performance data at absurd speeds. If an AI and a Human Pilot both decide that a full power, sustained speed turn, is the best course for the moment the AI will get there more quickly and hold it more precisely.
"Spurts"

-Pilot
-Aerospace Engineer
-Army Medic
-FMS Systems Engineer
Offline

hornetfinn

Elite 3K

Elite 3K

  • Posts: 3183
  • Joined: 13 Mar 2013, 08:31
  • Location: Finland

Unread post29 Jul 2020, 11:59

I think the biggest advantage what fully unmanned drones could have is that they would not need years of constant training and enough flight hours to keep those skills up. So O&S costs could be very small (just enough to keep everything running) and you could just upload a "pilot" to a jet within seconds and the skills would not deteriorate even if it didn't fly for decades. This would also allow them keeping up very high operating tempo a lot longer. Maybe we will also see maintenance done by AI controlled robots. Some day something that will very likely happen but I bet we will have pilots in fighter jets at least until F-35 is phased out.
Offline

sprstdlyscottsmn

Elite 4K

Elite 4K

  • Posts: 4878
  • Joined: 10 Mar 2006, 01:24
  • Location: Phoenix, Az, USA

Unread post29 Jul 2020, 12:42

So many of us are thinking about "the mission" that we completely forget about the lack of training and skills maintenance needed. Excellent point.
"Spurts"

-Pilot
-Aerospace Engineer
-Army Medic
-FMS Systems Engineer
Offline

madrat

Elite 2K

Elite 2K

  • Posts: 2763
  • Joined: 03 Mar 2010, 03:12

Unread post29 Jul 2020, 13:17

Drones do not need to win 1:1. Drones can offer 10:1 exchange ratio or better, with inferior capabilities, and inefficient technique, because it only needs to win once. The human will fatigue.
Offline

basher54321

Elite 2K

Elite 2K

  • Posts: 2054
  • Joined: 02 Feb 2014, 15:43

Unread post07 Aug 2020, 22:04

WASHINGTON ― An artificial intelligence algorithm will face off against a human F-16 fighter pilot in an aerial combat simulation in late August, the U.S. Defense Advanced Research Projects Agency announced Aug. 7.

The simulation — the third and final competition in DARPA’s AlphaDogfight Trials — will take place Aug. 20. The event will be virtual due to the ongoing coronavirus pandemic.

The AlphaDogfight Trials was created to demonstrate advanced AI systems’ ability in air warfare. Eight teams were selected last year to participate in the final competition that runs from Aug. 18-20. The competition is also part of DARPA’s Air Combat Evolution, or ACE, program, which was started in 2019, and seeks to automate air-to-air combat as well as improve human trust in AI systems to bolster human-machine teaming.

“We weren’t able to host the finals at AFWERX in Las Vegas as we’d originally planned with fighter pilots from the Air Force Weapons School at nearby Nellis Air Force Base,” Col. Dan Javorsek, program manager in DARPA’s Strategic Technology Office, said in a statement. “We are still excited to see how the AI algorithms perform against each other as well as a Weapons School-trained human and hope that fighter pilots from across the Air Force, Navy, and Marine Corps, as well as military leaders and members of the AI tech community will register and watch online. It’s been amazing to see how far the teams have advanced AI for autonomous dogfighting in less than a year.”

[...]

Source: https://www.c4isrnet.com/battlefield-te ... alflow+AIR
Offline

mmm

Enthusiast

Enthusiast

  • Posts: 68
  • Joined: 30 Jul 2013, 10:58

Unread post08 Aug 2020, 02:00

By being a good contrarian I say why not, close maneuvering dogfight is most simple of all tasks to be taken over by AI. You deal with little terrain and minimum computer vision challenge, not that AI is not already used to save human from ground. 25 year old game with somewhat simple and very predictable flight model already routinely dogfight with guns only to almost perfection, on consumer hardware, and now they can do abstraction with "Moore's law" times more processing power. I mean human has to use "experience" to guess what's the end state after a maneuver, but how does it compared to simulate with high fidelity flight model? They also have advantages like not tunnel vision in a 4 ship furball, even if you have DAS with spherical vision pilot is not going to take all that information into account and modify plan every 1/50s. With the ability to do abstraction AI don't necessarily need measured velocity and range data and remain more accurate than human "eyeballing". To use a potentially controversial analogy I think dogfight is quickly becoming the "calculator problem" for AI now, something human is not going to match the efficiency for very soon.

Human can remain in the cockpit, there are still many good reasons for it(but with DEW probably against as well), but it might not be that long before the standard procedure in a maneuvering dogfight becomes "press the automatic dogfight" button on the touch screen and sit it out.
Offline

zhangmdev

Active Member

Active Member

  • Posts: 182
  • Joined: 01 May 2017, 09:07

Unread post08 Aug 2020, 04:53

Please don't use the buzzword AI, which is a cheap marketing gimmick. There is not "I" in that. Just a bunch of machine learning algorithms trying, or pretending, to mimic how human brain, billions and billions of neurons, works. Current technology cannot reliably drive a car on paved roads without occasional fatal crashes. After some of the richest corporations pouring billions upon billions of US dollars into this area of research, for decades, still the fancy "autopilot" needs a human behind the wheel, just in case it goes crazy and decides to run into a parked truck or something. The problem is not the sensors are not good enough, or the computers are not fast enough. The heart of the problem is the computer has no idea of what a car, a human, a tree, a cow, or anything, is. It tries to discern objects by numbers. Pretty hilarious to us human, if you think about that, .454 is a cow, .545 is a moose. Human brain works in a quite different way. Exactly how it works, how information is processed, how memory is stored and retrieved, how concepts are formed, what is consciousness, scientists still have no idea.

