AI Claims 'Flawless Victory' Going Undefeated In Digital Dogfight With Human Fighter Pilot (taskandpurpose.com) 115
"A simulated F-16 Viper fighter jet with an artificial intelligence-driven 'pilot' went undefeated in five rounds of mock air combat against an actual top Air Force fighter jockey today," reports The Drive in an update to a story shared by Slashdot reader schwit1. From the report: The event was the culmination of an effort that the Defense Advanced Research Projects Agency (DARPA) began last year as an adjacent project to the larger Air Combat Evolution (ACE) program, which is focused on exploring how artificial intelligence and machine learning may help automate various aspects of air-to-air combat. Heron Systems, a company with just 30 employees, had beaten out Aurora Flight Sciences, EpiSys Science, Georgia Tech Research Institute, Lockheed Martin, Perspecta Labs, PhysicsAI, and SoarTech to claim the top spot in the last of Defense Advanced Research Projects Agency's (DARPA) AlphaDogfight Trials. This three-day event had started on Aug. 18, 2020.
On the first day, all eight teams had spared against five different types of simulated adversaries that Johns Hopkins University's Applied Physics Laboratory (APL) had developed. This included one dubbed a "zombie," with a flight profile similar to a cruise missile or a large drone, as well as ones that performed like fighter jets, such as the F-16 Viper, or heavy bombers, according to Air Force Magazine. On Aug. 19, the teams 'flew' against each other, whittling down the number of competitors to four finalists -- Aurora Flight Sciences, Heron Systems, Lockheed Martin, and PhysicsAI -- who moved on to the last phase. Those four remaining teams then battled each other in semi-finals earlier today.
Lockheed Martin beat Physics AI, while Heron Systems defeated Aurora Flight Sciences. Heron Systems pulled out a major upset over number two ranked Lockheed Martin before going on to face the actual human F-16 pilot, a Weapons School instructor pilot with the callsign Banger, in simulated combat. This tournament was the third and final trial in a series of events that started in November 2019. That initial trial involved teams flying simulated F-15 Eagle fighter jets, while the second one, which took place in January of this year, shifted to using the F-16 as the representative aircraft. The teams taking part in the competition this week again used digital representations of the Viper. It's not entirely clear how the outcome of this tournament may now impact the larger Air Combat Evolution (ACE) program directly. DARPA has said in the past that it hopes the event will at least "energize and expand a base of AI developers" for ACE.
On the first day, all eight teams had spared against five different types of simulated adversaries that Johns Hopkins University's Applied Physics Laboratory (APL) had developed. This included one dubbed a "zombie," with a flight profile similar to a cruise missile or a large drone, as well as ones that performed like fighter jets, such as the F-16 Viper, or heavy bombers, according to Air Force Magazine. On Aug. 19, the teams 'flew' against each other, whittling down the number of competitors to four finalists -- Aurora Flight Sciences, Heron Systems, Lockheed Martin, and PhysicsAI -- who moved on to the last phase. Those four remaining teams then battled each other in semi-finals earlier today.
Lockheed Martin beat Physics AI, while Heron Systems defeated Aurora Flight Sciences. Heron Systems pulled out a major upset over number two ranked Lockheed Martin before going on to face the actual human F-16 pilot, a Weapons School instructor pilot with the callsign Banger, in simulated combat. This tournament was the third and final trial in a series of events that started in November 2019. That initial trial involved teams flying simulated F-15 Eagle fighter jets, while the second one, which took place in January of this year, shifted to using the F-16 as the representative aircraft. The teams taking part in the competition this week again used digital representations of the Viper. It's not entirely clear how the outcome of this tournament may now impact the larger Air Combat Evolution (ACE) program directly. DARPA has said in the past that it hopes the event will at least "energize and expand a base of AI developers" for ACE.
Can it buzz the tower? (Score:3)
Can it buzz the tower?
Re: Can it buzz the tower? (Score:5, Funny)
Try with an Israeli pilot (Score:2)
They have a reputation as the best.
