AI slays top F-16 pilot in DARPA dogfight simulation | Hacker News Reader

AI slays top F-16 pilot in DARPA dogfight simulation

breakingdefense.com

331 pointsMontagFTB423 comments

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.

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.

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.

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 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.

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

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.

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 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

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.

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.

(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.

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.

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

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 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

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.

Also not being limited by G-forces or endurance.

> 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?

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

> 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.

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...

The fifth generation of fighter aircraft marked a significant departure from the idea of a fighter jet being anything like what we think of as a fighter jet.

They're now designed and employed as fighter 'platforms'. Instead of zipping around the battlefield guns and missiles ablaze in 1-on-1 combat, they're low observable long-range systems designed to understand and disrupt the battlefield by employing their electronic and long-range missile systems.

These AI are a natural extension of that. I imagine the 6th generation of fighter platforms will be commanding swarms of fighter drones to do the fighter part of their role.

I'd imagine one of the (forget the name, the big planes with radar that buzz around the battle space), could handle a swarm of drones from a long distance. Not sure that it'd make sense to have a speedy jet out there doing that.

Gremlins concept, deployed from C-130?

AWACS is the role, E-3 Sentry is the current plane. It's a 707 full of radar, but it's also a giant flying target. If you're going to link a bunch of drones to one platform, the drone controller is going to need a lot more survivability.

I believe you're thinking JSTARs? https://en.wikipedia.org/wiki/Northrop_Grumman_E-8_Joint_STA...

Or maybe the E-3 Sentry aka AWACS, but the JSTARs matches the "big plane" better and has a giant synthetic aperature radar for imaging large swaths of the battlespace. We deployed with JSTARs when I went to Iraq as a Shadow 200 TUAV Pilot myself.

>but the JSTARs matches the "big plane" better

Sounds like you would know better than me, but aren't both the E-3 and E-8 based on a 707 platform?

And don't you need both? If you're doing an air superiority mission you need an AWACS. If you're doing CAS you need JSTARS, right?

Yes, you're actually correct here. I interacted often with the JSTARs crew (we were on the same base in Mosul for a few months) and not as often with the AWACS crews. JSTARs was the first thing I thought of when you said big plane with radar though as it is literally a big plane with the sole purpose of SAR. The AWACS is more or less a similar thing, but with additional EW (electronic warfare) kit and sweeps the sky.

They're complimentary although the AWACS is more relevant today than the JSTARs is. Most of the hardware in the JSTARs hasn't been updated in decades and it is starting to show its age. We had trouble integrating our realtime video feeds from our UAV with the telemetry feed from the JSTARs because it is just clunky and old.

The problem is going to be keeping an AWACS close enough to command and control the fighters while not risking getting it shot down. An F-22 or F-35 (or even an F-15) is going to be a lot more survivable than an AWACS.

> is going to be a lot more survivable than an AWACS.

AWACS have vastly superior radar and electronics. It's going to see threats long before they are in range. And it's going to direct interceptors to deal with those threats long before they are in range.

I don't believe an AWACS has ever been shot down.

This lists a few times when AWACS detected and directed the destruction of threats https://en.wikipedia.org/wiki/List_of_aircraft_shootdowns

Many advanced nations now have "AWACS killer" missiles with 400+km ranges. I think this is why the sensor suites on 5th generation aircraft are so good.

An AWACS has never been flown in an area where it is likely to get shot down. That RADAR is a nice juice target for a HARM, and the maneuvers you can do in an AWACS is not nearly what you'd be able to do in any fighter. I don't understand what you are arguing. Are you suggesting an AWACS would be more likely to survive an attack than a fighter?

The longer the distance, the greater the risk that the control data links will be disrupted by the adversary's EW.

Slightly off topic, but how do they disrupt gyroscopes ?

FTA

"The overarching ACE concept is aimed at allowing the pilot to shift “from single platform operator to mission commander” in charge not just of flying their own aircraft but managing teams of drones slaved to their fighter jet."

Such a concept is well under development as the Air Force Loyal Wingman program. Boeing unveiled protoypes not long ago [1].

https://www.boeing.com/features/2020/05/boeing-rolls-out-fir...

Sounds like a Protoss carrier fleet from StarCraft.

Let's go back even further https://www.youtube.com/watch?v=BeZNoEVycOc&feature=youtu.be...

(Ur-Quan Dreadnought from Star Control II)

So, Ender’s Game?

The US Air Force is already looking at this. I believe the plan wouldn't be to tie it to a specific "generation" of fighter, and will allow fifth, and maybe even fourth generation fighters to lead a "swarm" of fighters.

https://afresearchlab.com/technology/vanguards/successstorie...

The British Navy were already looking at fully automated dronecraft carriers back in 2015-16 too. Using smaller drones as armaments instead of missiles and the like.

John Boyd must be rolling over in his grave

Boyd wasn't just a pilot but also a strategist. One of his biggest adherent is the USMC. I don't imagine him rejecting this outright. I would love to know what he would have thought of AI because the potential speed with which AI can digest observations and make a decision could be an order of magnitude greater than a human. Boyd's EM theory was heavily influenced by analysis he and his collaborators did using early computers. I can imagine him being awed by AI and embracing it.

I can't wait for the first time an AI squadron gets a sortie phoned in and it turns out no one double checked to make sure the info was good.

Are there any metrics on pilot's making abort calls and not attacking anywhere?

This is likely to cause several interesting problems in my mind. For instance you may be about to sacrifice a lot of current talent to actually "fight" these planes, since you're now mainly focusing on the ability to coordinate and micromanage the battlefield instead of being physically/mentally able to assume the physiological role as primary control unit of the aircraft.

Also, to be honest, these AI controlled fighter platforms scare the shit out of me because there is now potentially fewer human decision points in the system.

Like it or not, you can end up with many times the destructive power in the air orchestrated by 1 guy without having the requisite sanity check of "Excuse the hell out of me, sir, but you want me to bomb WHAT?"

The capability to look at a situation and decide to call off is a feature of warfare that I think is frankly underappreciated.

We may very well be working unintentionally toward creating a world where a small

Holy hell, it'd be nice if I finished my thought. We're creating a world where fewer and fewer people are capable of marshaling more and more destructive power.

This does not bode well in terms of the law of large numbers being able to temper the extreme characters that setup may invite.

In the extreme scenario where an entire mission involving 1000 aircraft is lead by a single commander it is entirely possible that the lone commander was a spy and turns against his own country.

In my opinion this system is exactly the scenario that they were talking about when a bunch of ML researchers and stuff signed that letter saying it was necessary to ban autonomous AI weapons.

And it is bizarre that no one seems to be recognizing this.

But why would that human commander have to be in the air, putting his life on the line more than when they would be further away and on the ground/in a ship?

The only answers I can think of are a) that having human eyes in the sky still has advantages, or b) that long range communication is too unreliable (either as is, or because of possible enemy interference).

You may need an AWACS to direct the battle, but would it have to have human in board?

But is either true? If so, what’s the reason? Is there a c) “we can’t tell the air force yet that pilots who actually fly are a thing from the past”?

> But why would that human commander have to be in the air, putting his life on the line more than when they would be further away and on the ground/in a ship?

Latency + Bandwidth.

... in a standardized test that has nothing do to with real enemy encounters - the complete title.

We've seen this before, even for a lot more limited and controlled environment like the game Dota the "AI" can beat the humans a few times before the humans learns to exploit it's many weaknesses.

Some program that self learned a (admittedly impressive) number of reactions based on seeing/playing a huge number of simulations is not intelligence so not AI. It cannot reason on the spot and will fall for the most ridiculous of traps. For example the AI that beat the professional Dota players fell for running in circles around a tower forever while getting slowly damaged to death. Even the most simple of mammals (which we do not consider intelligent) would react to the pain at some point and bolt.

