Could an AI-Powered Drone Defeat a Human Pilot? (thedrive.com) 157
America's Air Force "is hoping to pit an autonomous drone equipped with an artificial intelligence-driven flight control system against a fighter jet with a human pilot in a little over a year," reports The Drive:
[T]he general concept of a fully-autonomous unmanned combat air vehicle (UCAV) capable of air-to-air combat, as well as air-to-ground strikes, hold great potential to fundamentally change the character of aerial warfare... At its most basic, a UCAV would be able to perform many of the same functions as manned aircraft, but would be able to make key decisions faster and more accurately, taking into account much more information in a shorter period of time, without any concern about being distracted or confused by the general chaos of combat. They can also be networked into swarms that work cooperatively to maximize their combat effectiveness at any given time far beyond what a human-piloted formation could.
The head of Air Force's Joint Artificial Intelligence Center "implied that the scheduled faceoff was very much an aspirational goal," according to the article, "and that it wasn't clear if it would happen as planned."
They also remember when Elon Musk responded to a tweet suggesting there should be a competitor to the Air Force's F-35 program. "The competitor should be a drone fighter plane that's remote controlled by a human," Musk wrote, "but with its maneuvers augmented by autonomy. The F-35 would have no chance against it."
The head of Air Force's Joint Artificial Intelligence Center "implied that the scheduled faceoff was very much an aspirational goal," according to the article, "and that it wasn't clear if it would happen as planned."
They also remember when Elon Musk responded to a tweet suggesting there should be a competitor to the Air Force's F-35 program. "The competitor should be a drone fighter plane that's remote controlled by a human," Musk wrote, "but with its maneuvers augmented by autonomy. The F-35 would have no chance against it."
Comment removed (Score:5, Insightful)
Re: (Score:3)
And when you're talking about a machine that's not subjected to G Forces that a person is, the limits are that of the aircraft itself.
This is where the advantage of an AI (or even remote controlled drone) will kick in. Things that make humans black out/grey out/red out, things that provide so much g force that a human can't move against it to control the air craft, etc.
Re: (Score:2)
The question that remains to be solved is how do you define "loyalty" for an "AI". And what you do when that changes.
Re: (Score:2)
Pilots also can make some moral and political decisions that an AI will not. They can refuse an criminal order, such an order to strike Congress while it is in session.
Re: (Score:2)
And pilots could also choose to strike Congress while it is in session without any such order.
Re: (Score:2)
Morality, an AI for a drone is little more than a bug brain. In fact that is the smartest way to train hunter killer AI's teach small drones to hunt bugs, the brains would be in a remote bug hunter server and the drones would be out there hunting specific bugs. Probably fire a pneumatic dart on a retrieval string, go out and harpoon the desired bug, retrieve the dart and drop off the bug at the counting station. Speed counts for training, a lot will be learned about target acquisition, in flight manoeuvrers
Re:Wingman (Score:5, Informative)
Yeah, even the F22 is capable of stunts that would leave the human inside dead. And I'm sure when you get rid of all that meatsack support stuff the plane is gonna be a whole lot lighter.
Re: (Score:3)
Re: (Score:2)
Re:Wingman (Score:5, Funny)
How about we mount a bunch of them on the piloted aircraft, and just deploy them as needed? If we can get them cheap enough, they won't need weapons, they could BE weapons. I'm thinking what we could call those. Brizziles. McDissiles. Hey, how about missiles?
Re: (Score:2)
Re: (Score:2)
Re:Wingman (Score:5, Insightful)
Faster responses and turning don't matter in modern combat; Whoever gets missile lock first wins. The cockpit is not the expensive part of the plane. The F-22 is expensive because it's good at its job. Older planes with less capable electronics are indeed cheaper; they'll also never know the F-22 is there until they're hit by a missile.
Re: Wingman (Score:2)
Re: (Score:2)
Re: (Score:2)
Does missile lock work on something as nimble as the missile though? The plane is limited by what the human can endure. A drone is limited by the G forces the plane can endure before it falls apart.
The only thing as nimble as a missile is another missile. Of course, "missile" is a very broad term, but nothing with wings and jet engines is going to avoid a modern rocket-propelled air-to-air missile. Also, a missile designed to shoot down other missiles is generally expected to succeed these days, Modern missiles are far more accurate than they were 40 years ago.
Re: (Score:2)
Faster responses and turning don't matter in modern combat; Whoever gets missile lock first wins.
