Hacker News Reader

Comment by grandinj | Hacker News Reader

The more interesting result comes when you combine AI in a jet with :

() Not needing 2-5 years of expensive training. () not needing space for the human and it's associated safety systems (*) being completely willing to sacrifice the aircraft to "win" an encounter or otherwise achieve it's goals.

In other words, an air force does not need AI that can completely dominate human pilots. It simply needs AI that is "mostly similar in performance", and when combined the advantages mentioned above, means that such aircraft will be significantly better in terms of fighting capability.

Of course, that is just raw combat, not other areas of flight where greater human intelligence is of more value.

View on HN

I think that days of big expensive aircraft like modern fighters / bombers are numbered. They will likely be replaced with a swarm of drones, which are smaller, cheaper and harder to detect. For $100M (the price of just one fighter) you can field 10000 cheap drones and literally blot out the sun over the field of battle.

Complemented, and AI-piloted, but not replaced. There are significant positive scale factors in aerodynamics. Larger jets have much better thrust to weight ratios, allowing them to move faster, travel further, maneuver better, and carry heavier and longer ranged weapons and sensor platforms.

Light drones and big expensive aircraft will both exist in the future and play separate roles. They also won't interact much, aside from a fighter group spearheading a drone group. A drone swarm is too diffuse to be a worthwhile target for a fighter's weapons, but a fighter is too fast to be engaged by a drone swarm.

Air battles of the future will likely be conceptually similar to air battles today, but with light drones replacing bombers with guided bombs. An initial wave of fighters will contest air superiority. If they achieve it they will use it to launch long-range attacks to disable a small number of anti-drone hard points. Once the way is clear a wave of smaller, slower drones can swarm in and bomb a large number of targets with precision.

Bigger fighters definitely still have a place, but I'm not sure manned ones do. Imagine a dogfight between a manned fighter jet with a maximum G limit of 15G and an AI fighter jet with a limit of 30G - it won't even be close.

I said AI-piloted in the first sentence.

To add another response, (a) dogfighting is not something 5th+ gen fighters need to be good at and (b) no fighter is ever going to tolerate 30G in anything except maybe a straight line. Fighters are big and have big shear stresses and would tear apart long before then.

It isn't feasible to design an AI fighter jet with a limit of 30G. The airframe necessary to handle that strain would be way too heavy, seriously limiting the fuel load. And how would you avoid compressor stall in a 30G turn?

Dogfights no longer occur anyway. What matters is early detection.

> I think that days of big expensive aircraft like modern fighters / bombers are numbered. They will likely be replaced with a swarm of drones, which are smaller, cheaper and harder to detect. For $100M (the price of just one fighter) you can field 10000 cheap drones and literally blot out the sun over the field of battle.

Maybe that's true of regional militaries.

The US wants to be able to project power - often quite far from bases. The drones you're talking about just don't have much of a range.

I'm not saying they won't be used - I'm sure they will be. But there will also be much larger and more expensive drones as well.

I could see a hybrid approach, where a bunch of drones are loaded into a mothership, possibly a repurposed bomber with a bomb bay full of drones. That gives you local control of the swarm, and a way to deliver them over much longer range.

CARRIER HAS ARRIVED

Headline: America switches to Protoss, rushes Fleet Beacon.

You've pretty much described the F-35 doctrine that the US is developing. The only major difference is that an aircraft carrier or some other ship holds the drones before launch, which makes sense as the you already need the carrier to launch the flight controller craft (the F-35)

Come on, let's be serious. What makes your idea distinct from existing weapon systems? Instead of making up new names for old things just look up what is serving the same purpose as your idea. "Drones" are just guided bombs or missiles. "Motherships" are just fighter jets or bombers.

This may result in a situation that strongly favors the defender. If it takes ten drones to shoot down a fighter but they're only 1% the cost, countries that otherwise can't project air power internationally may become capable of defending themselves at home.

> This may result in a situation that strongly favors the defender. If it takes ten drones to shoot down a fighter but they're only 1% the cost, countries that otherwise can't project air power internationally may become capable of defending themselves at home.

This exists already. Those "drones" are more commonly called surface-to-air missiles.

The limited factor is targeting: as has been well publicized, modern aircraft have been engineered with stealth in mind, which means they need to be quite close to radars before they can be detected.

The solution, which was discovered in Yugoslavia, is to use low frequency radar to detect the stealth aircraft, and wait until they get close enough (defense in depth is necessary here!) before turning on your target grade radar at the last moment.

As you might imagine, this is a difficult and dangerous process, and so has only been successfully accomplished once.

So you're saying ground based defense systems should engage in stealth as well? Attackers use stealth but can also easily spot and fly around known defense systems.

This reminds me of The Diamond Age by Neal Stephenson...maybe not nanobots yet...but still..

I suspect there will be a fairly long lasting middle step: an AI piloted "mini jet fighter". Say 4x smaller, but faster and more maneuverable by virtue of not having to protect a human pilot, and still capable of carrying existing air to air and surface to air missiles. It will sort of be an auto-piloted aircraft carrier for the missiles, which at some level are the "real aircraft" these days already. The missiles need a larger, fast and hard to shoot down vehicle to carry them into the region of combat.

Anyone who plays RTS games knows this is about resources, speed of construction, how much damage the unit can project, and how much damage the opponent can absorb.

No it’s about the military industrial complex milking the US defense budget for all its worth.

I think that ignores that these drones would be much cheaper, no? Certainly in bang:buck.

I'm a bit skeptical. For one, how much of a modern aircraft is actually really dedicated to cockpit space, maybe 5% on a small fighter? How much extra fuel and thrust is really required to lug along a ~80 kg human (okay maybe double to account for supporting systems)? Keeping in mind a single Sidewinder missile weighs 85 kg.

For a drone (even a swarm of them) to compete with a modern jet fighter, it'll need a reasonably powerful engine and sufficient fuel to keep it up in the air long enough to: seek out and engage the enemy, carry a decent payload, and reach a suitable altitude. That all sounds like a lot of fuel and engine power to me, especially if you're expecting this drone (swarm) to operate outside of the immediate vicinity of the launching area.

I see more potential for drones to act as 'screen' for jet fighters once stealth technology is made irrelevant by advances in radar. I foresee the use of, low cost vehicles that resemble fighters (in the ways that count), but carry minimal ordinance (if any) to keep the costs down. They could also provide auxiliary functionality like electronic warfare and scouting.

I mean you could in theory develop a drone that exploded on impact and operated within a limited area, but you've just re-invented the guided missile, with a little extra smarts.

You're asking the wrong question. Not how much space the human takes, but how much performance the plane loses because it contains meat jelly.

I'm not disputing the notion that primary fighters might one day be completely autonomous. My point is that autonomous fighters won't lead to the end of the modern jet fighters as we know them. They'll still be the big expensive machines we know and love (fear?) because most of the factors driving that trend are unrelated to whether the pilot is a human or an algorithm.

It seems you misunderstood GP. Their point is that machines have higher tolerances and won't just die of too many g. Unlike weight, this is not obviously irrelevant.