My theory is that AI will not exist until we reach AGI. Because with specialized AI you can always fall outside it's area of "expertise" and behave like a stupid bot.

As much as I agree with your assessment.

You only get ONE chance to beat AI in real life dogfight. So you better make sure that your strategy to exploit AI is going to work.

Ejection is a thing. At this point it’s also about who has better missiles and more of them.

The USAF missiles will make guns obsolete doctrine of the 1960’s was premature not completely wrong.

It's almost always about who has better missiles and longer range radar/sensors. No point in getting into a dogfight if you can shoot the opponent out of the sky from 100 miles away.

Until it isn't, because the enemy has already captured and compromised your government back home.

There is probably a way they know his, but can you really fire a missile at something on a computer screen miles away and be fully confident the thing you are shooting at is enemy military aircraft? And not say a civilian aircraft ?

I guess if its going fast enough its almost certianly a military aircraft. Can the radars in the military jets identity the type of the aircraft before it gets blown out of the sky?

https://en.wikipedia.org/wiki/Identification_friend_or_foe

That lets you distinguish friend from not-friend, but not enemy from airliner.

Another way is to announce a no-fly zone and then assume anything within the zone is hostile. (Yes, civilian planes can still fly there, but they've literally been warned at that point)

I assume this would happen pretty quickly when a war went hot.

Radars on the F-15 can determine the type of turbine on an aircraft...

In general yes you can make that distinction but more importantly this is a problem for peace time.

No one would authorize a non-visual shoot down of a target in a non-active AO and not during an actual conflict.

If you are at the stage where you have factions duking it out in the skies especially these days were talking about a major conflict in which case there won’t be any civilian jets flying there anyhow.

This is why people train. Once AI piloted planes are prevalent on the battlefield, there will be a huge effort to capture enemy systems and develop tactics against them. The USA has been doing it for decades (https://en.m.wikipedia.org/wiki/Have_Doughnut), I’m sure that China and Russia have as well.

But how do you know which version of AI are you fighting?

Its not really all that simple, as you have limited time/resources to determine what exactly you are up against.

And you get to test your hypothesis once. You win or lose after that.

The biggest thing you seem to be missing is that none of your arguments are unique to fighting AI. You still only get one chance against a human pilot. You don't know how good they are individually, you don't know if they have any special tricks, and you don't know whether they've got all the available upgrades on their jet, or if they have some old or broken equipment. You do know the average skill level of the enemy pilots, and what equipment they might have, so you make your decisions by assuming that they are a competent pilot with the best equipment that is likely to be on that jet. Warfare still happens even when you don't have time to gather up perfect information and make a perfect decision.

One other thing to note is that there will not be a unique AI in every individual jet; that would be impossible to maintain. There will be one version rolled out to each type of jet, maybe with a few customization options that can be changed during mission planning. When a new AI version comes out of testing it will get rolled out to every plane in the fleet that has the required computing power- some jets used for training/reserve/national guard duties won't have the latest hardware upgrades and won't get the newest version. So a pilot will be able to ID a jet, know where it is generally strong and weak, and then try to target the enemy's weakness while avoiding the strengths. Sometimes they'll lose, but once enough people have survived encounters with AI jets (either from exploiting stolen jets or from plain old combat) the best practices will get pushed to the other squadrons. Just like what already happens.

Except an AI tweak which changes behavior enough can be rolled out - let's face it - probably via a wi-fi update within minutes (aviation is, ironically, pretty simply compared to self-driving cars - we're not talking about big algorithms in a lot of cases).

The side with AI planes is just looking up K/D ratios on their telemetry from the planes. The other side is losing pilots.

That's now how wars work. You get time to learn and adapt to the enemy strategy. One plane down doesn't mean anything if you learn how to exploit the enemy's entire fleet of "AI" with 1 casualty.

In case of a prolonged conflict you need to consider the AI and it's programmers as a joint system. The AI by itself may be exploitable, but the programmers will adjust it to counter anti-ai strategies.

We don't program what are called AI's nowadays. We curate data, sometimes supervise the learning process, and do our best to detect when a model is bad.

If you think there's any intention or well defined predictability to these things, I've got a bridge to sell ya.

"That's not how previous wars worked" would be a more accurate take. Considering how few aircraft there are in modern militaries and how long it takes to manufacture modern fighter aircraft and missiles, one side will be out of high tech aircraft/missiles before the other and then it's game over.

This is one of those things I kind of wonder about the reality of. It's absolutely true that you're unlikely to be able to manufacture a lot of what you'll use during the war.

But simultaneously, it's not like the United States is really trying to fix this problem. The modern generation fighter aircraft are boondoggles with massively inefficient manufacturing designed to win senators and congresspeople votes - not act as the backbone of a hardened supply chain.

They're being built the way they are because the US doesn't seriously expect to face any enemy who can substantially deplete its forces, or is going to have any ability to strike the mainland.

My suspicion is that if there was an actual belief in the need to have reliable supply lines, it could be done and manufacturing could be streamlined a lot. You could definitely imagine a nearly fully automated factory rolling these things off very quickly.

> "My suspicion is that if there was an actual belief in the need to have reliable supply lines, it could be done and manufacturing could be streamlined a lot. You could definitely imagine a nearly fully automated factory rolling these things off very quickly."

Unlike previous wars, cruise missiles are a mature technology, making it much easier to precisely zap an opponent's manufacturing facility and infrastructure than before. High tech gizmos also have a lot more dependencies that go into them, making it easier to disrupt by knocking out a few or even just one of them.

> Some program that self learned a (admittedly impressive) number of reactions based on seeing/playing a huge number of simulations is not intelligence so not AI.

Is this really that different than what humans do?

After watching dozens of AlphaStar commentary videos [1] over the last few months, I was more or less thinking the same thing: the AI has basically evolved a massive ruleset: "Do X. If you see Y, do Z."

Nonetheless, I decided to start playing again myself. Poking around, I saw a recommendation to go through one guy's sort of "training course" [2], and guess what? A lot of it comes down to the same kind of thing. "Send your first overlord to scout their natural. If they haven't expanded to their natural, build one -- ONE -- spine crawler in your natural."

How much of our "intelligence" is really anything more than pattern matching + search? And during the actual dogfight, how much of what the human pilot was doing was anything more than simply pattern matching from their own vast experience racked up in a simulator?

[1] https://www.youtube.com/playlist?list=PLVRQoOk_ltE3Fr1ofRE0Y...

[2] https://www.youtube.com/playlist?list=PLFeZeom2b4Dlt63qmkPO8...

> Is this really that different than what humans do?

Yes, we have built in mechanism that react to situations we never encountered before. You will never be able to simply sit calmly while something is damaging you for example.

Reacting to damage is something you learned in previous generations. You've encountered that situation billions of times (you just don't remember them).

We are quite different as we can evaluate new situations based on our understanding of the principles behind something like a build order in Starcraft.

Clearly it can surpass peak human skill at StarCraft, but it’s important to understand what Deep Mind is doing differently.

One example is memory. Deep mind doesn’t update it’s strategy when playing the same player repeatedly or even over the same game. It can operate at near peak human performance indefinitely, but it avoids being exploited via deep understanding of the exact rules in play. It was also playing on the ladder under random names to avoid people developing specific counters.

> One example is memory. Deep mind doesn’t update it’s strategy when playing the same player repeatedly or even over the same game. It can operate at near peak human performance indefinitely but it avoids being exploited via deep understanding of the exact rules in play. It was also playing on the ladder under random names to avoid people developing specific counters.

This is because many games without perfect information have "rock-paper-scissor" situations. An AI designed without this in mind, lacking a source of randomness (or memory of the last encounter fed into the next game), is going to pick "rock" every time and start losing to the human who realizes and exploits this.