You could rewind to the late 1950's and your statement would sound almost identical to the Air Force's thinking at the time. Who needs guns and maneuverability when missiles, range, and speed will rule all? Such thinking led to the F-4, a gunless, enormous, non-agile "fighter" armed only with missiles and a big radar.
The problem with this thinking is not military but political. Politicians in Washington decided you couldn't use BVR weapons because you might shoot down a Russian instead of a Vietnamese an
Re:Wingman (Score:5, Informative)
You could rewind to the late 1950's and your statement would sound almost identical to the Air Force's thinking at the time. Who needs guns and maneuverability when missiles, range, and speed will rule all? Such thinking led to the F-4, a gunless, enormous, non-agile "fighter" armed only with missiles and a big radar.
All of that is an urban legend. The F-4 was not gunless because the AF didn't think it needed them for a dogfight. The F-4 was gunless because it was the next in line of a long series of gunless planes going all the way back to WWII. It was an interceptor designed to take out bombers. Its predecessors used rockets, not guns. It replaced rockets with missiles, because missiles were certainly better at shooting down bomber than unguided rockets!
Later, the F-4 was pressed into an air superiority role it wasn't originally designed for. They added a gun, but missile technology also improved. After the F-4 got a gun, the missle/gun kill ratio was something like 20/1. And missiles have gotten much better since.
Unforeseen consequences (Score:2)
I think a more fun question is what will happen politically. There's a huge voting block of veterans that consistently votes for increased military spending. If automation reduces the number of soldiers needed, that could drastically reduce such votes.
Re: Unforeseen consequences (Score:2)
Re: (Score:3, Insightful)
The whole question seems a waste of time, just fire missiles either from the ground or air to air. Do we actually have dog fights anymore? they have Beyond Visual Range missiles.
Re: (Score:3)
Missiles can be defeated with chaff, flares, stealth, jammers, maneuvers and probably soon, lasers that fry or confuse the sensors of incoming missiles.
Re:Wingman (Score:5, Insightful)
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
...so that it would account for ship or submarine captains doing something you would never expect, That's the beauty of a human mind - it often makes irrational decisions, and that eliminates any predictive or "smart system" from really responding until it's too late.
If you've got two incoming missiles then no amount of irrational decisions are going to save you. You're 100% reliant on the ships AI systems to shoot them down.
Re: (Score:2)
Wait, you think you shoot down air-to-air missiles? Really?
The way to survive lock-on is to fly about 120 km - that will exhaust just about every air-to-air missile out there. Whether that be in a straight line (stupid - you'll lose the race, they are 2+ times faster than you) or in loops and corkscrews (better - use that velocity to force the missile to travel larger curves), the ONLY way to survive is to fly 120 km without getting hit - about 80 seconds or so, at full afterburner kick.
Re: (Score:2)
Missiles can be defeated with chaff, flares, stealth, jammers, maneuvers and probably soon, lasers that fry or confuse the sensors of incoming missiles.
By that logic, enemy fighter could be defeated with chaff, flares, stealth, jammers, maneuvers and probably soon, lasers that fry or confuse the eyes of the pilot too.
What one could do to protect the pilot and to prevent the pilot being fooled could apply just as well, or even better, to an AI controlled vehicle (be it missile or jet). The AI can incorporate multiple sensors, optical, IR, LIDAR, RADAR, etc, while human mainly rely on visual sight.
With missiles, the advantage is even greater as the AI only
Re: (Score:2)
Whatever AI you would put in a drone, put it in the kind of drone called "a missile" for best effect.
Re: (Score:2)
Missiles aren't designed for multi-hour flight, they're designed for short fast flights.
A missile has to be attached to something on longer missions.
Re: (Score:2)
Drones are ideal if you need "loiter time", not combat effectiveness. There are missiles with arbitrarily long range, however.
Re: (Score:2)
Re: (Score:2)
Do we actually have dog fights anymore?
We almost certainly will, for similar reasons that bayonets lasted long after guns were invented. And even in the Iraq war, knives were used during the fighting.
Re: (Score:2)
Since Vietnam, I think 1 military plane has been shot down with guns (in air to air combat), and something like 5 helicopters and such. Hundreds of planes have been shot down with missiles. It's silly to talk in absolutes when it comes to warfare, where everything is a tradeoff. For the past 40+ years, replacing the gun with an equal weight of some other ordnance was always the right choice.