I understood their point perfectly well and agreed that there may be a point in time where autonomous aircraft replace manned combat aircraft. My point was that such an occurrence wouldn't lead to a significant reduction in aircraft size or cost.

They absolutely will just die from too many G. You're just replacing a family at home mourning the loss of a pilot to the accountants throwing themselves out of windows because these damned airframes keep ripping themselves apart or otherwise incurring excessive maintenance costs because the maneuvers they engage in are pushing the airframe itself to it's material limits. Believe it or not, the art of engineering is striking the right balance to minimize the overall cost of the process, and trying to keep from just stunting the cost to a different coffer.

That's a good point. I always considered this in the opposite direction though. Once you have a sufficient number of drones I don't think any conventional aircraft will have sufficient weaponry to deal with them effectively.

Use the Sidewinder missiles as an example. You're going to run out of munitions very quickly and be left with far poorer capabilities.

I just don't see a drone swarm sufficient to the task of taking down a single jet fighter in disputed territory as being economically feasible (for the reasons I outlined previously).

Maybe it could be done defensively over territory you controlled, but the 'drone swarm' in this instance really does just sound like a ground-to-air missile.

I'm always baffled when I see people talk about "drone swarms" as if they have never used a drone or RC aircraft. Most drones can't fly for more than 30 minutes and then they have to charge for significantly more than 30 minutes.

The smaller your drone the lower its effective range or loitering time. Therefore a transport vehicle such as a jet has to bring the smaller weapon into effective range. The transport vehicle usually also has much better sensors so that it can effectively search for targets. Guided weapons receive additional targeting data from the transport vehicle even after they have been deployed and are heading toward the target.

When people talk about medium sized "kamikaze drones" that are carrying enough fuel to reach a given far off destination they don't realize that these already exist and are called cruise missiles and tomahawk missiles in particular are quite cheap. For the cost of a F-35 the US can buy 100 tomahawk missiles. I still remember that attack where 60 cruise missiles were launched from a single navy vessel at a military airport in Syria with essentially no risk of counterattack.

> I'm always baffled when I see people talk about "drone swarms" as if they have never used a drone or RC aircraft. Most drones can't fly for more than 30 minutes and then they have to charge for significantly more than 30 minutes.

Most drones are cheap civilian toys. Military drones have had much greater endurance for decades.

> Therefore a transport vehicle such as a jet has to bring the smaller weapon into effective range.

If you are using short-endurancd drones for some reason (e.g., missiles), the carrier can just be a long-endurance drone.

> When people talk about medium sized "kamikaze drones" that are carrying enough fuel to reach a given far off destination they don't realize that these already exist and are called cruise missiles and tomahawk missiles in particular are quite cheap.

That they already exist and are cheap and effective is proof of the point, not an argument against it.

> Not needing 2-5 years of expensive training.

It doesn't stop there. You train one AI, you've trained them all. A minor performance increase instantly upgrades the capability of your entire fleet. An AI that can learn from real world situations will likewise convey that learning to your entire fleet instantly.

They say smart people learn from the mistakes of others, but we often have some trouble applying this. An AI should do better.

Of course, the corollary is that bugs will affect your entire fleet. I wonder if a future cyber-security role will be to look for AI blindspots to exploit en masse without prior warning (like flying with the sun behind you, but more arcane).

The absolute worst case scenario would be somehow tricking the AI or taking control from it so jets start attacking/kamikaze-ing their own side en masse. I suppose this is theoretically already a risk for some UAVs and perhaps even some advanced human-piloted jets, but a hijacked autonomous fleet would be a wild sight to see.

No doubt some organizations are researching how to perform and defend against things like that. The future of asymmetric warfare in particular could get really crazy.

If I’m approaching an AI equipped force with my own AI force I’m driven to think about the ‘meta game’.

In this context I would sacrifice multiple aircraft to ‘taint my the enemies training data with solutions which are a disadvantage in a future fight scenario.

Good point, but the meta will surely be played by the other side, too. They'll expect that their enemies will want to try to "obfuscate" their true operations to confuse things that have been trained to recognize and predict certain actions.

That's why this probably won't actually be in use for many decades. You need something a bit closer to AGI than what exists now. Until then, you'll probably have limited use cases, maybe where movement and obstacle avoidance are autonomous but everything else is programmed or remotely controlled.

You could imagine small, agile flying assassination drones which can navigate cities and houses to kill targets with short-range projectiles. Very Black Mirror-esque, but far more ethical than the current approach of Call of Duty 4-style blowing up a big patch of land from the clouds with a Hellfire missile, including all the civilians in proximity.

AI doesn't learn in it's production form. That's training. You'll get the same deterministic response pout of every fighter, even if there is So.e level of sensitive dependence on initial conditions.

Human pilots do mot suffer this problem. A human can arbitrarily change things up for better or worse. Going the AI route is basically going the T-34 route. Suddenly quantity (via reproducibility) takes on a quality all it's own. My issue with that philosophically is does going down that road actually solve any pressing problem?

Which to me it rings a bit hollow and empty unless there's so.e other task we'd prefer people that would be pilots be doing.

General AI probably would learn in real time by most definitions, but setting that aside for a moment I was more referring to gathering production data then using it later. You can gather then re-train, isn't that what Google is doing with our voice data for example?

AI can also fly past black/red out thresholds. Even in current aircraft that would be add capability. But having that in the design phase may open up all sorts of possibilities.

I'm sure someone in these programs has speculated what air combat would look like between AIs where body-based thresholds are removed and machine learning has developed new tactics that are only applicable in a pure AI arena. I'd be really interested in seeing those speculations.

I can't help but think a drone aircraft is essentially the same as a guided missile except the exploding pieces are modular rather than exploding itself and thus the majority of it is re-usable.

We already shoot missiles at other missiles. Drones will make that cheaper. In other words i think the biggest change will be in the economics of war. War will become much cheaper in the same way SpaceX's reusable rockets made space travel much cheaper.

I agree war will become a numbers game even moreso than it already is, by taking out the 'incalculable' losses of human life on both sides of an engagement. But modern combat systems are definitely not cheap!

So the enemy's goal is to convince people on our side that using drones in combat is utterly beyond the pale, something only a monster would do. This prevents them from being developed, lest the military feel severe backlash even from within its own ranks, and ensures we're committed to human-flown craft while the enemy gets to field drones.

Just because drones make sense in a war between nations doesn't mean they can't be abused by governments against civilians. There is no shortage of governments that attacked their own citizens and with drones it becomes even easier.

...it's not the enemy, it's anyone with any appreciation for the physical forces to which a warfighting machine is subject, and the economic factors at play that make production of those systems as disastrous to the nation employing them as not doing anything at all. Keep in mind, a large contribution to the collapse of the USSR was the incredible financial strain created by defense spending in response to the Star Wars defense system initiative.

It was an economic victory for the U.S. more than anything else. There is always a balance. It's just like the hoopla over hypersonic cruise weapons. Sounds horrifying. Until you realize the trumpet bell makes countering them easier rather than harder as long as you detect them early, because there is simply no way to reorient quickly at those speeds without being torn apart.