Imagination plays a significant role for us humans. We also are capable of maintaining incomplete conceptual mental models. Some of us experience epiphanies. Etc.

Pattern matching and search are brute force approximation of actual “intelligence”.

As far as I understand, AlphaZero had three main components:

1. A model of how Go worked, constructed by humans

2. A neural net for generating interesting moves, and a neural net for evaluating game positions.

3. A Markov search algorithm that used 1+2 to do pattern patching and search.

I will say that so far, #1 -- constructing mental models which can then be searched -- is something I haven't seen an AI do yet.

But looking at the actual work that I do -- from coding to deciding what to cook for dinner to deciding how to lay out my extension -- I'm pretty sure that behind the scenes it's exactly the same as 1+2+3 above; and that if I could download my mental model and my utility evaluator into a program, it could (with training) do all of the above faster and more consistently.

After AlphaZero, DeepMind went on to develop MuZero (https://en.wikipedia.org/wiki/MuZero, paper: https://arxiv.org/abs/1911.08265), which is able to learn to play games without being provided an explicit model of their rules. "When evaluated on Go, chess and shogi, without any knowledge of the game rules, MuZero matched the superhuman performance of the AlphaZero algorithm that was supplied with the game rules."

> A lot of it comes down to the same kind of thing. "Send your first overlord to scout their natural. If they haven't expanded to their natural, build one -- ONE -- spine crawler in your natural."

I think your general question is still a good one, but it's worth noting that from a human perspective, Starcraft matches only start off this way. Very quickly, the game state becomes complex enough that decision trees break down and intuition becomes the driving process for high-level human players.

The extent to which AI can begin to compete with this sort of intuitive human processing is most interesting to me. As it relates to this article, I think it matters a great deal if the experiment has constrained the system to the degree that it ceases to operate in that intuitive realm that high-level Starcraft matches operate in.

> Very quickly, the game state becomes complex enough that decision trees break down and intuition becomes the driving process for high-level human players. The extent to which AI can begin to compete with this sort of intuitive human processing is most interesting to me.

As far as I can tell, we're already there. Go is a game where search + explicit rules work really poorly, and intuition is a heavy factor in how humans play. AlphaGo was the first major success where neural nets were able to replicate this "intuition", and actually become far better at it than humans.

You can definitely see this "intuition" factor if you watch the AlphaStar games. Basically, if you have two armies with "middle of the road" units, AlphaStar always knows when to take a fight. You can sometimes see it take its army up to a defensive line, then turn around and go home; but if it decides to attack up that ramp, it always wins, even in a situation where a human would be less sure.

The exception is with unusual unit compositions; for example, it seems in training to have only encountered 1-2 battle cruisers at a time. So there are games where over and over again it throws an expensive army full-throttle into a group of 10 battle cruisers, and over and over gets slaughtered.

But again, I think this "intuition" factor, even in humans, is not much more than a sophisticated pattern matching, perhaps with background search.

> The exception is with unusual unit compositions

Could be wrong, but I believe this is actually a case-in-point moment for the other side. AI as it exists now is flummoxed by unfamiliar input, whereas a human expert is more likely to be able to react in a way that is at least directionally correct.

Put another way, if AlphaStar is confronted with a Terran opponent that can build Zerg units, how likely would it be to win the game? How likely would an expert human be to win that game?

No, what AlphaGo does should never be considered "intuition". It is simply using perfect information to optimize its probability of winning. We have no idea what intuition is or how it works. That's why we have learned literally nothing from AlphaGo. In fact, players who have played against it have often gotten worse. Intuition is very obviously much more than pattern matching. You need thought and probably some form of language to employ it, both of which we have very little understanding of.

The korean fight it beat human For many games afterwards.

Some insight gained like going to three three which is a taboo like old day playing three three alone.

Intuition or not ... learning we from computer go.

What the problem is they are perfectly known game. Air battle is not a perfect information game. That is why you need human decision.

> It is simply using perfect information to optimize its probability of winning.

Perfect information would be the knowledge of all possible outcomes of a given move. That's not even possible in chess, and is fantastically less possible in a game like go. That's why, until AlphaGo, there had never been a go computer program that had ever even come close to beating a professional go player.

Let me emphasize that: As of early 2015, nearly 20 years after Deep Blue beat the world champion Kasparov at chess, there had never been a computer go program that had come close to beating a professional go player. The game starts with literally 361 possible moves, and each move thereafter decreases the possible moves by exactly one. The search space is just so massive that it's impossible, by brute force computation, to do a search with any meaning.

What's needed is an intuition about two things. First, for a given a board position, which color is more likely to win? Secondly, given the hundreds of moves you could make, which ones are the best ones to explore? Go players have some rules, but primarily they develop an intuition for these two things by playing and looking at hundreds and hundreds of games.

AlphaGo's primary architecture was the same. It has two neural nets, which spit out board evaluations and move suggestions. These nets don't have explicitly coded rules, but were developed by playing millions of games. How is that any different than our wetware neural networks?

> In fact, players who have played against it have often gotten worse.

I hadn't heard this, but this is probably not unusual. Take my StarCraft example: Suppose someone had gotten to Diamond league mainly by perfecting their early rush strategies, but then hit a wall. Then they watch that video series which says the foundation of a GrandMaster strategy is having a solid economy first. If they decide to take this advice, it will entail a complete re-building of their skillset; they'll almost certainly drop in their rankings before picking back up again.

You could imagine the same thing happening in go: for hundreds of years, certain principles have been believed to be true. AlphaGo regularly violates these principles. As people are exploring this alternate strategies, they will inevitably get the "new" principles wrong while they're learning.

On the other hand, the early versions of AlphaStar clearly dominated mainly by having inhuman micro capabilities (ability to precisely control individual units). A human trying to replicate a strategy that relied on inhuman micro would inevitably fail.

Similarly, it might be that certain moves in go are good moves if you can do the kind of massively deep search that only AlphaGo can do. If that's the case, then of course humans trying to imitate AlphaGo are going to fail. A similar thing happened in chess: computers historically have been amazing at tactics and weak on strategy. A human playing a computer should focus on a good strategy, because they're never going to beat a computer at tactics.

But none of that changes the fact that what AlphaGo is doing, with two neural nets that spit out answers based not on explicit rules but based on millions of games worth of experience, is indistinguishable from human intuition.

I think you need to look up the definition of "perfect information games". It is not what yoU asume it to be, when you say "knowing all possible outcomes".

The “one” spine crawler thing is a bit overstated in the context of an actual game. It’s not like it’s 100% necessary to build it, build it in your natural, or only build one depending on what you’ve scouted, the opponent’s race, and who you’re playing against.

Like the sibling said, the game begins like a decision tree but quickly falls outside that purview.

Well you have to remember who he's talking to. You can sort of tell that lots of people, if they think they're about to be rushed, would make loads of spine crawlers as a defense; the emphasis on "ONE spine crawler" to delay an attacking army just long enough that you can build your own army in response (assuming you see it when it starts to cross the map) with the minimal impact on your economy.

And my point is that basically what he does in the series is start with a super simple set of rules, and then slowly add more subtleties and sub-clauses. That's obviously at some level how he works when he's actually playing -- he has a general plan, then a whole load of patterns that he matches on in how to respond. I'd be willing to bet that he rarely encounters a genuinely new situation for which there is absolutely zero precedent -- absolutely nothing in his experience which is similar enough for him to take an existing response and adapt it for that purpose. How is that different to what AlphaStar is doing?

The main difference is likely to be in how they learn afterwards. If he encounters a new situation in a game, afterwards he will use his model of the game to try to come up with new responses. AlphaStar will most likely have to do something much closer to brute force to find an appropriate response.

I'm not an expert but I think in the future you will have swarm of drones against another swarm of drones. No human involved (in the air at least).

The best algorithm will win the war.