The only good reason now to even put a gun on a plane is insurance against everything we now know about air combat
Re: (Score:2)
Re: (Score:2)
A dogfight with missiles is still a dogfight.
I guess it's technically possible to screw up badly enough for that to happen, but doctrine is around preventing that. Missile range is quite long these days, and if you need to get up close to visually inspect/confirm something, your wingman is way back ready to lock onto anything that proves hostile.
Re: (Score:2)
Maybe that will prove to work in practice (do you think we'll ever have a war with a country powerful enough to contend with the US Air Force in the air? Hopefully not), but in the rest of the world there were dogfights as recently as last year [wikipedia.org].
Re: (Score:2)
Re: (Score:2)
"The only good reason now to even put a gun on a plane is insurance against everything we now know about air combat changing"
Forgetting the A-10 Thunderbolt and the AC-13?. For both, the gun is the primary weapon.
The A-10 has a 30 mm rotary cannon
The AC-130 has a 25 mm rotary cannon, a 40 mm cannon, and a 105 mm howitzer.
Re: (Score:2)
The context was air combat though. Sure, for ground targets.
Re: (Score:2)
Re: (Score:2)
The Browning M-2 is way more than 50 years old. Is still effective.
Not over tens of miles.
AC-130 for potshooting 'peasants'? Are you part of the aristocracy and you think 'Gentlemen warfare is aircraft only?'
No, I'm just pointing out that it's easy to shoot with short range weapons at targets that don't shoot back. That's a very different thing from fighting for air superiority, which is what fighter airplanes are for.
New thing called SAM, the days of meat pilot days are almost over.
SAMs? But why not flak cannons? Suddenly guns are not enough? Interesting turn indeed!
Re: Wingman (Score:2)
The drone is a cheaper way to carry missiles near their target than a fighter.
Re: (Score:2)
"Wars seem to end nothing."
The Ba'ath Party (both Iraq and Syria arms) was directly modeled on the Nazi party. In Syria, it still exists. In Iraq, it has morphed into Daesh.
Pretty much no we don't (Score:2)
Missiles don't fly for hours, cover your 6:00 (Score:2)
For striking a land or sea target some distance away, sure cruise missiles often make more sense than a strike fighter.
Where you do use a fighter, having a missile onboard is a poor substitute for having a buddy covering your six. The missile isn't a wingman and isn't going to keep the enemy from getting behind you.
Air to air combat is still a thing, whether with guns or with short range air-to-air missiles. It hasn't been that MUCH of a thing because the US Air Force has sufficient capability that a) only
Re: (Score:2)
Re: (Score:3)
We already have those, they're called 'guided missiles'.
Re: (Score:2)
I'll have to see how well it will do on backup vocals to my "You've Lost That Loving Feeling".
Off course (Score:2)
Re:Off course (Score:5, Insightful)
if a drone doesn't care if it lives or dies, it will easily defeat a human.
Indeed.
Also, asking "Can a drone beat a human pilot?" is the WRONG QUESTION.
The proper question is "Can a pilot ($5M to train) flying an F35 ($94M each) be defeated by however many drones you can buy for $99M (likely several dozen at least)?"
Re:Off course (Score:5, Insightful)
Re: (Score:2)
Re: (Score:2)
If it's just a matter of winning at a low cost, then why bother making the drone anything more than a glorified missile
First, let's state the obvious: China is the only potential adversary that will ever challenge American air superiority. Russia is sliding backward. All other countries with significant airpower are our allies.
A conflict with China means defending the two Island Chains [wikipedia.org].
We need to be able to engage Chinese aircraft at a range too far for them to launch their own ordinance. A fire-and-forget missile is not going to work. We need a drone that can be sent out on patrol, loiter in the target intercept zone, a
Re: (Score:2)
That's almost beside the point. The F35 is supposed to kill you before you see it, and (it was argued) it didn't *need* to be great at dogfighting.
Now I've always been dubious of that claim; as the old Yiddish saying goes, Mann Tracht, Gott Lacht (man plans, God laughs). But let's take it at face value and consider how the same argument might apply to a drone. Freed of the need to carry a pilot and cater to his needs, could potentially be stealthier, have a larger payload or a greater range.
The drone could
Could an Ai-DRONE beat a human? Doubt it (Score:3)
I seriously doubt that AI by itself is good enough to take on Humans. However, an AI controling multiple drones would destroy manned planes so quickly.