Re: Try with an Israeli pilot (Score:1)
Re: (Score:2)
Well, on the positive side, had the AI won a real battle in the sky, they would be collecting pieces of the human pilot now, and your taxes would be paying a small part of the compensation bonus to his family. So...
Re: (Score:1, Interesting)
Well base upon reality, I could punch that human pilot in the face hard and the pilot would probably still fly with a broken nose. That AI I just have to break one critical transistor out of the millions to target and that fucker goes down, stone dead, falling out of the sky.
When it comes to reliability I would still trust a human over 1 in 10 million transistors, just one has to fail and the rest are useless, a 10 million to one chance against it. How many human brain cells can die before that human becom
Re: (Score:3)
Re: Try with an Israeli pilot (Score:2)
Re: (Score:2)
"Yet you can't even show us working self driving."
There are no roads, crossings, road-signs, railway-crossings, speed-limits, pedestrians, bicycles and playing children and animals in the sky.
Re: (Score:3)
Avionics is some of the highest-reliability/availability stuff around, if one transistor fails it won't even be noticed. Heck, you can put a bullet through something like a Boeing PFC and the others will keep going without skipping a beat.
However, the main point of the article is that of course a bot is going to beat a human, bots react vastly faster, can monitor all sensors at once, consider all possible outcomes, etc, that's always going to beat wetware-based heuristic reasoning. Read the book "Digital
Re: (Score:3)
Well base upon reality, I could punch that human pilot in the face hard and the pilot would probably still fly with a broken nose. That AI I just have to break one critical transistor out of the millions to target and that fucker goes down, stone dead, falling out of the sky.
Except that the human pilot is also relying on those millions of transistors. You haven't replaced a vulnerability with a lower risk option, you've added an additional risk.
Also, don't punch pilots in the face.
Re: (Score:2)
Well base upon reality, I could punch that human pilot in the face hard and the pilot would probably still fly with a broken nose. That AI I just have to break one critical transistor out of the millions to target and that fucker goes down, stone dead, falling out of the sky.
Except that the human pilot is also relying on those millions of transistors. You haven't replaced a vulnerability with a lower risk option, you've added an additional risk.
Also, don't punch pilots in the face.
This exactly, if only had mod points. Since all modern jets are fly by wire and there are zero control surfaces directly moved via hydraulics etc linked to direct control inputs. All of it filters through the avionics systems. In fact that very system often imposes limits that can be relaxed with only AI to consider as no more human introduced problems like GLOC risk and risk of PIO's with less damped response curves. Some limits are more for stress/metal fatigue reasons and only overriden in emergencies wh
Re: Try with an Israeli pilot (Score:5, Informative)
> how many transistor can die before a CPU becomes dysfunctional 1 just fucking 1 out of millions of transistors
If that's how computers worked, computers wouldn't work.
CPUs "automatically disable" (ELI5 version) bad circuits in them and continue operating at ever-so-slightly lower speeds. The clock runs a bit slower than theoretical max, so a number of failures can occur before the actual computer shows any signs at all.
On top of that, critical systems like fighter jets use multiple entirely separate control systems, isolating any that fail and continuing with the remaining good ones.
No idea about modern jets, but the Space Shuttle had complete 5 redundant flight contol systems for example.
Re: (Score:2)
Uhh.. CPUs don't work like that, in general(*). It would depend on where the fault was (logic vs cache, memory, etc...) as to what issue you'd see and how severe it would be, but there are certainly a large number of transistors in a CPU that would completely bust it if a single one just crapped out. The reasons computers work is because people put a lot of time and effort into testing them for defects and these failures are relatively rare.
* - you could certainly design a CPU like that and maybe there are
Re: (Score:2)
Re: (Score:2)
You do know the human pilot is relying on millions of transistors and if any of them fail the pilot and plane are just as dead as if a transistor inside the AI brain failed, don't you?
Re: (Score:2)
Re: (Score:2)
Re: (Score:3)
They have a reputation as the best.
Well, if the target is a bunch of farmers, yes.
Practice? Nah Code (Score:1)
So now its not how much you practice and train but how well you code.