Further, no one is saying we should stick with manned fighter craft, only that they won't likely be the silver bullet everyone seems to think they will be, will be fraught with their own weaknesses, may not in the long run be cheaper than man operated, and may Coe with unforeseen doctrinal changes as the enemy adapts.

War is complicated, and rarely does who has the cooler gun perfectly translate to who wins or loses.

Gee, I can't wait to see which poor country one of the big players will attack next /s

Modern warfare is eventually just an economic game. How many missiles can you afford to launch, and how many missile defences can you afford to build? (/ how much can you afford to spend on military R&D?).

I'd honestly be quite surprised if AI research discovers novel and useful dogfight tactics. The interplay between speed, height, position, and direction has been pretty thoroughly mined for ideas, there's been little in terms of updates since Boyd's 1964 Aerial Attack Study (I found a copy at http://oplaunch.com/resources/aerial-attack-study-1964.pdf if you're curious)

Then again, the AI system Eurisko won the Trillon Credit Squadron game twice in a row, each time with novel strategies. You might say that dogfighting is more constrained by laws of physics we already know, and so this won't happen. But wouldn't you hate to be proven wrong?

I would bet the inverse. If the majority of tactics is spearheaded by one person from 1964 - that probably means there's a lot of innovation.

I found reading Boyd's biography, so much effort was made before decision makers would use his tactics, even when they were clearly superior.

Even board game like Go and Chess have experienced new and novel tactics from AI despite being much older, with lower barriers to entry. Inversely, dogfighting is new and the barriers to entry are so high its unlikely its close to optimal.

If that's true then why has nobody invented a missile that is equipped with a metal storm [0] as its warhead?

[0] https://youtu.be/wKlnMwuCZso

Probably not as much as you would think. Modern airframes are still not capable of sustaining as much acceleration as missiles. You just can't outmaneuver today's missiles.

maybe this would lead to smart missiles or a platform to launch them that can loiter near the battlefield, directed by a pilot or controller nearby

Smart loitering munitions already exist.

Loitering in the sense of expending fuel load until a hypothetical target arrives? Or loitering in the sense it's there for reconissance and isn't intended to directly destructively interact with the target?

There's a big difference between a Predator with Hellfires, a tomahawk, A Global Hawk, or a man portablerecon/kamikaze drone, and finally Project Pluto, the ultimate "F you" device likely ever contemplated.

> Modern airframes are still not capable of sustaining as much acceleration as missiles.

Are those limits what they are because there wasn't any point in designing stronger airframes if the pilot was just going to black out under higher G forces?

If you take the vulnerable human body out of the equation I wonder if airframes strong enough to withstand that acceleration would actually be feasible.

It's sometimes possible to overstress airframes (even ones that use FBW, and should in theory always operate within limits) in high-G maneuvers, as they have stress limits (g limits), which depend on the airframe as well as load-out for fighters.

It's mostly a matter of mass. You would need crazy amounts of power to make a large aircraft make the same kind of turns that a small missile can do effortlessly, besides that it is much easier to make a small structure very strong. Try killing a flea if you want more proof of that.

A plane is, conceptually, just a platform from which to launch ordnance closer and at a higher starting potential energy than you would be able to from the ground.

Given that a plane, then, needs to be capable of carrying multiple pieces of ordnance (otherwise, it's pretty much just a guided missile), it's pretty much fundamentally going to have a higher mass than whatever missile you're launching at it.

That sounds like it means dogfighting doesn't matter at all. You just need an autonomous missile platform with good enough ECM to avoid getting splattered by the enemy missile platform.

That's the basis behind F35's design and why it can't outmaneuver a F16 in a visual range dogfight but is never expected to get into one.

>That sounds like it means dogfighting doesn't matter at all

There has not been an aerial dogfight in this millennia. The last ones that would really count happened over Bosnia in the 90s. I think a Rafale took down a 1960s era MIG-23 in the Libyan Civil war, but modern fighters are either too hard to hit or too expensive to risk.

> If you take the vulnerable human body out of the equation I wonder if airframes strong enough to withstand that acceleration would actually be feasible.

Take the human body out, and the airframe is a (possibly, bus) missile.

I don't know why this keeps getting brought up. Humans may be a factor in the maneuverability of an aircraft but the days of dog fighting are over. There are effective countermeasures against guided missiles that don't involve high G forces.

Today it's all about spotting the enemy first and then the usual logistics of deploying as many weapons on the enemy as possible. An aircraft that can carry 4 missiles and runs out should head back so that it can sortie tomorrow again with a fresh load of missiles instead of starting a desperate dogfight. An expensive plane that can only carry 2 missiles but is guaranteed to return home is more valuable in a war than a cheap one that carries more missiles but gets shot down frequently.

Just remove the aircrafts altogether, use only a cloud of AI-enabled missiles....

There's a saying that drone warfare enthusiasts keep reinventing the guided missile. Proponents point out that drones are cheap and disposable compared to the value of the targets they attack. But a drone can travel further, faster, and with more armor and ordinance if it doesn't carry fuel to return or equipment to land.

The problem with missiles is that they have a limited amount of propellant. Most missiles will spend the majority of their flight time gliding toward their target after the initial boost phase. Fighters can spend hours in the sky, and fly thousands of miles away from their home base. Missiles are limited to a range of several hundred miles and a few minutes max, even with advanced loitering technologies.

If it's loitering, or intended to come back, it isn't functionally a missile.

And why in the hell would you fire a missile without a target that would require it loiter unless you were doing something completely mad like Project Pluto? That just doesn't really make sense at all. Though it reminds me of a torpedo from a Michael DiMercurio book that had to be sedated due to it's carbon based computing matrix being smart enough to basically guarantee a hit on an enemy combatant submarine, but not smart enough to understand IFF.

Loitering missiles are excellent at destroying targets of opportunity; a radar site that is only active temporarily, a convoy of armor moving from its lager, etc etc etc.

Isn't that what a Predator/Reaper is for? Maybe I'm lacking imagination, but a loitering missile seems very difficult to make.

So we’ll have AI controlled missiles instead? With small missiles attached to them to fire at enemy aircraft.

High-G tolerance is not a dogfighting panacea. Even without the pilot, there are issues when you want to take a large aircraft into 10+g range. Structural strength is one thing but there are also aerodynamic limits. The higher the G the higher the stall speed. Superhuman or AI, trying to pull 10g in thin air will result in flow separation (ie stall) and departure from controlled flight... ie you lose control and drop like a rock.

Missiles don't care about stalling. They are on a suicide mission to get as close to the target as possible for the nanoseconds necessary to detonate. Missiles, being small and dense, are much stronger than any aircraft.

Pilots do black out during high G maneuvers and the planes seem to be fine, so this is already a limitation for a human+plane combo that a plane alone would not have.

Planes generally can't pull more Gs than the humans inside without needing to go through an extensive maintenance process.