Re: (Score:3)
Every fighter today carries a swarm of drones for just this purpose. They're highly optimized, hard to d better other than the software. We call this special kind of drone "a missile".
But drones with jet engines, wings, and useful amounts of fuel are no cheaper than manned fighters. The cockpit isn't the expensive part of the plane, and you can't make anything useful in air combat much smaller than existing small fighters.
We will eventually use drone fighters to protect the lives of pilots, but that's th
Re: (Score:2)
Re: (Score:3)
If the mission is simple enough and enough training data can be generated you can get pretty good results out of "AI" (really CNN, DNN, and other techniques part of the field of deep structured learning). Complex missions or missions you could not prepare for are going to be a problem. But your drone is still cheaper than the cost of training a fighter pilot. An intercept mission, especially in night conditions, might be a huge advantage for an human-less aircraft. Tighter maneuvers, slimmer airframe, nove
Acceleration (Score:2)
An unmanned plane can out dogfight something that carries a human pilot, especially if its got enough power and thrust vectoring.
Re: (Score:2)
An unmanned plane can out dogfight something that carries a human pilot, especially if its got enough power and thrust vectoring.
Indeed. A human will blackout at 5 g's. With training and a g-suit, 9 g's is possible.
A drone airframe can be built to withstand 100 g's.
So a much, much tighter turning radius.
Re: (Score:2)
I'm curious about the turning radius - without thrust vectoring, isn't there a practical limit as to the ability of control surfaces to be acted on by the air flowing over them?
Re: (Score:2)
This hasn't been relevant for decades now. Air combat is won by whoever gets a good lockon first.
no brainer (Score:2)
Re: (Score:2)
This is a no brainer, absolutely empathically yes they could. The question is really is, are the programmers good enough.
Yeah after a couple dogfights its disk will get full - mid-third-battle the AI will freeze because it'll need swap space but the programmer put everything in one partition.
Re:no brainer (Score:4, Informative)
The question is really is, are the programmers good enough.
Not if they are like the ones I work with, who think that because they can program they KNOW everything.
G forces (Score:2)
No contest. Human can only win if object analysis and classification is involved. Maybe if the task were to select targets amount friend and foe, the human has a chance.
Re: (Score:2)
That's what IFF is for.
No G constraints, should fly like a UFO. (Score:2)
Beyond visual range it should be decent too if it is good at detecting and avoiding incoming missiles (which, again, without G constraints it should do reasonably good at if the AI is trained well enough and has good situational awareness)
All comes down to how good the AI is and it's sensor suite / integration, and if it can be "unpredictable" enough for a huma
Only 1 drone? (Score:2)
When survivability of failure becomes a purely economic question which doesn't involve a pilot everything becomes too much cheaper for manned fighter programs to be anything but a boondoggle.
Can the manned plane win against a dozen drones which can fly twice as fast with half the radar profile?
I don't see how jamming is a problem either, you put a high gain antenna on it which tracks the known path of a command drone or satellite. With double digit ghz signals it won't see the jammer.
If it can beat a chess champ, it can beat a pilot (Score:2)
And computers are only going to get better and smaller. It's inevitable.
Re: (Score:2)
It only beats a chess champ by brute forcing a solution to a game where at any one point in time there are always a predictable and finite number of choices available.
There are not a finite number of choices available in real-life combat, because there are no real rules beyond those that the laws of physics might impose.
Re: (Score:2)
Re: (Score:2)
Perhaps it would be better to say that even if the number of choices is finite, there is no defined upper bound on the number of choices that are actually available. This is, of course, functionally the same as the number of choices being infinite, even if the latter is not literally true.
Re: (Score:2)
Re: (Score:2)
That will work until (Score:2)
Re: (Score:2)
They adapt. Remember, resistance is futile.
Human might win round one. If the drones AI is in communication with others, it can learn the new human strategy and devise a counter.
Re: (Score:2)
Once true AI gains weapons control, the last time that will happen was in the last movie written by a human.
Stealth (Score:2)
Visual aid for thread. [youtube.com]
I particularly enjoyed the otherwise-gratuitous apple biting scenes. Dog whistles are for dogs. I prefer Elect whistles.
tower this is ghost rider requesting a flyby (Score:2)
tower this is ghost rider requesting a flyby
Negative, Ghost Rider, the pattern is full. (Score:2)
Wrong question (Score:2)
.
