Comment on First known test dogfight between AI and human pilot carried out, US military says
Blue_Morpho@lemmy.world 2 months ago
AI will win if not now, then soon. The reason is that even if it is worse than a human, the AI can pull off maneuvers that would black out a human.
Jets are far more powerful than humans are capable of controlling. Flight suits and training can only do so much to keep the pilot from blacking out.
Maneuverability is much less of a factor now as BVR engagements and stealth have taken over.
But, yeah, in general a pilot that isn’t subject to physical constraints can absolutely out maneuver a human by a wide margin.
The future generation will resemble a Protoss Carrier sans the blimp appearance. Human controllers in 5th and 6th Gen airframes who direct multiple AI wingman, or AI swarms.
Plus the ai has no risk, outside of basic operation.
Humans have an inherent survival instinct to which drones can just say “lol send the next one I’m dying cya”
To fight optimally, AI needs to have a survival instinct too.
Evolution didn’t settle on “protect my life at all costs” as our default instinct, simply by chance. It did so because it’s the best strategy in a hostile environment.
It’s the best strategy because it takes decades to make a fully functional human, and you need humans to make more humans, plus there is the issue of genetically sustainable population sizes, etc. A fully functional aeroplane can be made much quicker, in a factory that can spit out several of them in a day. They are more expendable.
Only if the goal is reproduction. You need to survive to reproduce.
If the goal is maximum damage for the least amount of economic cost then a suicide (anthropomorphizing the drone here) can very much make sense.
No one would argue that a sword is better than guns or bombs, because you still have the sword after attacking.
Jets are a lot more expensive. What's at risk is all these resources for the jet going down the drain.
Huh? Jets are far more replaceable than a human operator who takes years of training and has “needs”.
Ya know unless your military is running on cold war fumes or something and you can’t afford to.nuild an airframe you already have in production
Training a combat pilot used to cost (in early 2000, not sure now) 10M€ for a NATO member.
Find me a modern jet that costs so little. Regardless of what politicians say, human life has a price… and it is waaaay below a jet (even including the training)
I’d imagine a jet purpose built for an AI would be a lot cheaper than a human-oriented one. Removing the need for a cockpit with seats, displays, controls, oxygen, etc would surely reduce cost. It would also open the door for innovations in air-frame design previously impossible.
We keep talking like we’re discussing the future, but autonomous drones are already fighting in the skies of Ukraine.
Begun, the drone wars have.
Jets are in many ways expensive because they can’t be expendible. They also make an bunch of compromises to accommodate keeping a human alive.
For the cost of a single f22, you could put up 60 Valkyries. I think I know which side I would bet on.
Can’t they literally pull turns that would snap the pilots neck?
Can anyone confirm if AI has a neck?
My neck. My back. Lick my inputs & my headphone-jack.
You deserve an updoot for this. Enjoy.
And a mouth ( ͡° ͜ʖ ͡°)
ಠ_ಠ
But the supervising wingmen do.
Maybe if you were sitting sideways in the cockpit and did it very abruptly with the flight control computer disabled (only a few jets can even disable it). It’s the sustained G loading that makes you black out or red out.
A skilled and fit pilot can pull ~9G in a Viper for about 30s.
A computer can pull ~9G for as long has the plane has the speed to pull that hard, or it can pull as hard as it can until the plane snaps in half, because computers don’t suffer from g-LOC.
Not so much f16s but the more modern planes can do 16G where the pilot can’t really do more than 9G. But once unshackled from a pilot a lot of instrument weight and pilot survival can be stripped from a plane design and the airframe built to withstand much more, with titanium airframes I see no reason we can’t make planes do sustained unstable turns in excess of 20G.
We might have to start submerging pilots in breathable fluid.
Here comes the juice!
Jets are far more powerful than humans are capable of controlling. Flight suits and training can only do so much to keep the pilot from blacking out.
Can they be piloted remotely? Or would that be too dangerous with latency
Yes they can. Before AI the US was expecting to move to remote piloted jets
Expecting to? We’ve been using remote piloted jets for twenty years.
That’s not the case yet for fighters, just things like predator drones and global hawk
Latency, signal interference, and limited human intelligence are all limiting factors in that strategy.
If the enemy interferes with any of those, the enemy wins.
This was is already being fought with autonomous drones. By the end of it, the robots will be unrecognizable to us now.
You're better off with drones
What’s the difference? A remotely or AI-piloted fogger jet is just a big drone.
Drones are designed without cockpits. Retrofitting remote-control into an F-16 does not seem like the best choice to me.