People stop bullying your nerds at school if you don't want to loose your next war ;)

Yes, only thing is... usually it does not work that way in reality, and when they breech enemy defense drones, they will continue with a quick genocide.

> I think in the future you will have swarm of drones against another swarm of drones

Unless the swarms are easily exploitable like the current "AI"s.

> I'm not an expert but I think in the future you will have swarm of drones against another swarm of drones.

Unlikely. You will rather a swarm of drones against an enemy that has no such technology.

A country that can have a swarm of drones would very likely also have nuclear weapons and deterrence is a thing.

Canada isn't going to be building nuclear weapons ever again[0], but we're going to have drone swarms. They don't melt the face off of a bunch of civilians. They can selectively target. They're cheap.

[0] Most people don't know this but Canada is one of the few countries that built nukes (in conjunction with America and the UK) and then gave them up. I think we still technically own the seven nukes, but America is in possession of them.

IIRC Daech is using swarms of drones against the Russian air base in Syria (they probably don't have AI inside though).

I think AI powered swarm of drones is certainly feasible for a small team of motivated engineers and modest funding.

The thing that would let them at a disadvantage is the quality or lack thereof of their missiles guidance system.

Why do people talk about swarms of drones but never of swarms of fighter jets? That's incredibly weird terminology considering drones are never deployed in this manner.

> It cannot reason on the spot and will fall for the most ridiculous of traps.

> My theory is that AI will not exist until we reach AGI.

You are confusing the definition of AI for AGI so of course you think that. AI doesn’t need to have true understanding to be considered AI, it just has to have the appearance of intelligence.

>We've seen this before, even for a lot more limited and controlled environment like the game Dota the "AI" can beat the humans a few times before the humans learns to exploit it's many weaknesses.

I recall that there was AI beating top Dota players. I'm not familiar with humans figuring out the AI and exploiting weaknesses. Every search I've done just shows articles about the AI winning. By chance do you know where I can read up on humans figuring out the weaknesses of the AI?

https://openai.com/blog/more-on-dota-2/ mentions some of the strategies to beat the bots.

You can also check the ml subreddit for discussions of exploits:

https://www.reddit.com/r/MachineLearning/comments/bfq8v9/d_o...

https://www.reddit.com/r/MachineLearning/comments/6t58ks/n_o...

https://www.reddit.com/r/MachineLearning/comments/bcumrs/d_o...

https://www.reddit.com/r/MachineLearning/comments/6u304t/n_m...

I also recall a post from one of the pros talking about his strategies to try and beat them but I can't find it.

After AI beat 1v1 the top players they gave it to casuals at the international if I remember correctly: https://www.reddit.com/r/DotA2/comments/6t8qvs/openai_bots_w...

> The general strategy is to win by claiming first tower. At 0:00, you aggro the enemy creep wave so that they start following you. Then you walk around in a circle around the jungle, and the enemy wave will start to form a congo line that will follow you around. You then path around the jungle so that on the next wave spawn, you can aggro the wave again and continue to walk around in circles. The AI will burn glyph when your creep wave hits the tower, and for some reason it can't really decide between chasing you or defending the tower. So after about 5 minutes of doing this, your creep waves will eventually destroy the tower and you win the 1v1.

This reminds me of how 2b2t PvP has evolved. At first everyone was just wearing fully enchanted diamond armor and had golden apples that regenerate HP faster than you can lose them. Hacking is allowed on 2b2t. Because of dupe glitches getting large quantities of specific items is possible. Ender crystals that are normally used to resurrect the ender dragon can be smashed to create an explosion which kills players in full diamond armor instantly. Suddenly it was all about who could deploy his crystals first. Then totem duplication + auto equip allowed you to survive instant deaths until your inventory no longer contains totems.

No matter how impressive each of these techniques were, each of them had a counter that was highly effective.

I think there's value in having a huge learning set, but there are also limitations, as you're alluding to.

But I also don't think we're miles away from computers learning some sort of reasoning structure. Some sort of causality-type thinking where you have hierarchies which at some level are reasonably simple, because humans can only fit so much. But when the computers figure it out, they don't have that problem.

At the moment, yes, there's a huge corpus of patterns and you can make some smart decisions just by being able to learn from the huge library, but it's the difference between knowing that one move tends to beat another, without knowing the why. For instance in sports, you have man-marking vs zone-marking. The naive thing to do is just tabulate how often a team did one or the other vs how often they won. Then break it down even more by who they were facing and various stats. But if you don't have a theory of marking, you're a bit lost for explanation, even if the tabulation clearly says zone marking tends to win. A causal explanation might sound something like "man marking allows the other team to pull you out of shape and gives them the choice of which players face which". It might also tell you that sometimes it's actually smarter to man-mark, eg when there's some player you really feel is dominant and needs to be taken out.

I gather that people are working on this causality type AI though, so no doubt we'll see something interesting soon.

Without a clear definition of AI that everyone agrees on, we will never reach AI. If an AI is considered intelligent only if it is "General", then are we, as human, even intelligent? I would argue strongly that we are missing the 'G'.

> "Some program that self learned a (admittedly impressive) number of reactions based on seeing/playing a huge number of simulations is not intelligence so not AI."

Seems like with every breakthrough the goalpost of AI gets moved.

I believe a researcher coined a phrase for this but I can't remember what it was.

> I believe a researcher coined a phrase for this but I can't remember what it was.

There's an article on Wikipedia called "AI effect", I don't know if that's what you had in mind.

https://en.wikipedia.org /wiki/AI_effect

It doesn't get moved so much as we discover that a certain mathematical manipulation plus all our previous mathematical manipulations don't amount to an AI.

It also seems like with every breakthrough this complaint is raised without addressing the underlying issues with the new AIs shortcomings.

My hot take: true AI is so far out of reach of our ability it's not even funny, which makes the whole field either a search for the fountain of youth at worst, or at best a search for tools to inform humans or to replace humans in rote tasks. See https://youtu.be/orMtwOz6Db0

Ummm, no? AI means a clear thing: artificial _intelligence_. Depth-first search is not intelligence. Auto-generated algorithms that succeed on most of the training data is not intelligence. And that is obvious from the errors "AI" makes that a human would never do.

For example, would you fool even a child with this? https://cdn.mos.cms.futurecdn.net/s4DuKgTLnS4cngTyiqVkNC-970...

Someday, this line of argument is going to conclude that humans are not intelligent.

>Dota the "AI" can beat the humans a few times before the humans learns to exploit it's many weaknesses

Lets retry this experiment but the loser of any games get shot in the head, and the next player only gets basic telemetry while also shitting themselves. A human Dota player with a gun pointed at him will likely perform differently too.

Note that the AI player doesn't get killed if it loses, because it is software. It just doesn't get full telemetry and diagnostics.

Yes, they are all specialised AIs but we make them so fast nowadays that it doesn't matter they don't generalise yet, we can have as many as we want and they can be really useful in many fields. ImageNet can be trained in 30 seconds from scratch, and fine tuning takes just a few extra epochs. We don't need to wait for AGI, most low hanging fruit have not been picked yet.

Not sayin they are not useful. They are. But they are bots. Not AI.

OpenAI Five lost just 42 out of 7,257 matches. It's true that it has exploitable weaknesses, but it did not simply fall apart.

One of the major weaknesses of fighter aircraft is the pilot. Not their skills or abilities: the requirement for a soft, squishy human that deforms under G forces to pilot the thing is a disadvantage. The person in the cockpit is the most breakable part of the system. Everything else on the jet is capable of withstanding far more material strain.

Just being able to replace a human-driven fighter jet with one that is piloted by an AI - even a comparatively dumb one - would be an advantage for fighters. The AI driven jet would be more maneuverable straight out of the gate.

It's also by far the limiting resource on aircraft production. It takes 18 years to grow a human to the point where they can pilot a jet, and then a few years more to train them. Many years more to train them really well. In any sort of war of attrition, you'll run out of trained pilots long before you run out of planes.