Back to aviation. Besides takeoff and landing, machine can do pretty much everything about civil aviation, which is not much to take about. Keep that altitude, on that route, beep me if anything is in close proximity, here you go. Far easier than the mythical self-driving car. But still there is no commercial airliner flying pilotless, because there is no faith in computers to handle emergencies. If all engines are lost, the plane has to land on a river in the middle of a city, I am pretty sure no computer can do that. Wait, what is a river? or a city?

Military aviation is far more complicated than the civil one. For example, communicating with grunt on the ground, to get the coordinate of target, do you trust Alexa do that? More difficult questions, is that a guy carrying a RPG, or a kid playing with a stick? Is that a woman digging potatoes, or a guy planting IDE?

If you need some rain on the "AI" parade, there is a book
https://www.amazon.com/AI-Delusion-Gary ... 0198824300
Offline

basher54321

Elite 2K

Elite 2K

  • Posts: 2054
  • Joined: 02 Feb 2014, 15:43

Unread post08 Aug 2020, 11:56

mmm wrote:By being a good contrarian I say why not, close maneuvering dogfight is most simple of all tasks to be taken over by AI. You deal with little terrain and minimum computer vision challenge, not that AI is not already used to save human from ground. 25 year old game with somewhat simple and very predictable flight model already routinely dogfight with guns only to almost perfection, on consumer hardware, and now they can do abstraction with "Moore's law" times more processing power. I mean human has to use "experience" to guess what's the end state after a maneuver, but how does it compared to simulate with high fidelity flight model? They also have advantages like not tunnel vision in a 4 ship furball, even if you have DAS with spherical vision pilot is not going to take all that information into account and modify plan every 1/50s. With the ability to do abstraction AI don't necessarily need measured velocity and range data and remain more accurate than human "eyeballing". To use a potentially controversial analogy I think dogfight is quickly becoming the "calculator problem" for AI now, something human is not going to match the efficiency for very soon.


It is possible but still a lot of unknowns seems simplistic - until you start working on the many challenges there are at something so complex.

With a system like DAS that can provide some parameters constantly (assuming it reliably can) then this could provide enough info to give an advantage to the numbers game.

If however AI has no parameters after the merge and has to make guesses based on optical only (like a set of eyes) that that is an entirely different ball game.


Probably misleading to try and compare with AI in flight sims because apart from not being AI in any real sense and having simplistic flight models that gave them an advantage - they also know the state of every object in the system. So they have access to things that cannot be ascertained like current weight of a player aircraft and everything else (e.g. player is at a weight I cannot out turn) . So it is entirely illusionary which is understandable because they are supposed to be mimicking human BFM in pre DAS era aircaft where eyes would be used and so have all these cheats to compensate.
Offline

outlaw162

Elite 1K

Elite 1K

  • Posts: 1494
  • Joined: 28 Feb 2008, 02:33

Unread post08 Aug 2020, 16:55

I think dogfight is quickly becoming the "calculator problem" for AI now, something human is not going to match the efficiency for very soon.


I wonder what 'numbers' would be in play in an even scissors where the options come down to colliding with the other aircraft....or losing.

AI processing: "Oh no, not a .604, need input." :shock:
Offline
User avatar

element1loop

Elite 1K

Elite 1K

  • Posts: 1534
  • Joined: 31 Dec 2015, 05:35
  • Location: Australia

Unread post09 Aug 2020, 06:02

zhangmdev wrote:Military aviation is far more complicated than the civil one. For example, communicating with grunt on the ground, to get the coordinate of target, do you trust Alexa do that? More difficult questions, is that a guy carrying a RPG, or a kid playing with a stick? Is that a woman digging potatoes, or a guy planting IDE?

If you need some rain on the "AI" parade, there is a book
https://www.amazon.com/AI-Delusion-Gary ... 0198824300


Dunno about that take. At the moment active and passive sensors and back-end systems ID and classify contacts independently, and prioritize them, so I don't see this as an emerging problem for algos, within air combat, its a hurdle within a declared combat airspace which we already solve. These systems will make mistakes, humans also make human-errors.

The other point is who says the system will have command of the drone? As of now remote flown drones don't, so why would we need to let a mostly autonomous drone make final decisions on what is, and is not a hostile? We can still reserve final PID classification before firing. The machine doesn't actually need to 'know', and it does not have to be left to decide on key decisions about what's fired at.

It's a tool a human uses.
Accel + Alt + VLO + DAS + MDF + Radial Distance = LIFE . . . Always choose Stealth
Offline

basher54321

Elite 2K

Elite 2K

  • Posts: 2054
  • Joined: 02 Feb 2014, 15:43

Unread post20 Aug 2020, 21:36

Have been watching some of the final day stream.

One of the DARPA guys was laying out how he thought the future was manned - albeit with a slight difference.

The AI will do what it does best which includes flying the plane and all tactical maneuvres including BFM.

The Pilot will do the cognitive thinking and battlespace management. So the pilot become a passenger of sorts and can go from jet to jet without having to train for flying it - well that is the goal / dream I guess.

The Human pilot lost 5-0 to the Heron AI that won the rounds Vs the other AI teams.

The human pilot was at quite a disadvantage in a 3D computer sim especially if they are not used to VR headsets - takes a while to get used to and nothing like flying a jet where there is feel and sounds to give reference to G etc.

They did state the AI had full state like Ye old computer games thus always knew the state of the humans jet object - also difficult to say the level of real physics or accurate F-16 modelling looking at the 1997 level graphics. :P

Probably quite an achievement for a years worth of AI learning overall... maybe.
Next

Return to Technology

Who is online

Users browsing this forum: No registered users and 3 guests