Even then, as has been said countless times in this thread, high-G maneuvers don't do shit against modern missile threats. They might be useful for dogfighting with guns, but given the range and accuracy of modern missiles, combined with proximity fuzes, it simply doesn't accomplish much.

Against a weaker enemy, you'll never have the need to pull Gs in the first place. Against an equal or stronger enemy, it won't do a damn bit of difference anyway.

Dogfighting with guns is a great way to stop an incoming missile. You just need the ability to do high-G maneuvers and aim well. Shoot the missile.

It helps that bullets as small as 12.7mm can be steered.

This comment is so far into Poe's law territory, I'm past the ability to tell if I'm being trolled or not.

In case I'm not being trolled: the way to defeat a radar-guided missile is to beam (fly perpendicular to) the radar source. For an missile with active or semi-active radar guidance, this means you need to fly perpendicular to it. And, preferably, fire lots of chaff.

This is fundamentally incompatible with the idea of flying straight at the missile trying to shoot it with guns. Head-on, a missile will close the last mile in under one second. During this second, you need to accurately track, predict the flight path of, and reliably put guns onto a target that's cold (all the fuel has been burnt) and has (in the case of an AMRAAM) a 7 inch diameter. And that very well may not even care about being shot up with bullets; at best, I think you'd kill the seeker, but within the last mile it's not changing course much to have to come close enough to kill you by the proximity fuze (of which I believe there are two, one in the nose and one in the tail). By the time the first of your slow-moving bullets has any shot of reaching it, it will easily be within a quarter of a mile.

This is not even a remotely plausible way to defeat an oncoming missile. It runs directly counter to the actions you already be taking to maximize your chance of survival. And on top of that, bullets are only going to damage the leading-edge parts of the missile which aren't particularly critical to its ability to kill you, at least not within the distance it will be by the time your hail-mary bullet salvo hits it.

We're assuming an AI here. It can aim. Bullets as small as 12.7 mm can even steer. Larger ones, such as those carried by modern fighter aircraft, can have independent seekers.

You said the missile had some amazing ability to turn, and now you're saying it has no fuel. You can't have it both ways. There is fuel, or there is no energy to spare. Being "cold" isn't generally possible; nothing cools instantly. Modern IR sensors do not rely on engine heat. Supersonic flight generates plenty of heat on the nose.

That last bit deserves more emphasis. Heat was a serious problem for the Concorde and SR-71, despite operating at much slower speeds. The missile is hypersonic. That is the speed at which air molecules come apart from the heat. IR detection will be trivial.

There is a circular circuit board that is very much needed for the proximity fuse. That is full width.

The speed of the missile contributes to its own destruction. Bullet impact will be devastating. The missile can not endure supersonic flight with a lopsided hole and a bunch of cracks, and it wouldn't fly straight even if it could. Tumbling means instant destruction, like Space Shuttle Columbia.

Shooting back is the action to maximize chance of survival. It takes priority; all other options are obsolete.

What you’re suggesting is possible technically but nobody does it because the idea of having to turn your entire plane towards a missile threat to neutralize it virtually guarantees your destruction if two missiles are fired at you in rapid succession. This is not a rare occurrence. At all. Yes, from multiple directions. That’s exactly how surface-to-air systems are set up.

It becomes slightly more reasonable if you turret the gun, until you factor in the weight of the necessary ammunition to neutralize multiple threats, the turret itself, and the independent radar used for target acquisition and tracking. A B-52 could pull it off, a fighter couldn’t yet. Not without massively compromising their payload and aerodynamics.

Chaff, flares, stealth and jamming are used because they actually work in practice. Jamming is a big one that people don’t hear about much. Wonder how the B-52 is still in service? Jamming. Wonder how wild weasels do their job? Jamming.

When you see a turret like that on a bomber then you can get excited for self-guided bullets shooting down dozens of multiple incoming missiles. But tbh at the rate we’re going it’ll just be a laser instead.

>Chaff, flares, stealth and jamming are used because they actually work

Don’t modern missiles use optical target acquisition in the first place?

BTW, tanks have active automatic ballistic defense systems for a while now.

Thanks have a much more robust complement of armor compared to an aircraft. They defeat the mechanism of armor penetration in munitions before impacting the armor. They still take a significant beating after the munition is "defeated". Airplanes don't have armor in the same way a tank has. Even small arms fire can take out an aircraft, generally speaking.

They now have active intercept systems: https://en.m.wikipedia.org/wiki/Trophy_(countermeasure)

The automatic ballistic defense systems a tank uses would obliterate an aircraft. They are basically shaped charges that explode outwards to deflect a munition. It only works because tanks have thick armor in a dense, heavy, package.

That is an old system, mainly used by the USSR, and mainly intended to act against long piercing projectiles of depleted uranium.

It is more common to have active electronic systems. For example, one such system fires grenade-like projectiles upward that then explode in a downward direction. The timing is accurate enough that the incoming attack is hit by the shrapnel. Another system is more like a shotgun.

The more reasonable comparison is ship-based defense systems. The USA uses the Phalanx CIWS, which fires simple dumb projectiles of the same diameter as those of a typical fighter plane gun. About 100 are fired to destroy an incoming missile.

Adding an AI to a modern fighter makes it an awful lot like a flying Phalanx CIWS. You have a nearly identical gun with a nearly identical radar.

Modern ATGMs are usually subsonic; none of the protective systems on tanks are designed to counter supersonic missiles or projectiles. There's a world of difference in countering a Mach 3 missile and a TOW or Hellfire.

Incoming speed doesn't help the attacker unless it beats the sensors, which operate at the speed of light. Mach 3 missiles impact Mach 0.1 shrapnel at about Mach 3.

That helps the defender. The faster the missile goes, the harder it gets hit by defensive shrapnel.

Speed makes the engagement cycle short; instead of having maybe 2 shots at an incoming missile, you get one. And if your system is too slow, it has no chance of reacting in time. Most systems are designed to defend against missiles, but are ineffective against cannon rounds (APFDS).

Now they automatically shoot counter-projectile: https://en.m.wikipedia.org/wiki/Trophy_(countermeasure)

It is beyond Poe. This is harry potter territory. Even if one had the magic bullets, you need to know where the missile is to within a few inches. Good luck with that. The missile is doing Mach4 and has the radar cross section of a small flightless ant. And your magic gun system has perhaps one or two seconds to track and engage that ant. Add to that the fact that your gun system is presenting a wonderful IR and radar-emitting target for the missile, and his swarm of buddies, to better home upon.

This is the sort of missile a fighter pilot is up against, 20 years ago. Look at what the sidewinder does at about 1:05. Good luck shooting that down.

https://www.youtube.com/watch?v=6YMSfg26YSQ

Why mention speed at all? Mach 4 is nothing. We can hit satellites and ICBM warheads going between Mach 20 and Mach 30. Speed of approach just means that the impact energy will be greater.

Impact energy grows with the square of the velocity. If the velocity doubles, that is like quadrupling the mass of the bullet, which is like turning 20mm ammo into 32.75mm ammo.