The Japanese weren't defeated in the Pacific because they ran out of planes, they were defeated because they ran out of pilots. At the time of the Marianas Turkey Shoot, Japan still had a large carrier force (9) and a large number of aircraft (750), but their pilots were all green, and so weren't very effective in combat. The Battle of Santa Cruz Islands is considered a strategic victory for the U.S. even though it was a huge tactical defeat, because it depleted the stock of trained pilots enough that Japan wasn't able to mount effective resistance for the rest of the war.

The Battle of Santa Cruz Islands is considered a strategic victory for the U.S. even though it was a huge tactical defeat, because it depleted the stock of trained pilots enough that Japan wasn't able to mount effective resistance for the rest of the war.

Didn't know this. How sad. The more I learn about WWII and its abject brutality, the more I marvel at how much US culture fetishizes it.

Related: this season of Revisionist History has a series about Curtis LeMay and the history of napalm that is equal parts fascinating and sickening.

> equal parts fascinating and sickening

That's how I felt when I watched the documentary about Robert MacNamara [0], when they started talking about how we are pretty sure that what we did to Japan _before_ the atomic bombs counted as war crimes.

0: https://en.wikipedia.org/wiki/The_Fog_of_War

That's most of what the podcast covers. They quote MacNamara in The Fog of War as well.

Please do recall everything mentioned above pertaining to Japan was after https://en.wikipedia.org/wiki/Attack_on_Pearl_Harbor

When else would it be? I'm unsure what your point is…

All the horrible Pacific WW2 brutality by the US followed a surprise attack on its homeland intended to kneecap its military abilities so that Japan could invade countries around it unimpeded.

Doesn't make any of it right. Just putting it in context.

The US did some disproportionately heinous shit after Pearl Harbor. For example: 2,500 people died in the attacks on Pearl Harbor. The firebombing of Tokyo killed somewhere between 90,000 and 120,000 people - most of them civilians, and most either incinerated or asphyxiated by incendiary bombs over the course of six hours. It's the single most lethal day in human history - even greater than the atomic bombs, where over half the deaths were lingering affects of radiation over weeks, months, or years.

Curtis LeMay, the Allied Air commander in the Pacific, freely admitted that he would have been tried as a war criminal for the firebombing of Tokyo had the Japanese won.

Don't rush to justify everything the US did just because we weren't the first to strike.

There where also fuel and key supply shortages - just the same as Germany had late war

WWII was brutal. It was an existential war that had to be won, and we did win, and at great cost and tragedy. I do not think Americans fetishize WWII, but I think we do think of WWII with some pride, not for the evil that was committed, but for the triumph over a larger evil. It is too easy to look back and criticize the hard choices leaders made when faced with terrible situations.

There was another there major difference between the Battle of Santa Cruz Islands and the Marianas Turkey shoot. The US went from flying the vastly inferior Wildcat to the mostly superior Hellcat. Japan wasn't able to produce a more modern fighter than the Zero in significant numbers for the rest of the war and by 1944 it was obsolete.

Funny, I thought airforces generally just recruited pre-grown humans called adults.

The Japanese had excellent pilots, but they did not invest in the infrastructure of making more pilots. Every excellent pilot flew until he was killed. The Americans pulled top pilots back to train the next generation of pilots, creating a massive talent gap over time. The lesson is one that applies to startups as well as to aviators: you have to invest in your people, not just in maximizing value delivered today.

Many years ago there was an AI game for PC called Creatures, using neural nets for the Norns, the characters you would raise in the game by training them. The devs who wrote it were originally writing AI software for fighter pilots, and continually won simulated engagements because the AIs had no body-based limitations, which opened up a range of maneuvers that would literally kill a human.

The best example of such is a hard dive: if you're flying level, and dive too quickly, the G forces drive all the blood into your brain, potentially causing a hemhorrage. The maneuver human pilots use for a hard dive includes a half barrel roll just to avoid this, and it's still less effective than suddenly pointing down. The AIs used hard dives effectively to shed human pursuers.

This led to the insight that bodies (in some sense) were a missing component of any human-like AI, so they made Creatures to test the theory. The neural nets had virtual bodies with hunger, fatique, and pleasure/pain receptors; the player interacted with a god-like hand cursor that could pet them or spank them. For the 90s, it worked pretty well on Windows 98.

What a clever dual market strategy. Consumer PC games as failed fighter pilot software. Capitalism is incredible.

The story I told comes straight from articles about the game as it was released, and the marketing definitely worked on me. Raising cuddly neural nets seems much more interesting when you believe that the AI was already proven in another area.

Humans also have a limit on what they will do. A military with more AI can engage in actions that a group of soldiers would refuse to do. As more of the military becomes AI driven, it effectively concentrates more power behind those who operate the AI.

Alternatively, humans will also do things that the military doesn't want them to do: like loot, rape, and kill non-combats. Not every war is WWII where military leaders actively directs troops to commit atrocities. More often than not, crimes are committed by troops against the wishes of their superiors.

https://www.youtube.com/watch?v=n7wluszxDrY

On the bright side, horses are now spared much of the direct experience of modern warfare, so maybe humans will eventually follow.

On the flip side, if the oligarchy-with-social-policies was a knock-on effect of mass conscription during the napoleonic wars, a military with more AI suggests an oligarchy without social policies.

CN covid responders' celebration: https://www.youtube.com/watch?v=daVCbgNsVEM

UK covid responders' celebration: https://www.youtube.com/watch?v=4bQG9aB8dWY

(to be fair, I think the latter was celebrating 72 years of NHS, so the choice of a folkloric airframe was likely deliberate.)

Excuse me while I run to incorporate SkyNet Systems, LLC

That's a solved problem already really, basic training gives you soldiers who will do what they're told...

Wouldn't that be an argument for remote controlled fighter aircrafts?

Latency between the jet and the pilot is an insurmountable problem. You either have a very laggy connection to a human pilot who is far away, or a wireless link to a more local pilot who is vulnerable due to being in a combat zone. Plus, it creates an EM signal that could be disrupted as a tactical weakness.

The latency, though, would be the primary killer. It's why the Air Force still needs to send drone pilots to Afghanistan - you can have a person piloting the drone during the mission out of a container in Nevada, but they don't have enough reaction time to safely and reliably land the thing in Afghanistan. Control has to be handed off to a local pilot who has a lower-latency control link to the aircraft.

THank you for sharing that. I hadn't thought of that in that way before.

Interesting to note however that in the actual competition [1], the AIs didn't seem to find benefit from vastly exceeding the performance limits of humans, with regard to G-forces. Maybe these limits come through in the design of the jet as a whole, and would be a different story in a built-to-purpose UAS. The winning AIs here mostly benefited from super-human precision in aiming.

[1] https://www.youtube.com/watch?v=NzdhIA2S35w

Maybe these limits come through in the design of the jet as a whole, and would be a different story in a built-to-purpose UAS

That's what I'd suspect. A jet that's purpose built to be flown by a robot would likely be much more acrobatic, as you're designing for the engineering limits of the materials, rather than a human.

Regardless of the real answer, it's not making me feel like climbing into a jet cockpit and picking a fight with an AI.

Guys, the AI has perfect state information. While this is certainly impressive, it is not what you think it is.

This was one of the big advantages the AI had.

I could tell that the human pilot had to reorient himself after encounters, while the AI never needed to waste time grasping the situation and was able to take advantage immediately, every time.

In the same vein, as a former fighter pilot was commenting, the advantage goes to the pilot who can keep the plane right at the very edge of its performance envelope--which an AI can do more precisely than a human pilot can.

Classical dog fighting is all about energy management, something which an AI can outperform humans in considerably since it doesn’t uses intuition and experience but real time data.