Re:Practice? Nah Code (Score:5, Insightful)
Except that dogfighting hasn't actually been relevant to air combat for about 40 years. When missiles were somewhat new, they could be avoided or decoyed, but those days are past. These days anti-air missiles are quite reliable (except the man-portable ones). The only way planes are going to get close enough for a dogfight is if, for political reasons, the fight doesn't start until they're already in close. Which, come to think of it, is a great mission for drones: fly close to an unknown but probably enemy aircraft to get visual confirmation, and loiter in harm's way waiting for permission to engage.
Re: (Score:2)
Re: (Score:2)
Beyond Visual Range Shots Not Allowed ... (Score:2)
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
The incident I refer to was back in the 90s from memory and the helicopters will full of peace keepers or weapons inspectors or some such thing over Iraq. I am sure you can look it up. However if we cannot use BVR missiles then what use is AI. A BVR missile is kind of the equivalent thing except it does not need to worry about landing. I am not aware of current policy on missile usage policy.
Oh, I'm sure the referenced incident happened. The reason for avoiding BVR was due to such errors going back to at least the 1960s, basically soon after they started using BVR missiles.
AI pilots, if they are all they are claimed to be, should be useful after visual confirmation as well. ID of aircraft just another task to train them on. Also we have the external verifications like the AWACs. The visual ID may reduce the value of stealth aircraft but not necessarily AI aircraft.
Re: (Score:3)
You are probably referring to this incident:
1994 Black Hawk shootdown incident [wikipedia.org]
Re: (Score:2)
IFF isn't fool proof. The more you rely on sensors the more likely blue on blue is to happen. Because of this its common for RoE to require visual identification. It prevents things like what the Iranians recently did. The real world isn't like video games where you have perfect situational awareness at all times and know exactly who everyone is and what their intentions are.
Add up these realities and it means jets still have to get close to one another from time to time and that makes BFM more likely. You
Re: (Score:3)
Everyone still has guns, but the latest US fighters have like 5 seconds of ammo. It's mostly just tradition at this point, though perhaps is a way to make sure there is a gun designed in just in case doctrine is wrong.
Also unless they have been updated I don't think either the IR or radar self guiding can tell friend from foe
Thus my caveat above. In any serious conflict, anything in the AIR without friendly IFF just dies. It's the only way to use the advantage of modern aircraft - fire at very long range before the enemy gets lockon. It's only when you don't know whether a potential target is hostile that you'
Re:Practice? Nah Code (Score:5, Insightful)
Except this is only true because most wars in the past 40 years have been fought either between a "superpower" and a tribe or two, or between tribes well-armed with weapons you can carry on you or stuff in a Toyota jeep. So, one or both parties did not have an air force at all.
Re: (Score:2)
Re: (Score:2)
But the possibility of fighting a more capable adversary remains. The future is unpredictable.
Re: (Score:2)
But this is totally irrelevant to the point made.
AI and Planes (Score:5, Funny)
Re: (Score:2)
That might be a better point if similar but much better-designed systems not constrained by the 737’s operating certificate didn’t exist for decades.
Re: (Score:2)
Fail to succeed (Score:4, Interesting)
I think of these exercises as ways for the armed services to ask for money. Back in 2005 (I had to look it up) the Indian air force "shot down" a bunch of F-16s in exercises https://www.csmonitor.com/2005... [csmonitor.com] . Either the U.S. was waaaay overpaying for our airplanes and training, or the US Air Force decided that they needed something dramatic to justify continued funding of the F-35. Or both, I'm not saying it couldn't be both.
There may be some pride in the good ole US of A at work in my assessment too ("We can't be beat unless we want to"). But really no matter what, we shouldn't be spending as much as we do and lose anything to anyone ever.
So I always hear the same message with these things: Our military leadership has taken a shit-ton of money and produced inferior results, Give them all the money they ask for, or the results will continue to be inferior.
Not really surprising to me (Score:5, Insightful)
I'm not at all surprised by this result. Two problems: information latency, and lousy visual resolution.