It's not like a small flightless ant on radar, and anyway a Mach 4 object will be hot from compression heating. Mach 4 is just below what it takes to rip air molecules apart. That's damn hot. IR tracking would be trivial.

There is no limit that says "one bullet only". Typically the Phalanx CIWS fires about 100 shells, and those aren't even equipped with seekers. It's just dumb 20mm shells, about the same as carried by the typical fighter plane, costing about $30 each. We use the Phalanx CIWS against incoming missiles. It works great.

BTW, that was an unimpressive video with no clear point. Yes, missiles can fly and go after targets. We all know this. It works especially well in the missile vendor's promotional video.

>We can hit satellites and ICBM warheads going between Mach 20 and Mach 30.

I think you're forgetting that a lot of anti nuclear defense systems use nuclear weapons themselves. When you only need to be within a few hundred meters of a target accuracy suddenly doesn't matter at all, especially when you have several hours to predict the flight trajectory of the enemy missile.

If laser based CIWS turns out to be viable on aircraft then we would see a focus on bigger aircraft that can defend themselves against missiles but even then they would not directly fly into a missile if they can avoid it.

Only the Russians do from what I can find, and only around Moscow in theory. Everyone else tends to take the judo of a kinetic impactor approach.

I really mean "we can hit". We use direct impact, without any explosive, to destroy ICBM warheads and satellites.

The systems to use nuclear warheads against incoming nuclear warheads were decommissioned half a century ago. I don't think that was ever in use against satellites.

Lasers are nice, but definitely not needed. Phalanx CIWS uses about 100 unguided bullets to destroy an incoming missile. Modern US fighter aircraft carry the same diameter ammunition. The plane is like a flying CIWS, minus the software.

Eh, maybe minus the software. If the software existed, it might not be public. Maybe the F-22 or F-35 already have the capability, limited mainly by pilot G-force survival.

Fantasy beats technology and physics any day.

Right: the fantasy Mach 4 missile that maneuvers with the engine shut down, does not experience compression heating from hypersonic flight, survives Mach 4 impact with tungsten 20mm shells, can track an enemy through a cloud of 100 incoming 20mm shells, and can maneuver with zero turn radius without losing energy.

I'd like some of that.

Phalanx CIWS was prototyped in 1973 and has been in service since 1980. Shooting down incoming missiles with machine gun fire has been viable for a very long time.

Back when I was playing the game the Moskit and its derivatives made the CIWS obsolete. The short engagement range meant that the debris of the threat would impact and mission kill.

The larger 30mm and RAM were created to extend that range. We planned to engage when the thermal plume was detected on the horizon.

The higher speed of ballistic threats meant that the pK of the SM-3 was pretty low. Luckily finding ships at sea is hard.

In a fight against a first line threat it was best to be in a boomer.

Has Phalanx ever been tested against a supersonic missile?

Also, there is no way to install such a system on a fighter airplane.

> Mach 4 missile that maneuvers with the engine shut down, does not experience compression heating from hypersonic flight

Well, if it's only Mach 4, it won't ever experience hypersonic flight, so that's pretty much a given.

> Phalanx CIWS was prototyped in 1973 and has been in service since 1980. Shooting down incoming missiles with machine gun fire has been viable for a very long time

Sure, if you are firing at much slower missiles and you've got a platform big enough to mount a rapid-rotating turreted Vulcan on, and even then it's not super reliable which is why it's being replaced or supplemented in newer deployments with systems based around the Rolling Airframe Missile.

I would suspect that that's due to there being no point in the past of engineering aircraft that can do acceleration that would kill the pilot inside.

Now that no pilot is necessarily needed, different choices can be made.

No pilot doesn't change the fundamental calculus that high-G maneuvers are virtually useless against an oncoming missile.

Even if you could pull an instantaneous 10G acceleration completely perpendicular to your direction of flight without shedding airspeed (think: sideways-mounted rocket motor), in the time it takes a missile to close the last mile, you've moved less than 100ft. And all the while, the oncoming missile is actively tracking you. Even if it misses that's no big deal, since it uses proximity fuzing to blow up at its closest point to you.

High-G maneuvers are great for close-in battles to outmaneuver another plane and prevent a guns solution. They're becoming increasingly useless against modern IR missiles like the AIM-9X, and they're absolutely useless against a missile approaching at ludicrous speeds from 75+ miles away.

Might also be useful for firing missiles, without having them to expend their fuel, giving them a better starting angle, less time wasted in maneuvering around...

> will result in flow separation

With thrust vectoring and a TWR > 1 this is less of a concern.

No, this is ridiculous. It's absolutely a concern. Having attitude control at stall speeds makes for impressive air shows but doesn't help even slightly against an oncoming missile.

Sure you can point the nose wherever you feel like, but from the missile's perspective all you've done is given up your ability to make abrupt changes in direction and made its job of predicting your future location easier. Without lift under the wings, you're spinning the plane around on its axis but you're not actually changing the plane's motion vector as much as if you'd made a coordinated turn.

And on top of all this, a missile doesn't actually have to hit you directly to kill you. Missiles are proximity fuzed, so simply blow up next to you. Even a fancy 10G maneuver (a maneuver with thrust-vectoring while stalled, by the way, is ~0-1G since you're mostly just a ballistic object) isn't going to help much against a missile at Mach 4 that can pull 50G.

It doesn't have to be a panacea, or even decisive, to be a factor worth mentioning.

pilots don't really take up as much space on these planes as you'd think. these planes are huge. the F22 is over sixty feet long. if you replaced the pilot, controls and screens, weight wise you might get to carry one more AAMRAM, but that sucker is twelve feet long.

These are large aircraft (though the F-16 is significantly smaller than an F-22) but the pilot + ejection seat + airtight pressurized compartment + life support systems + displays and controls really starts to add up in both the mass and volume budgets.

But the pilot requires more design compromises. The requirement to have a large canopy protruding from the fuselage for maximum visibility is a challenge to both building a stealthy aircraft and aerodynamics.

Also, currently the pilot and their circulatory system keeps aircraft from turning at more than ~9G (and then only for brief periods of time). Without a pilot, it would be possible to build aircraft with stunning maneuverability.

The maneuverability thing is mostly a myth. Sure it's theoretically possible to build a drone that can turn at >9G. But it won't have sufficient thrust to sustain that rate of turn without losing a lot of energy. And now the airframe has to be stronger to handle the stress, which means heavier, and that compromises mission effectiveness in other ways. And for all that it still wouldn't be able to out maneuver a missile in most realistic conditions.

A better option would probably be to design drones that are less maneuverable. Sure more will be shot down. But that will lower the price so just manufacture more.

Following that logic to its conclusion, you could save a lot of money by skipping out on the fuel and carrying capacity to make it back home.

Another commenter said it first, but pretty much any set of reasoning that gets you to autonomous drones eventually leads to simply making guided missiles.

A lot of the interfaces for the pilots add complexity that could be reduced in the case of an autonomous aircraft, not just weight or complexity of interface. But actual complexity in the design and development stages. They also add constraints on expanding the capability of the system as new controls have to go somewhere, either displacing existing ones, changing the "buttonology" of the existing interfaces, or being tacked on to a mostly empty (hah) corner of the cockpit.