With two identical airplanes, that's true. IIRC the AI also excelled in high-aspect head on shots which normal pilots tend to avoid, because for humans it's mostly about luck and less skill. When you pit two dissimilar aircraft against each other, it's up to the pilot to try to keep the fight outside of regimes where the enemy plane performs better.

A common comparison is between a one-circle and two circle fight, where the aircraft turn in opposite and same direction, respectively (and unintuitively!). Pure energy fighters like to keep going in a circle at the peak speed of its turn curve, when high-AoA capable planes want to cut into the enemy's flight path to get an early shot.

Another surprising example is a well-known HUD tape where a slow but fast-turning A-10 beats an F-16 in a training dogfight when the real-world solution would have been to not engage in a turning fight but hit the slower plane with a missile.

I can see the scary part being what happens when you network a bunch of these. At least until the EWAR guys figure out how to ruin it's day.

how about this: the scary part is when the individual nodes are rigged to be able to operate autonomously, and are able to elect new leaders or form consensus with peers and select new mission objectives as required by the situation if the prior command hierarchy is destroyed or otherwise uncontactable for an extended period

Uh huh.

A) Easier said than done.

B) How do you protect against malicious node injection and compromise?

C) Where's the off switch?

If you haven't solved the problem amongst carbon based life forms, you really need to think twice before jamming it into silicon. Boring? Yes. Less likely to end up boomeranging into your face? Most definitely.

Exactly, if you could replicate that perfect state information from real world sensors, then this become much more impressive.

Or introduce noise into the simulated state to reproduce real world sensor capabilities.

We know from John Boyd that getting into the opponent’s OODA (observe-orient-decide-act) loop is what wins these fights. If the AI has perfect state information, then it’s no wonder that it beat the human pilot who has to go through all four stages.

Having human pilots is going to play out the same way cavalry charges did in WW1. Old out of touch generals who romanticize how things where back in the day sending airmen to death needlessly.

The inability to change and adapt to new technologies plagues the US airforce and navy. Resulting in both of them being completely miss-equipped for what any war that doesn't involve fighting impoverished herdspeople would look like today. Only a fraction of both's budget is spent on systems that are actually good at anything other than showboating and being profit centers for mil-contractors. The majority of their budgets are captured by outdated manmetal (surface ships/pilot'd aircraft) most of which would be expended very quickly if a conflict ever did break out.

God forbid we ever have another conventional war but if we do you can look forward to having a couple dozen hypersonic glide vehicles (costing pennies on the dollar) sinking entire carrier groups in minutes. Swarms of drones just flying past over-priced f-35's, letting them expend all their munitions (which probably cost more than the drones), and still making it to hit their intended targets.

But the US navy's still got some kinda ok submarines so they got that going for them, the airforce.... idk I guess they can keep talking about "stealth" (fancy paint) and how its effective at avoiding detection from the underfunded military's of 3rd world dictators.

Militarized AI is nothing to celebrate. It should be banned or at least controlled with international agreements. 200K people died due to the atomic bomb, but AI weapons will not cause such apparent destruction. They will continue to be improved and integrated into the military and our culture. Because they will never be so scary to cause an outcry from the public or the experts, and politicians will not feel obliged to control them seriously.

Until it becomes completely accepted that flying robots kill human beings every other day... oh but wait. Already the case, never mind. It is okay I guess, as long as you are born in the country equipped with it.

I had an interview with Improbable in London, the CEO of their "Defense" research arm is a complete nutjob without any care for ethics or forward-thinking into the risk of what he is building. And that's a tech startup, I can't even imagine what's the mindset in the R&D labs of the old-school defence industry. Probably ridiculously entitled and self-congratulatory.

Does anyone have a good primer on (and I guess this is embarrassing to not know it) modern warfare and the statistics.

What I think I am looking for is a short Dungeons and Dragons style combat set of rules, and a blog post walking you through the issues.

I mean something like "Jet fighter A: can fire missile X with 50% chance of hitting other jet at range 200mi"

You not knowing in any level of detail is intentional. At best, the US military might have that under lock and key. These are more or less the most secretive projects a military has. If specs like "we sort of suck at x distance" were released, you give away a massive edge. Also, there is most definitely false info like that going around somewhere, so even if you find something, give that a lot of skepticism. Things like this are as much a war of intelligence as it is combat.

I remember a couple months ago, a video explaining the physics of a submarine being blown up showed up here. That was the first time I learned that a strategy for submarine warfare was to target an explosion under the center of the enemy sub, causing the water to rush up and creating a pressure differential that would crack the sub. Does that mean that is how sub warfare is done now? Or if that's the standard attack vector? Or that sub warfare would even be done with torpedoes now? No. The general public, and probably people not actively in the military with top secret clearance, only get bits and pieces of modern warfare strategy.

Edit: When I said the military might have those specs under lock and key, I meant my understanding is that the military is extremely compartmentalized and, aside from the highest ranking generals, there is almost no data aggregation on the details of what is capable.

I keep losing to lousy AI with my F-16C in DCS World all the time. :^)

No, seriously, this is sort of impressive. I don't like the taste of this being mostly relevant for war (read as: potentially killing people) but it's a technological advancement nevertheless. What exactly was the input for the AI, though? Just imaging or complete virtual situational awareness? Also from what I've taken from the video the challenge was to just point the nose vector at the contender as long as possible. Not sure if that is even relevant for a AA dogfight anymore these days.

Do you mind adding a bit of colour as to what losing looks like in DCS with an F-16?

I (Probably wrongly) thought modern air battles are just about getting within sparrow range and firing one off?

That's not fun, you just do dogfighting with cannons and/or shortrange IR-Missiles, like in that "AI" Sim.

Sparrows are history these days, but even more modern weapons can be defeated. It’s a lot about knowing what your weapon’s perf envelope is, and what the other side has to throw at you, and defending appropriately.

You could watch some of these DCS dogfighting videos, there is a mixture of guns only or short range missiles.

https://www.youtube.com/c/GrowlingSidewinder/videos

Here is one where the flightsim pilot dogfights a French Rafale pilot. https://www.youtube.com/watch?v=zgDZ4BLk0Pw

> "AI still has a long way to go before the Air Force pilots would be ready to hand over the stick to an artificial intelligence during combat"

That's making the brave assumption that the platform designer would wish to G-limit their craft to what ugly giant bags of mostly water can withstand.

https://drive.google.com/file/d/1uMWEEwzadeOFYHvjAKtMjXt_rwH... (page 96) and compare Michie's MENACE.

Inject pilot, win fight, catch pilot midair :)

Ah yes, the "loop zook", a standard air combat technique:

https://www.youtube.com/watch?v=k-wFI9vTqto

Very Epic. I remember this.

"With its weapons and defences under the immediate control of a positronic brain, it would be more manoeuvrable than any manned ship. With no room necessary for crewmen, for supplies, for water or air purifiers, it could carry more armour, more weapons and be more invulnerable than any ordinary ship. One ship with a positronic brain could defeat fleets of ordinary ships. Am I wrong?"

-- Elijah Baley, "The Naked Sun", by Isaac Asimov

PSA: The Campaign to Stop Killer Robots

Banning autonomous weapons is a very important step humanity has to take.

https://www.stopkillerrobots.org/

What happens if established liberal democracies get squeamish about killer robots, while rising states with alternative political models see them as a way to leapfrog existing disadvantageous colonial patterns?

What are your thoughts about land mines and use of chemical gas in warfare?

Here's a direct link to the point in the youtube video, where the fight between human (Banger) and AI (Heron) starts:

https://youtu.be/NzdhIA2S35w?t=16786

so they trained the AI at pointing the nose at the enemy, not at deflection shot? wonder why they didn't put this in a proper simulator.

also, the AI plane seems to have consistently turned in a tighter radius, which makes no sense in context.