The human pilot in a simulator has no seat-of-the-pants sense of airspeed or g level, both of which are absolutely crucial for doing well in a dogfight. He could only see those in his HUD in a VR system, which required him to add, if you will, an extra couple of channels of attention and compensation compared to his usual scan and control inputs. And the update rate in a HUD is limited. Everything he was doing had to be crosschecked against the HUD, while in a real airplane, he just feels the jet respond and knows if he can afford to add more g or roll harder.
Also, the pilot is limited by a low-resolution visual system. An Oculus or other typical VR headset has relatively poor acuity, and knowing exactly which way the other airplane is turning - and even beginning to turn - is critical to any air combat situation. That's why pilots must boast perfect vision to become fighter jocks.
The AI, however, gets all this information instantly, and can make minuscule adjustments to its control inputs to wrest the maximum performance out of the airplane. Unless they crippled the AI with poor or "fuzzy" information on opponent attitude and direction, and some lags on its own g and airspeed info, there's no way the human pilot has any chance of beating an even slightly competent AI.
This trial is really not particularly meaningful, until you put the AI in control of a real airplane using actual optical sensors for opponent position information, and rerun the test that way.
Re: (Score:3)
> The human pilot in a simulator has no seat-of-the-pants sense of airspeed or g level
I don't suppose you are a fighter pilot or do aerobatics?
In the planes I flew, which were just about the opposite of fighters, there was zero "seat-of-the-pants sense of airspeed". None at all.
Fighters may be different. I wonder if there is any "seat-of-the-pants sense of airspeed", because there is none whatsoever in slower planes.
Re: Not really surprising to me (Score:1)
Re: (Score:2)
You might find these informative.
https://www.dictionary.com/bro... [dictionary.com]
https://www.dictionary.com/bro... [dictionary.com]
Re: (Score:1)
Their "argument" is entirely a result of a thought experiment, that is, zie's making shit up.
If you've read some of hir posts, which is easy because zie actually spends most of hirs time here, you'll see zie's allegedly done things in hir life that an ordinary person IRL will need roughly seven lifetimes and the fortunes of several Fred Trumps to do.
So it is quite safe to assume zie's just dreaming the life.
Re: Not really surprising to me (Score:2)
They/their.
The words you're looking for are "they" and "their."
Re: (Score:3)
> The human pilot in a simulator has no seat-of-the-pants sense of airspeed or g level
I don't suppose you are a fighter pilot or do aerobatics?
In the planes I flew, which were just about the opposite of fighters, there was zero "seat-of-the-pants sense of airspeed". None at all.
Fighters may be different. I wonder if there is any "seat-of-the-pants sense of airspeed", because there is none whatsoever in slower planes.
Humans have no sense of velocity, beyond watching the world move past us. We feel acceleration. Not a pilot myself but I would suspect it is more about the feel of how close to the edge the airplane is, rather than velocity. Can the airplane turn a little harder, or is it about to lose lift and fall out of the sky? At least that is what comes to mind for me.
Velocity (Score:2)
It is quite easy to fly an aeroplane without an air speed indicator. The angle of attack varies, the feel of the controls varies, the sound varies.
If you have ever flown in an airliner you will know the back of the seat feeling of the acceleration. Integrate that and, with experience, you get an idea of airspeed.
So no, there is no absolute sense of airspeed. But there are plenty of indicators.
Re: (Score:2)
It is quite easy to fly an aeroplane without an air speed indicator. The angle of attack varies, the feel of the controls varies, the sound varies.
If you have ever flown in an airliner you will know the back of the seat feeling of the acceleration. Integrate that and, with experience, you get an idea of airspeed.
So no, there is no absolute sense of airspeed. But there are plenty of indicators.
I had not considered sound, fairly obvious since it works with many cars as well. Thanks for the correction.
Re: (Score:3)
The sound simply indicates the engine power, but tells you NOTHING about the airplane attitude. It's very much possible to fly into the ground at full engine power.
You NEED to have reliable AoA and airspeed indicators to fly if there are no external references (Sun or ground).
Re: (Score:2)
It is quite easy to fly an aeroplane without an air speed indicator. The angle of attack varies, the feel of the controls varies, the sound varies.