If we didn't have to care about those things, changing the existing infrastructure in an aircraft would be greatly simplified (not simple, but simplified).

Have them fly head on, looking out the nose, like a bob racer. Place the radar somewhere else.

The airframe can handle more Gs than the pilot can though, so if you don't have a pilot then in theory it opens up maneuvers that were not previously performable.

None of these maneuvers are useful. High G maneuvers rapidly bleed energy. Slowing yourself down for a missile is... not a winning strategy.

High-G maneuvers are great for an in-close adversary moving relative to you. They don't do much against an adversary approaching at Mach 4. At that speed, while pulling 10G, you'll make less than a hundred foot change in your predicted position while the missile closes the last mile.

And keep in mind, the missile will be actively tracking you during this period.

And also will proximity detonate, it doesn’t even need to hit you.

That's why I said "in theory".

Thank you for being a voice of reason in all of this.

In the context of dogfights, why do you even need an aircraft at that point? A swarm of small, cheap, explosive drones with a kamikaze AI sounds like the end of human-piloted warplanes to me.

I don't think dogfights are the future of air-to-air combat.

> A swarm of small, cheap, explosive drones with a kamikaze AI sounds like the end of human-piloted warplanes to me.

This already exists. We just call them surface to air missiles or air-to-air missiles.

Modern air superiority combat is point-and-click. Send your air superiority fighter over, shoot a missile from beyond the sensor range of the other craft, and then blow them up.

----------

Air support fighters need guns (see A10 warthog) and other ammunition. But will we really be using guns in the future of air to air combat?

Runways, SAM Sites, Fuel Depots are hit at the same time. You gain air superiority very quickly

Bombarding the opponent with rockets and artillery usually wins. We've been doing that for decades.

The issue is that advanced opponents (ie: Russia or China) will have air defense of their own. How do you expect to take down a SAM site? Its specifically designed to destroy air targets as they approach.

With what, with air-drones? Against a hardened anti-air specialty base?

This is where electronic warfare comes in handy: "jamming" the radar of a SAM site for example will blindfold them. But the battle of information (jamming, stealth, and advanced radar) is constant, there's no guarantee that your jamming or stealth technology will be good enough for the next generation of Radar.

And no. You don't approach a SAM site (or any area a SAM site is defending) from the air. US Patriot systems can take out SCUD missiles traveling at Mach 5.

And if your "kamikaze drones" are flying at Mach 5 with an explosive payload, that makes them effectively a SCUD missile.

There are reams of doctrine about how to approach a SAM site from the air. Look up Suppression of Enemy Air Defense (SEAD) or Destruction of Enemy Air Defense (DEAD).

Jammers (think emit a huge amount of energy on the frequency the radars use) and anti-radiation missiles (think radar homing) are a big part of it.

How many incoming bombs, cruise missiles, and other threats can a modern SAM system engage at once? According to wikipedia, a SA-20 carries 4 missiles per TEL, and each command vehicle can control up to 12 TELs. That gives the command vehicle 48 missiles to play with. There is a 0.7 chance of intercepting a "tactical ballistic missile," which is the equivalent to a SCUD, and is probably the easiest thing to intercept. If you fire two missiles at each TBM, you have about a 91% chance of intercepting each missile, meaning you only need to fire 7 TBMs at a site in order to have a fifty chance of one of the missiles making it through, if they fire two missiles at each one. In the meantime, the SAM sight has expended 14 very expensive surface to air missiles. BTW, the .7 figure is likely under favorable conditions, which most battlefields would not be, as there would be jamming, decoys, cyber attacks, and lots of other clutter. This is also for ballistic missiles. Terrain following cruise missiles and precision guided munitions would likely be more difficult to target. [1]

In modern warfare, the offense still has the advantage, if you don't run out of ammunition. [1] https://en.wikipedia.org/wiki/S-300_missile_system#S-300PMU-...

> And no. You don't approach a SAM site (or any area a SAM site is defending) from the air.

This is completely mistaken. SEAD (suppression of enemy air defenses) is one of the most basic and important missions the air force and naval aviators train for. The US absolutely does attack SAM sites from the air, and has done so going back to the Vietnam war. In 91 Saddam had a state of the art integrated air defense system that was systematically demolished by air in the initial days of the counterattack, long before ground troops got close to say Baghdad.

Google what the acronym at the bottom of the "Wild Weasel" patch means for a laugh.

Nice. :) For the lazy: https://en.wikipedia.org/wiki/Wild_Weasel

If all else fails, there's always the intercontinental ballistic tungsten telephone pole. Nothing short of a nuke can stop it.

The Earth does a pretty good job. Just don't stand in the way, and you won't get hit.

> How do you expect to take down a SAM site? Its specifically designed to destroy air targets as they approach.

> With what, with air-drones? Against a hardened anti-air specialty base?

Yep, air drones (or, to use the more common name, “missiles”, whether launched from the surface or launched from manned aircraft) are by far the most effective technique.

In fact, there are missiles optimized exclusively for that (anti-radiation missiles for taking out radars sites.)

> You don't approach a SAM site (or any area a SAM site is defending) from the air.

Yes, you usually do.

> US Patriot systems can take out SCUD missiles traveling at Mach 5.

SCUDs are ballistic missiles, which at any given speed are the easiest thing to hit, since they don't maneuver.

> And if your "kamikaze drones" are flying at Mach 5 with an explosive payload, that makes them effectively a SCUD missile.

No, it makes them hypersonic cruise missiles. Which no one, anywhere, has a particularly good counter for right now.

> This already exists. We just call them surface to air missiles or air-to-air missiles.

GP was effectively describing SAM with loitering capability and those surely don't exist yet.

*AAM (air-to-air missile)

Certain anti radar missiles have a loiter capability where they deploy a parachute, looking down and waiting until their target lights up.

Not as rare as you'd think. Even Iran seems to have them.

https://www.google.com/search?q=loitering+sam

> Air support fighters need guns (see A10 warthog) and other ammunition. But will we really be using guns in the future of air to air combat?

That transition has already happened. In recent conflicts the US has favored precision guided munitions over gun runs, even when sending the A-10 in somewhere. The A-10 is an internet favorite, but the reality is that it's expensive and obsolete. It's honestly a travesty congress keeps throwing money at it.

Bullets are cheap. There's a reason pretty much every 3rd world air force slings gun ponds under its ground attack aircraft. No reason to spend a multi-thousand dollar bomb when 20mm will do.

That no longer works when the adversary has any sort of credible air defense. Can't risk manned aircraft at low altitude any more.

OTOH, an plane with a very precised gun may "win" an encounter by circling above a well-defended area, having SAMs shot at it, shooting down the SAMs as they approach, while being cheaper to operate (and replace when it eventually does get hit) than the missiles shot at it.

While at the same time occasionally dropping bombs at targets to make ignoring it not an option.