The AI plane has a tighter turn radius because it's flying slower, which is part of the tactics.

Having said that, this simulation (like you said) seems to be assuming infinite bullet speed, and making that more realistic would change the tactics massively - probably in favour of the human.

No surprise. When making video game "AI", our challenge is not to figure out how to make a clever AI that can beat the player, but in making a system that can fail against the pathetic human in a believable way. Ironically, many of those failure modes involve having the AI do stupid human things, like forgetting to reload because stupid human brain/ears can't track exactly how many bullets have been fired just by listening to the BRRRRT of a machine gun. Or stepping out from cover at just the wrong time. Or calling to friends and thereby revealing position. Best one ever "I'm out!" or the variation "Reloading!"

Someone here said (paraphrasing) "but it only takes a few tries for the humans to learn how to beat an AI in games like Dota". In war, you're dead. There is no second try. Maybe, if they managed to get telemetry out, the survivors can analyze the behavior, but so are the winners.

There are also plenty of games where, no matter what, the humans do not win.

Once this is combined with vehicles that do not need to obey "don't squish the human" physics, and humans are done.

I'd be curious to know when the last actual read-world dog fight actually happened. Maybe there were one or two during the Gulf War? At any rate it seems to be going the way of the cavalry charge.

2019:

https://en.wikipedia.org/wiki/Dogfight#2019_Indo-Pakistan_ae...

Unless you're talking guns-only (the DARPA simulation discussed in this thread was guns-only).

> AI still has a long way to go before the Air Force pilots would be ready to hand over the stick to an artificial intelligence during combat, DARPA officials said.

It seems to me that rival states who also develop AI-enhanced systems for their respective fleets may do a different cost-benefit analysis and throw caution to the wind on that point.

A foreign fighter craft such as a MIG, traditionally with lower maneuverability when pitted against the likes of an F-22 (and similarly well-piloted), can eke out a significant equalizing advantage if flown by an advanced AI system; especially under the duress of protracted skirmishes.

And let's not forget the industrial cost of building AI-powered jets at scale vs. training highly-skilled human pilots en masse. That's already showing significant ROI in the form of Predator drones.

I think combat-ready AI fighter jets in the US will progress along a development roadmap that matches their growing need during acts of deterrence, rather than during full-scale war. But when multiple states have sufficient AI capability to go tit-for-tat without loss of life, then that would signal an ability to conduct more strategic "stress-tests" that carry with them a significant possibility of escalation.

Is the AI Pilot given perfect 3D state space information about the opponent? Or is it using sensors and vision to build a state?

In this test the AI pilots were given perfect state information. No doubt they will introduce a fog of war at some point, complicating the AI’s job considerably.

I believe they mention it having perfect state info. Really the human vs AI match is an effective publicity stunt, the real competition was among 8 different AI, so a more fair matchup.

This. Reality has to deal with a lot of imperfect information about the location of enemies.

There is a continuum - early spads, tossing bombs, shooting rifles - then machine guns (MG's) on axis, then gunner steered MG's, then axis rockets - then radar rockets - then AI rockets - then AI planes with AI rockets, then AI rockets with AI submunitions, and sub-sub munitions. Later AI's abd AI rockets capable of 100+ G turns. I see the fighter space fully non-survivable for manned fighters, with 100+ G opponents. Overlay this with mature jamming on 4 or more sublevels. I can see Tesla self driving AI packages, shaped to fit to be the teachable AI's that are NOW capable of running these shows. Same applies to high speed anti tank/truck/ship remote killers. High level management inertia in the Air Force hierarchy has been stalling these developments. You can bet Russia and China are on all this, so we must adopt ASAP while we have the electronics lead?

So, I’m curious what this means. Isn’t a dogfight with guns in F-16s kinda a completely implausible combat scenario in a modern world? I’m certain that air-to-air drones are going to be a big part of future defence, but I’m curious to what extent they need to be much more than flying missile launchers. Anyone in the industry have any thoughts?

It is unusual, but it does happen.

https://en.wikipedia.org/wiki/Air_engagements_of_the_Gulf_Wa...

There have been attempts to make fighters (F-4) without guns as that was considered obsolete by military middle management (pentagon), but then they had to add them later because actual combat showed it not to be the case.

In fairly recent engagements between F-15s and Mig-25s/29s - dogfights with guns have occurred and were decisive.

You don't always get to detect threats beyond visual range and shoot missiles.

For a more in-depth account to get an idea of how ROE (rules of engagement), and actual conditions play into these scenarios:

https://www.thedrive.com/the-war-zone/35765/confessions-of-a...

I kind of doubt 3 decades ago counts as modern war.

The F-22 - the top-of-the-line US fighter right now - first flew in 1997.

The youngest B-52s in service are well over 50 years old.

People forget that military tech changes slowly. B2 was 1980s, M1 1970s, Su-27 is early 1990's. Perhaps the newest aircraft would be the J-20.

Can you point to any gunfights with F-15s? I try to keep track of this (since the Bekaa Valley fights) and I was unaware of any gunfights.

It's like beating humans at chess, then Go. We don't desperately need AI to be able to do that specific thing. It's useful as a test - one of many surely to come - to see how far along we are in the process of replacement. You have to start somewhere.

It's a unit test.

stealth coating and low return shapes can't spoof eyes and flares and chaff can't spoof a bullet

Isn't it unfair to the human pilot, as the AI can pull G-Force maneuvers that'll otherwise kill the human pilot?

From the article, the pilot was sitting in a simulator. It’s also unfair to the pilot that he or she was not given all appropriate measures to avoid a gunslinger dogfight using long range weapons, teamwork, and stealthiness, but that wasn’t the point of the simulation.

Exactly. If two fighters are in a dogfight in 2020 it's because an entire network of systems has failed.

This exercise simply isn't reflective of modern tactics.

Then why do we train for it constantly in coalition exercises?

Sometimes entire networks of systems fail.

Because having the training is preferable to nothing in the same way that having poorly equipped, poorly trained fresh recruits manning your trenches is better than nothing.

Open warfare is rare. Most fighters patrolling near potential adversaries have complicated rules of engagement that give up their bvr advantages.

So you’re in your fancy fighter, patrolling a no fly zone, and you’re vectored to do a viz id on something that turns out to be a helicopter. Now you’re sweating it in case what could be an aid flight is actually going to fire an rpg at you as you pass, or if the whole thing is bait to bring you over aa or an ambush by some outclassed fighter that’s gonna magically arrive on your six...

When war starts someone always gets the jump on someone. Look at the strategic picture. That's one airframe. It's acceptable losses in exchange for not shooting at things you can't see prior to open hostilities.

Yes. He was also wearing an operationally unrealistic VR helmet. So, he wasn't flying a real platform for which he would have had good muscle memory for real flight controls.

This is actually a big deal. Very small differences between a sim and the real platform could be showstoppers for pilots (e.g. the contrived example of the 'fire' and 'radio' buttons being swapped in the sims vs. the real planes). Sims that are used for real combat training have to be exceptionally high-fidelity in all aspects that affect a given training task.

Wouldn’t it work that way IRL too?

Warfare is only sometimes about good sportsmanship. During the part where you’re trying to kill each other, it’s totally fair game to pull more G if you can. It’s also fair to use a better jet with better missiles of you have them.

Once (if) the other guy has ejected and is under canopy, then it’s important to be a good sport again- no shooting at him while he’s defenseless, and if you capture him then you have to treat him with dignity as a POW.

Beer missiles are the best!

Thanks for the catch. I edited my typo so now everyone will be super confused when they read your comment.

And if he loses an artifical leg during the ejection process you may let a replacement be parachuted in:

https://www.rafbf.org/news-and-blogs/operation-leg-%E2%80%93...

See, there are weird war stories that if we turned into a movie, people would think it's Hollywood embellishing.