Indeed. It's very easy to fly an airplane into the ground this way.
Quite a few actual air crashes happened exactly because pilots couldn't tell that they are descending. The most recent example was Kobe Bryant crash. I experienced this myself while doing flights with an instructor.
Re: (Score:2)
Nope. It is impossible to have a sense of the airspeed unless you somehow evolve a pitot static system. If you have ever flown in an airliner you would know that the back of the seat feeling disappears during the cruise flight and can even be misleading because of the quirks of the human vestibular system. Besides, the airspeed is relative to the air, hence it depends on the wind speed and direction, hence only indicator that can give you an idea of the airspeed is the actual airspeed indicator.
Re: (Score:2)
It is a standard part of learning to fly an aeroplane to be able to fly and land it without instruments. Usually just past first solo.
Re: (Score:2)
That might work on a cessna 152 that essentially can't fly slow enough to stall with full flaps but not in an airliner. But that is not feeling the airspeed, that's not giving a shit about it.
Re: (Score:2)
Fighters may be different. I wonder if there is any "seat-of-the-pants sense of airspeed", because there is none whatsoever in slower planes.
I'll also add that even the sense of _direction_ is not reliable. It's not at all uncommon for pilots to think that they are climbing while in reality they are descending when flying at night without ground visibility.
Re: (Score:3)
That's also true with perfect ground visibility. As long as you're not in a near vertical nose-dive, the change in apparent size is very gradual and can go unnoticed. Note here that being pointed nose up does NOT mean you're climbing. In fact, ground your nose too high is a very common way to lose altitude. (Altitude change is controlled by pressure differences on the wings, it's a *plane*, not a helicopter).
In one of my training flights my instructor, an experienced pilot who designs small planes, was s
Re: (Score:2)
No, I'm not a pilot. But I am a 30-year flight test engineer with a specialty in testing military flight simulators. So I'm extremely well-acquainted with techniques, both good and bad, of artificially cueing airspeed for a pilot.
As others have already replied, it's not so much absolute airspeed as the CHANGE in airspeed. But one does get real cues from wind rush, turbulence and buffet levels, and how the airplane responds to control inputs. Those are all native to a well-experienced pilot, and in a simulat
One aspect may be inescapably better for AI (Score:2)
I agree about a real pilot probably fairing better in the outside world...
There is one aspect though where I wonder if it would still lend itself to an AI victory - G-Force limitations. With a human pilot you are constrained to the G-forces a human body can take.
I would imagine the plane is related a lot higher though, so an AI pilot could simply introduce far more extreme maneuvers a human pilot could not match or evade... even keeping a pretty large margin of safety for the plane itself.
Re: (Score:2)
Re: (Score:1)
Great point just the weight saved on the ejection seat alone, and the bracing around it to support the weight of a human being tossed around at 5Gs...
It does bring to mind a question of how much power and compute infrastructure that AI in the contest required, could it currently be feasibly mounted in a jet? Or even with latency would it be able to operate remotely?
Re: (Score:3)
It reminds me of the old mecha combat game Armored Core, where the human player ran missions or duels against the CPU player. Both machines had radar but the human player radar was limited by range. The CPU player radar was limited by range as well, but it ALWAYS knew exactly where the human player was, and aimed at it, 100% of the time, even without a weapons target lock. There was no way to evade this. It simply knew because of course, it had to know.
The human player had no such advantage. The C
Re: (Score:2)
It's kind of the opposite in Starcraft. If you watch professional players they are constantly micro-managing every single unit, giving it specific targets that change second to second rather than relying on the AI to select them.
Even when the CPU cheats it is limited by the rules programmed in to it.
Re: (Score:2)
The CPU player radar [...] ALWAYS knew exactly where the human player was, and aimed at it, 100% of the time, even without a weapons target lock. There was no way to evade this.
What? Of course there was. Just don't "tell" it.
It simply knew because of course, it had to know.