The only thing that makes it bigger is the cockpit. There’s no difference.
But can a drone fly like an F-16?
If an f16 is remote controlled, it’s technically a drone.
AI will win if not now, then soon.
This article didn’t mention it but the AI pilot did win at least one of the engagements during this testing run.
I’d jump in and insert a major caution here.
I don’t know what is being done here, but a lot of the time, wargaming and/or military exercises are presented in the media as being an evaluation of which side/equipment/country is better in a “who would win” evaluation.
I’ve seen several prominent folks familiar with these warn about misinterpreting these, and I’d echo that now.
That is often not the purpose of actual exercises or wargames. They may be used to test various theories, and may place highly unlikely constraints on one side that favor it or the other.
So if someone says “the US fought China in a series of wargames in the Taiwan Strait and the US/China won in N wargames”, that may or may not be because the wargame planners were trying to find out who is likely to win an actual war, and may or may not have much to to with the expectations the planners have of a win in a typical scenario. They might be trying to find out what would happen in a particular scenario that they are working on and how to plan for that scenario. They may have structured things in a way that are not representative of what they expect to likely come up.
To pull up an example, here’s a fleet exercise that the US ran against a simulated German fleet between World War I and II:
en.wikipedia.org/wiki/Fleet_problem
Fleet Problem III and Grand Joint Army-Navy Exercise No. 2 edit
During Fleet Problem III, the Scouting Force, designated the “Black Force,” transited from its homeport in the Chesapeake Bay towards the Panama Canal from the Caribbean side. Once in the Caribbean, the naval forces involved in Fleet Problem III joined with the 15th Naval District and the Army’s Panama Division in a larger joint exercise.[9] The Blue force defended the canal from an attack from the Caribbean by the Black force, operating from an advance base in the Azores. This portion of the exercise also aimed to practice amphibious landing techniques and transiting a fleet rapidly through the Panama Canal from the Pacific side.[10]
Black Fleet’s intelligence officers simulated a number of sabotage operations during the course of Fleet Problem III. On January 14, Lieutenant Hamilton Bryan, Scouting Force’s Intelligence Officer, personally landed in Panama with a small boat. Posing as a journalist, he entered the Panama Canal Zone. There, he “detonated” a series of simulated bombs in the Gatun Locks, control station, and fuel depot, along with simulating sabotaging power lines and communications cables throughout the 16th and 17th, before escaping to his fleet on a sailboat.
On the 15th, one of Bryan’s junior officers, Ensign Thomas Hederman, also snuck ashore to the Miraflores Locks. He learned the Blue Fleet’s schedule of passage through the Canal from locals, and prepared to board USS California (BB-44), but turned back when he spotted classmates from the United States Naval Academy - who would have recognized and questioned him - on deck. Instead, he boarded USS New York (BB-34), the next ship in line, disguised as an enlisted sailor. After hiding overnight, he emerged early on the morning of the 17th, bluffed his way into the magazine of the No. 3 turret, and simulated blowing up a suicide bomb - just as the battleship was passing through the Culebra Cut, the narrowest portion of the Panama Canal. This “sank” New York, and blocked the Canal, leading the exercise arbiters to rule a defeat of the Blue Force and end that year’s Grand Joint Army-Navy Exercise.[11][10] Fleet Problem III was also the first which USS Langley (CV-1) took part in, replacing some of the simulated aircraft carriers used in Fleet Problem I.[12]
That may be a perfectly reasonable way of identifying potential weaknesses, but the planners may not have been aiming for the overall goal of evaluating whether, in the interwar period, Germany or the US would likely win in an overall war. Saying that the Black Fleet defeated the Blue Fleet in terms of the rules of the exercise doesn’t mean that Germany would necessarily win an overall war; evaluating that isn’t the purpose of the exercise.
NegativeLookBehind@lemmy.world 2 months ago
I think the same will eventually be true for AI, especially when you give it weapons
Kyle_The_G@lemmy.world 2 months ago
I think theres a movie about that
JimboDHimbo@lemmy.ca 2 months ago
It’s name is Stealth, starring Jamie Foxx.
I can’t believe these idiots went ahead and gave skynet a fucking jet.
pezhore@lemmy.ml 2 months ago
Who knew such a bad movie would be such a good cautionary tale?
atocci@kbin.social 2 months ago
intensely_human@lemm.ee 2 months ago
It’s already true for AI. Just observe OpenAI trying to control what their AIs talk about. The mechanisms of control they’re trying to employ are leaky at best.