The defender in this scenario has a seriously disadvantage, in that the local SAM sites can send many missiles at the same time, and the anti-missile suite of the plane has to defeat every single one. A gun that would succeed in this scenario would be a massive technological leap. The US Navy uses gun-based anti-missile setups, but they are more of a last ditch sort of defense that can defeat a small number of incoming missiles by putting an absurd amount of bullets in their way.

The plane is the attacker, unless someone invaded their home and put SAM sites everywhere while they were out.

> can send many missiles at the same time, and the anti-missile suite of the plane has to defeat every single one.

That's what makes unmanned planes special: It doesn't. It just has to force enough missiles to be fired that the missiles are more expensive/valuable than the plane when it ultimately gets shot down.

(Value can be different from financial cost, depending on e.g. supply shortages, influence on the rest of the war, etc.)

Pretty much all ground attack aircraft that aren't just multi-role fighters with precision munitions hanging off of them are crap if you don't have air superiority.

We can and we will risk manned aircraft at low altitude against a credible air defense. The war just has to be meaningful enough.

Nazi Germany had a credible air defense. We sent bombers anyway. At one point, the expected lifetime (survival) of a bomber crew was a month.

It has been some time since we fought a war that was worth that risk.

There is an interesting angle on modern great power conflict. Even in a non-nuclear war, all modern combat systems, munitions, and manufacturing bases, would be destroyed or expended within the first weeks by precision munitions. Pro-longed non-nuclear conflict would likely require tradeoffs towards lower-tech components that could be manufactured in more locations at lower risk of bombardment.

They're ultimately just aren't that many aircraft or cruise missiles in modern militaries relative to WW2, and they each cost 2-3 orders of magnitude more than their WW2 counterparts (although the cost increase matches the capability increase).

You can bet we will be risking manned aircraft and low altitude for a very long time to come. I don't think you understand military priorities in times of war.

Sure they will take the risk at low altitude when there's no other option. But the preferred technique now is to orbit at medium altitude and dispense precision munitions. Why expose the aircraft to ground fire when it's not necessary?

You'll want a two-stage system of some sort anyhow. There's a fundamental tradeoff between the ability to carry the fuel needed to gain height, maneuver into the correct position/heading/speed, loiter, etc, and to have the extreme power-to-weight ratio and maneuverability for a suicidal final attack run.

A plane that is flying high enough and can travel fast enough is essentially immune to ground-based interception. Your launch has to spend a significant amount of time and energy just to get high enough, and launches like this are extremely loud to the defending plane's sensors. If you launch too early they can just turn around and fly out of range; too late, and they can overfly and outrun your range.

Essentially, the position + velocity interception state space for smaller kamikaze designs does not necessarily include ground level & stationary for a large portion of the design space. And once you start up-scaling things to get the endurance necessary to attempt intercepts, then you lose the maneuverability and cost advantage for making kamikaze attacks, and attaching parasitic missiles or drone aircraft for the final attack run makes a ton of sense.

Please show me a swarm of small, cheap, explosive drones that can't be avoided by using radar and outdistanced/outmaneuvered. To do that you need medium to large and expensive drones.

Well, as others have mentioned, the current version is called "missiles". They require a platform to launch from.

I would imagine a human-less plane would also be able to pull off manuvers that might be too stressful for the human body to handle.

Right, humans can only handle 9Gs, I see no reason why removing the human wouldn't allow pulling way more Gs

Say you're the president of Elbonia, and you get a report that an unidentified aircraft is in your airspace.

Without a human pilot in a fast jet, how do you get up there to take a closer look at the situation to decide what to do?

Modern warfare is limited in nature, with rules of engagement, etc.

Iran just shot down an airliner because they had poor ROE and control of their systems.

Just because something looks like a threat on sensors does not mean it is a threat. It could be that natural phenomena are masking things. It could be a ruse, such as an adversary is disguising a civilian aircraft to look military, or vice versa.

This result with dogfighting is interesting, but is only a very small part of what fast jets are useful for.

What about cameras?

> Iran just shot down an airliner because they had poor ROE and control of their systems.

So did the USS Vincennes. https://en.wikipedia.org/wiki/Iran_Air_Flight_655

Also, higher maximum G limit. AIs should be inherently much more maneuverable that human pilots because humans can only pull -1 to 9 Gs, AI would only be limited by the aircraft itself.

This is of marginal value. Pulling Gs might help you in a close-in dogfight, but when your primary threat is a radar-guided air-air missile at 75mi+ or a SAM site at 200mi+ of range, Gs don't really matter unless you've failed multiple other possible means of defeating the incoming missile. And even in this case, high Gs aren't going to be much more than a hail mary.

Of course it will- autonomous aircraft that can zigzag to counter threats.

But nowhere near as fast as a missile that weighs 1% and has a much higher power-to-weight ratio.

No, that's not even remotely close to how aircraft defeat modern radar-guided or even heat-seeking missiles.

Even when you take the fleshy bits on the inside of the plane out of the equation, a plane that's capable of carrying multiple missiles and/or bombs is simply not going to be able to out-turn a missile that can pull 50G+ approaching at Mach 4 with any sort of regularity.

Pulling hard Gs is all but pointless because even if you cause the missile to "miss", the missile doesn't care about actually impacting you. They're proximity-fuzed and will happily detonate right next to you instead.

Didn't say that's how modern aircraft do it. They dont. I envisioned future drone aircraft that behave like an aircraft and missile hybrid with a smaller radar cross section.

A large part of getting a missile to target is predicting where the target will be. Zig zagging hard and rapidly ruins the prediction and you don't have to do it all the time- only when it knows it's illuminated.

IIUC The missile pretty much compellingly maintains it's advantage here for as long as the "hybrid" approach maintains non-missile-like design features. Which leads you to either giving up on the approach, or just observing that it's hard to hit a missile with an equivalent missile.

I think the issue is going to be that the AI can pull G-forces higher than what a human can do and also never having to be re-trained.

This means a swarm of "cheap" craft that focus on deploy 20 to take anything out is going to be the optimal. And without the need for all those human systems (glass cockpit, AR consoles, ejection systems, seats, etc) we're good.

For non-combat operations they can be flown similarly to drones.

I'm not sure they wouldn't require retraining. If all the pilots are AIs, the tactics would change, necessitating retraining. Then there would be the search for better AI tactics. You'd still be in a position of a tactics race, driven by simulation and machine learning probably, so the evolutions would iterate faster, and the deciding factor would become the computing power each air force could throw at it.

With perhaps the exception of the training, wouldn't most of those problems (and some other ones mentioned in this thread) be something that can be solved using remote controlled, but still controlled by humans, fighter jets?

There must be a reason that I'm missing why this isn't already a thing, anyone got an idea?

gamers worry about the latency of a wireless controller sending a signal from their couch to their TV.

for a fighter jet with a remote pilot, you're looking potentially hundreds of miles of range, and the latency of the video transmission to the pilot combined with the latency of the controls signal back to the plane. i can't imagine that being acceptable for any sort of dogfighting scenario.