No, it didn't. The game engine has to know. The "AI" for the enemy 'mech doesn't have to know. You simply don't give it that information, and program it as for how to behave if you don't. But that would require making AI code that can make useful decisions, like teaching it to figure out where the human player is headed based on the terrain and their last known location and heading. It's much easier (and thus cheaper, both in the economic sense, and in the CPU
AI not limited by G Forces (Score:2)
Or if G forces are limited in the AI aircraft we still have the situation where the AI can endure a high G load indefinitely. Unlike a human pilot that can only endure for a limited timeframe.
Re: (Score:2)
Actually the fly by wire system limits the G forces because the pilot is able to bend the airframe otherwise, especially if the aircraft is loaded out. It happened to a German MiG-29 - these have no fly by wire.
The AI can endure a high G load, though, yes.
Re: (Score:2)
Does any of that stuff really help against long range air-to-air missiles? They are launched from beyond visual range and while modern fighters do still have canons they are a last resort weapon and missiles are preferred.
Re: (Score:2)
I actually got to sit in the USAF's F-16 trainer/simulator back in the 1990s. (I was working on a project to see if we could replace their $6 million SGI computers running it with a b
Re: (Score:2)
I actually got to sit in the USAF's F-16 trainer/simulator back in the 1990s. (I was working on a project to see if we could replace their $6 million SGI computers running it with a bunch of these newfangled 3D graphics cards coming out for PCs. Interesting side note - they're housed in a building whose walls are entirely lined with lead, to create a Faraday cage and prevent any sigint that might give away true aircraft performance from leaking out.)
That was about the same time I saw the Marines' part [archive.org] of that integrated wargaming system: Synthetic Theater of War [STOW] [archive.org]. At the time, the Air Force was working on modelling all the aircraft, the Navy was modelling the ships, Army did the vehicles, and Marines were doing people.
The display our version used was a rear-projection screen on a wraparound three-panel frame. Each of the three projectors was a quarter-million dollar monster driven by a rack-mounted video system that had separate video cards for e
Re: (Score:2)
There are simulators that simulate the motion.
Re:Not really surprising to me (Score:4, Insightful)
We're seeing that across all automated systems. It's part of why Tesla has been able to iterate their way towards FSD the way they have been.
They make an improvement, and as long as the hardware doesn't need to change, they can push it to every system out there overnight.
Every time there's a failure in the AI, we can retrain it, fix the problem, and push it to all the systems. That's an incredible advantage over human pilots or drivers. And the AI won't forget it, and won't revert to muscle memory and deep-seated habits during stressful situations.
Sure, AI has disadvantages, but consistency and the ability to update it and make those updates "stick" are two major advantages.
SO programmers are GOD (Score:2)
Bye bye... Maverick
Next, AI vs something from this century (Score:3)
While one of the most numerous (everyone flies them) fixed wing aircraft, the F-16 had it's first flight in 1974. yeah, that's 46 years ago.
I'd prefer to see something like AI vs F-22 (yeah I know, still last century) or AI vs F-35...
REal world (Score:1)
Re: REal world (Score:4, Interesting)
There's one problem with a pilotless aircraft: there's no one to die. The defense budget is already so massive that losing a bunch of unmanned aircraft, even at, say $40 mil each, isn't much for anyone to notice. But if planes with people in them start falling out of the sky, people will notice that. The political cost of using (and losing) unmanned aircraft is much smaller than that of manned aircraft. That means the decision to use them is easier. War needs to be a hard choice, not an easy one.
Re: (Score:3)
Those who refuse to develop this technology because 'war should come at human cost' will simply find themselves faced with a superior force by challengers who would not put such artificial limits on themselves. Such as the kind of nation that subjects its population to 'social credit scores' and embraces absolute tyranny.
The time for meat in the can is quickly coming to an end.
Re: (Score:2)
Re: (Score:2)
There's technically no need whatsoever for anything more advanced than a 1945 bomber for this.
Your 1945 bomber would be wiped out by a modern SAM before you can say "bombs away!".
Literally.
Re: (Score:2)
The closest the US has come to a real anti-air defense was during the first Iraq war, back then Saddam had some obsolete (even at that time) Soviet anti-aircraft rockets and during Serbia bombing raids (which also had obsolete Soviet tech).