But more strategically... at what point does one stop sending humans up against drones? Why not just let the drones fight the drones.
So, really, the question becomes, at what point would we want to send up humans to fight against drones, if ever we would want to do that?
Very prescient 1992 film (Score:2)
No (Score:2)
Isn't this great? (Score:2, Insightful)
We can't solve climate change but the killing machines just keep getting better.
Eventually, yes. (Score:2)
Could an AI-Powered Drone Defeat a Human Pilot? Not this year, or next, but eventually, yes. Musk is wrong; an AI piloted drone won't be able beat an F-35. But perhaps a couple of generations along.
People have the wrong attitude toward "artificial intelligence". There isn't any intelligence other than the very HUMAN intelligence of the programmer, and its only advantage will be that it follows its logic tree very quickly. SF author James Hogan wrote a book (Two Faces Of Tomorrow) featuring an AI compu
Re: (Score:2)
Re: (Score:2)
There isn't any intelligence other than the very HUMAN intelligence of the programmer, and its only advantage will be that it follows its logic tree very quickly.
You should look at some of the games played by AlphaZero. Nobody taught it how to play chess. Well, they gave it the rules, but that's it. Not even a "try to defend your pieces" or "try to get open lines" or anything like that. Just "a pawn can do this, a rook can do this, checkmate is defined like this". With only those rules, they let it play against itself millions of times. It came up with winning strategies that even the best human chess players would have discarded immediately. Like giving up an entir
Maybe. Maybe not. (Score:2)
On the other hand is what someone else pointed out: when you remove the G-factor limitations of a human pilot, then suddenly the machine has a certain advantage in that it can push the aircraft into maneuvers that would kill or render unconscious a human pilot.
But when you're talking about the current crop of algorithm machines they keep calling 'AI', the problem is there is no cognitive capability, no intuition, no creativity, and no understanding of how humans
Simple yes (Score:2)
Make a realistic flight simulator as is used for developing automobile AIs. Then pit human players against AIs in competition for some time. As the AI eventually dominates human players, make it real.
The most important aspect of the simulator will be to support adding new realistic drones through code. This would allow the creation of superior drones that would be far more maneuverable drones than could car
The answer is yes without a doubt (Score:2)
We're fucked. (Score:2)
Gamers know (Score:4, Interesting)
Developers are constantly tuning game "AI" such that human players can remain competitive. Thus there are almost always various levels in the game to establish "hardness".
If we ever find ourselves pitted against a weapons platform that is computer controlled and has no artificial limits set, I fear we as combatants would find ourselves outmatched and outclassed, every time.
This is a technological advancement that IS going to happen. So, in that regard, I would much rather see that advancement aimed at our foes, such as they are, then at us or an ally.
Re: Gamers know (Score:2)
So it would _only_ have real world limits? Could you be underestimating those a bit? There is no other_team Boolean in real life.
I think you are also overestimating the benefits of fast reaction time. It would have to be smart too. Will it know it could be flanked? Will it do anything to avoid being ambushed? These are real life concerns. On the other side, can it stalk a target, or set an ambush? Can it use the weather and terrain to its advantage?
That's all silly anyway because we're not even remo
Re: (Score:3, Interesting)
Re: (Score:2)
The article specifically is a report that this about to be tested for the first time.
Re: (Score:2)
The difference is that you aren't specifically going to be facing a set of rules that are controlled by a script. That is not what Artificial Intelligence is, in actual fact.
Yes, you will be facing a computer program but that program will essentially be granted the capability to perform withing the operating envelope of the airframe and weapons loadout. Any innovation you as a human can achieve, can also be achieved by a true AI. This is what true AI is after all. I write programs called scripts for basic n
Re: (Score:2)
Anyone who has played a video game like Elite:Dangerous will know that AI piloted flying enemies are lame...
Perhaps it's because the computer generated opponents are artificially limited in their capabilities. In Elite: Dangerous and all the way back to the first MUDs, computer opponents have been limited in their response times, limited in the type of reaction, limited in the scope of that reaction, limited in the aggressiveness of that reaction and etc.
Imagine how fun any computer game would be if you could never defeat the computer. Difficulty levels exist for this reason. Once you win the game at normal diffi
Re: (Score:2)
Imagine if that plane's "pilot" was a true AI and not limited by the imagination of the writers. Horrifying!