Ironically, game controllers can have a latency of more than 20ms. Light travels more than 6000 km in a vacuum in 20ms. That's a pretty good excuse for gamers to complain about that sort of latency, I think.

Bluetooth is bad enough that videos have to be delayed so that audio can catch up. Completely worthless for games unless you use proprietary codecs like aptX.

I wouldn't say it's that bad, I flew a drone controlled via bluetooth with my phone, for instance.

Buffering audio is actually a great answer to disconnections. Plus, it's less energy-intensive to fill the buffers at once than to stream it as needed. And since it's pre-recorded anyway, you already have the whole buffer. Yes, initial transfer can take a while, though.

But in theory, the play/pause operations shouldn't have latency, and the bluetooth device could clearly communicate its state. I'm not sure how audio is actually implemented in the bluetooth stack, though, I've only ever done serial communication with it.

That's before taking into account enemy attempts to jam, spoof, or otherwise disrupt your control link.

remote control has a rather extreme weakness against advanced enemies.... jamming.

AI scales better to how many jets you want to use. AI improvements can also instantaneously be rolled out to every jet. There is no need for continues live communication in "hot" combat. Less video leaks when they kill a bunch of civilians again.

Also think about the export market! Countries that now can buy the ruthless-pilot with the plane! And like with so many exports, you can give them a shittier version of the AI, so they do not become a danger to your forces!

Or just a version that can be hacked or revolted by giving them a signal. That is why chinese network equipment is so dangerous in national defence. You may not know the kill code. But then who will buy them.

Reliable comms is far from guaranteed in a war, and if there's a gen. 4 fighter dogfight involved, it presumably a median sophisticated adversary capable of EW.

Also opening up the ol' remote desktop port on an F16 is a major cyber risk.

Also it isn't an either/or type situation. You can have both. The Air Force is working on "Loyal Wingman" where you can have a squadron with a manned system helping strategically target and direct their "unmanned wingmen" ai based planes.

The biggest advantage I bet (in a real AI fighter) is removing the person from the cockpit. Modern jets have been able to fly harder than humans can reliably stand for a while, being able to pull higher Gs means tighter turns.

The real question is how much does gun dogfighting ability matter now because guns are almost vestigial on modern fighters. Dogfights now are missile affairs and this test was a gun based one.

https://www.businessinsider.com/f35-pilot-f-35-can-excel-dog...

AI can run the show locally, with a remote operator making the actionable decisions. Military drones already operate this way, without much onboard smarts.

Not something I'm excited about, but likely inevitable.

I think what we're actually going to see is something like assistant drones that go with the pilot. So using the f35 as a command&control system to give orders to the ai drones. And/or just have them backthem up in combat anyway.

The issue with AI is that combat can change so quickly. It's hard to make the right calls.

Let the AI handle Air to Air defense while the human pilots take out what targets they can read by their discretion etc.

This is explicitly part of the plan for the f35. It also frees up the drones from having to carry sensor equipment, and lets all of the drones benefit from a single sensor suite too heavy for the drones themselves to carry.

Not to mention those pesky G-Force limits that humans are subject to. They black out long before aircraft has reached its limits.

I'm almost excited to blow another hundred billion dollars on designing a next-gen aircraft without any G-Force considerations. Current jets designed around a human pilot can already achieve mind-boggling performance of over 10G on multiple axes, what would a blank slate aircraft with none of those limitations be able to do?

Not that much more, but you'll be able to control the aircraft at all times because the 'pilot' doesn't black out in high-G situations. Maybe save a little bit of weight, smaller radar cross section.

You don't think there'd be a bunch of compounding design decisions available to engineers? Pilots currently black out in high G situations so the flying envelope is basically constrained to acceleration / turning radii that are at the edges of pilot abilities. Sure if you removed the pilot in current planes, you'd be able to hit the extremes of that envelope, but if you were to redesign it without ever thinking of a physical pilot, you could have a much wider envelope. What would they be capable of?

Something like this but with a ~full-size airplane:

https://youtu.be/AXSfFLGeVZA?t=53

> Of course, that is just raw combat, not other areas of flight where greater human intelligence is of more value.

You don‘t need the AI to replace every plane or pilot, you could have a traditional plane and pilot to do all the human things and an AI or two to protect the human or engage the targets. Mixed AI/human squadrons.

Would you even need a good AI or equipment to be sufficiently disruptive? I imagine a swarm of crudely moving, low cost drones would win most encounters with more complicated targets. This feels like a boon to smaller nations who can quickly gain an edge by capitalizing on existing research.

Or mixed in with human pilots as a force multiplier. I think that is how they will start being used, so that there is always a human observer. At the start.

Not having a squishy human in the machine makes doing a lot of crazy maneuvers a lot more doable.

Wouldn't that make the jet just a fancy missile ?

Also not being limited by G-forces or endurance.

(works both ways)

and the enemy doesn't need to beat the AI in dogfight, just the infinite side channel attacks on the training set.

WW3 might be won by adding a few high intensity LEDs to some old MIGs.

https://www.businessinsider.com/clothes-accessories-that-out...

this is a tough one, to make an exploitable attack you would need at a minimum to have a copy of the flight AI and be able to simulate possible encounters. Given the relatively finite number of side-channels available, it's probably far cheaper for the AI builder to run such simulations and adversarially test against the main exploitable variations such as blasting the aircraft's sensors with bad/confusing data.

There just aren't that many methods of interacting with a plane >1km away, and even fewer at the 50-100KM ranges that modern fighters are built to fight at.

> to make an exploitable attack you would need at a minimum to have a copy of the flight AI and be able to simulate possible encounters.

I'd disagree. Data is more important than the model. But also you can do black box attacks on AI pretty effectively. What you do is train your AI to perform a similar feat and then figure out its weaknesses. Those weaknesses are likely to translate to the other AI. Robustness is a difficult task.

As for blasting confusing data, well that already happens to human pilots. There are radar jammers, targeting jammers, visually occluding devices, etc. But hey, if an AI can operate under 90% of typical dog fight situations, then that's a big reduction in needed pilots. Then those pilots can be all trained to be aces.

If weaknesses can be known in advance, then they can "just" be incorporated into the training environment. Unless you can feed the sensors "false" data, there aren't necessarily that many possible exploits.

You should read up more on adversarial machine learning. If you could solve things so simply you'd get a lot of papers really quick (and lots of money).

When you shoot down an aircraft it leaves a wreck on the ground. You can extract the AI and find weaknesses that way.

I’m not sure how that business insider piece is applicable. Fighter jets have a huge range of available telemetry, not just visual spectrum cameras. High intensity LEDs might fool a camera but they won’t fool radar.

I don’t think the problem is fundamentally different from the pre-AI state of electronic warfare and countermeasures.

Ah, but you need to compute to make that happen. More compute comes with greater trade-offs in terms of potential side channels to exploit. You're absolutely right.

It's just another step in the arms race ladder.

Or just some old migs with high intensity lasers designed to constantly paint the cockpit. Humans have exploits too

LASER -homing missiles might negate that.

Reply on news.ycombinator.com