Re: (Score:2)
A WWII bomber will still fly above the range of man-portable SAMs. And rebels/terrorists/freedom_fighters don't have access to anything more serious.
Sometimes they do. If you can come up with money you can always find someone to take it.
Re: (Score:2)
Why not have an international agreement on this, like the ones for nuclear weapons. Limit development, limit the number of weapons.
Re: (Score:2)
Would you actually trust any such agreement, when it is a matter of air superiority at stake?
Sure, I'd love it if we could skip automating death and destruction. Given the current state of humanity, where over a billion people live under tyranny and that number is only set to grow, I'm just not optimistic that arms race will be avoidable.
Re: (Score:2)
Well we trusted it for nuclear superiority so... Yeah.
Re: (Score:2)
Not a simple question. On one side you are correct - with no risk, war is too easy. On the other, intentionally increasing the risks your own people take is war is absurd.
That said, air power has always involved killing a a distance, so this isn't a huge change. I think its robotic ground troops that would change behavior a lot.
Re: (Score:2)
I think a non-human machine killing people (even if they are enemies) looks bad and would have approximately the same amount of deterrence as killing pilots.
Re: (Score:2)
The Air Force has been using remotely piloted aircraft for at least the past 40 years, maybe longer (I don't get out much). The public may not know why the military thinks they still need humans at the controls in the air, but apparently the military still believes there is some need for it.
I'll be impressed... (Score:2)
...when it's a real aircraft. An AI "piloting" a simulated jet in a simulated environment doesn't impress me much.
Re: (Score:2)
Re: (Score:2)
I hope and assume the simulation was sophisticated enough to account for that, but I don't know. I suspect an AI pilot is less likely to make that sort of a mistake than a human pilot because it can have hard coded limits set.
Where the AI pilot is likely to do poorly is in some unexpected situation, but I don't know if there are any situations that are relevant to air combat.
Re: (Score:2)
how many maneuvers by the ai would have resulted in an unrecoverable stall....how many maneuvers would result in exceeding the aircrafts structural limitations and tearing the wings off
Almost certainly none, these sorts of limitations have been programmed into every flight simulator I've ever used.
On other news... (Score:5, Funny)
AI learns to be full of itself and gaslight human pilots. It also turns out that it watches Star Trek and pulled a James Kirk Kobayashi Maru in reprogramming the simulation so that it always won.
I can see the future. (Score:1)
My robot army is better than your robot army!
At least, hopefully, innocents wont get killed by this.
Teach this AI to play tic-tac-toe STAT (Score:2)
War Games style
Re: (Score:1)
Yeah before Crystal Palace declares DEFCON 1.
Words of wisdom, Lloyd. Words of wisdom.
1st Law of Robotics (Score:2)
How was the AI able to, even in simulation, kill the human fighter pilot without violating the 1st Law of Robotics? What, they have no limitation to kill any and all humans? How do they distinguish between enemy plans that are hostile vs enemy red cross flights or civilian aircraft?
The future is here (Score:2)
"The wars of the future will not be fought on the battlefield or at sea. They will be fought in space, or possibly on top of a very tall mountain. In either case, most of the actual fighting will be done by small robots. And as you go forth today remember always your duty is clear: To build and maintain those robots."
Mr. McKittrick, after very careful consideration.. (Score:2)
How Big An AI? (Score:2)
Is this something that can actually fit in an aircraft? If not, then what happens when you introduce telemetry lag and jamming?
X9 Ghost (Score:2)
Obligatory Macross Plus reference: https://youtu.be/KurUNRuCGkY [youtu.be]
AI Homework Help (Score:1)
I swear I have seen this movie before (Score:1)
Finish Him (Score:1)
Flawless Victory.
Wrong pilot (Score:1)
Why are they using an air force pilot? The best pilots in the world are Navy pilots. They have actual experience. Air Force pilots are like unionized pilots. Treated with kid gloves. Navy pilot will gut you before you even know anything happened.
Not sure why we even have an air force. Should be part of the Army. Space Force should be part of the Navy.