They did make it into a movie, but not sure if it covers the parachuted legs incident: https://en.wikipedia.org/wiki/Reach_for_the_Sky

There is a movie about Bader: Reach for the Sky:

https://en.wikipedia.org/wiki/Reach_for_the_Sky

I can't remember if it includes the bit about parachuting in a replacement artificial leg... I can't imagine there was much desire to portray Hermann Göring in a positive way in the mid 1950s!

This is true. A Bridge Too Far both includes incidents that actually happened but were deemed unrealistic by critics, as well as omits certain details specifically because they would be assumed to be unrealistic.

The one movie I would love to see is one about the battle for Castle Itter toward the end of the war, where Americans, German soldiers who had defected to the resistance, a deserting SS officer, and a group of prisoners including two former French prime ministers, two French generals, Charles de Gaulle’s sister, and a pre-war French tennis champion all teamed up against an SS force attempting to recapture the castle and the VIP prisoners within. Which sounds like a completely absurd movie but it actually happened.

That's really cool. But I gotta wonder - why didn't the Germans just give the guy a new artificial leg instead of letting the Brits parachute drop one in? Seems like it'd be easier for everyone.

There may be specific fit issues for a new artificial leg. Most prosthesis require quite a bit of fit adjustment.

Those are usually custom fit to the individual and the Germans were probably happy not to take the effort of repeating all that work.

But what if the simulation is a Spielberg movie and if you let the AI go, later on he'll kill Tom Hanks?

It's like the Trolley Problem but weighted by how much you liked Joe Versus the Volcano.

Is that second part still a thing? I know ww2 got hairy, but as a regular civilian I don't know a ton of context about whether this was isolated or the start of a trend.

https://en.wikipedia.org/wiki/Laconia_incident

Some thoughts:

A) The "law of armed conflict" isn't a formal law, it's a collection of treaties recognized by most countries, and each country can have a slightly different interpretation. But most of the written stuff only goes back to the Geneva Conventions, and before that it was just tradition that kept everyone in line. Basically, if your side didn't follow tradition then there was a chance that you wouldn't be treated well if you got captured. So people mostly behaved themselves.

B) All (civilized) forces agree on the idea that it is unethical to attack non-combatants. There are grey areas, but everyone agrees that a pilot is no longer a combatant after his plane is disabled. If they land in your territory you can capture them and hold them prisoner, but you can't kill them, even if they were dropping bombs on you a few hours earlier. Similarly, if you sink a ship you are not supposed to kill the survivors.

C) Accidents happen and communication between enemies isn't totally reliable. Although you aren't supposed to kill people engaged in life-saving or rescue operations because they are non-combatants, what are you going to think if you see an enemy ship or plane approaching your defenseless friends? Sometimes people do bad stuff intentionally though, and hope that the lawyers can make a decent case if they get caught. Considering that evidence often gets destroyed and that dead men can't give testimony, they often get away with it.

D) A non-combatant can lose their protected status by engaging in combat. If a downed pilot pulls out his sidearm and starts shooting while you're trying to capture him, then you can shoot back. You can't shoot a clearly marked medic trying to save someone's life, but if that medic picks up a gun then you can shoot them. Paratroopers are practically defenseless while they are under canopy but they are considered combatants, so you can shoot them in the air.

E) AI probably won't be granted personhood until after they demonstrate sentience, so computer pilots probably won't get any of the courtesies discussed above.

Source: this was made very clear to to me at SERE school and is made clear to everyone at any SERE school regardless of branch. The exact phrase my instructor used was, "don't start bopping people on the head if you get caught behind enemy lines, but if it looks like the enemy isn't playing fair then you might as well get to bopping."

I saw the page and assumed the AI had boundaries on maneuvers that reflected the real-world boundaries on those aircraft.

Otherwise, the AI can easily win -- fly at light speed to behind the opponent, fight, fly at light speed away. The only thing to do with a simulation like this would be to put boundaries on the simulator.

The boundaries are on the human body, not on the AI. So it's more about speed and certain turns etc.

There are definitely also boundaries on any aircraft. There are speed limits, acceleration limits, etc.

The AI is just controlling the aircraft, so any simulation where the AI isnt bound by the aircraft's limitations would not be a good setup.

This isn't different from simulation based training of autonomous vehicles. Imagine if in an AV simulation, the AV controller could opt to accelerate 0-60 in 1 second, or if it could suddenly go 300mph to skip an up coming collision, etc.

Indeed. Reminds me of the Machine People and Hesperus in House of Suns discussing how they can push the limits of the spaceships that humans physically could not survive.

- House of Suns: https://en.wikipedia.org/wiki/House_of_Suns

All is fair in love and war.

This is 4 years old story repeated again> https://magazine.uc.edu/editors_picks/recent_features/alpha....

I believe in that 2016 test the AI had control over a UAV. In this test it was an F-16. While similar, the dynamics are not equal.

I watched this demo and noticed that the major advantage the AI pilot had was that it had near perfect gunnery. The human pilot had superior BFM and would have blown the AI out of the air if he had the reaction time for an instantaneous gun solution like the AI. A semi-autonomous mode where a half action trigger automatically maneuvers into a gun solution for the human pilot would have won him the match handily.

DARPA has been fighting the last war for ~30 years and this is no exception. AI will always have an advantage when all the variables a known and can be calculated, because if we're honest, modern air combat in a radar dense environment is pretty much just a series of math problems. (between 4th generation fighters anyway)

I'm all for seeing if AI can come up with solutions to tactical problems that pilots can't, and as a training tool this might be very useful, I just don't see much value in this particular coverage of the topic.

Air Force General Thompson interviewed Elon Musk at the Air Warfare Symposium this year. [1] It was, in a nutshell, a reality check for the Air Force's combat pilots, who were present in the audience, that they were no longer top dog, and had to either get on board with the new mission of drone warfare, or get off the train.

1. https://youtu.be/sp8smJFaKYE

Isn't this simply an issue of domain? Of course an AI will win in the simulator; have that AI drive a real plane against a real pilot in the real sky and see what happens.

Interesting competition. I just finished "Ghost Fleet" which is chock full of this tech. Eye-opening stuff when you look deeper. https://en.wikipedia.org/wiki/Ghost_Fleet_(novel)

> All eight teams will compete in each of the three scheduled trials. They will use the JSBSim open-source flight dynamics model for simulation environment and the open-source FlightGear Flight Simulator for the visual system. The simulated aircraft for the Trials is the publically available, unclassified FlightGear version of an F-15C fighter aircraft.

Interestingly they are using the FlightGear flight simulator.

I know I'm not supposed to detract from the main point, but something is a bit odd about this company:

A) It's classified as a Small Disadvantaged Business (SDB).

B) All its work seems to be for the government, and this is reinforced by its mission statement.

C) "The Federal Government’s goal is to award at least 5% of federal contracting dollars to SDBs."

I wonder to what extent it's valuable work vs just compliance with a kind of charity program?

Ah, WW2 style fighters are among my favorite games.

I don't think its realistic, but its still an advancement of AI. I'd see this as an advancement like "AlphaGo" or "Deep Blue" moment. But for a more obscure game with somewhat realistic implications (only somewhat, WW2 style fighters don't exist anymore in the modern world).

It’s important to note this is just within visual range dogfighting/BFM. Also just guns, and also perfect enemy state information. Important progress but it’s super early. Beyond visual range/BVR, coordination with wingmen and many other higher-level tasks still to come (or are still secret).

Watched the stream for this. Meat is not even close to competitive, and this with the AI limited to meat Gs and an aircraft that's designed around meat pilots. I feel sorry for the first meat pilots to find themselves "fighting" these systems.

I think the use of the term "Meat" highlights my greatest fear with robots fighting our wars for us, doing our killing for us. Our dismissal of our humanity, compassion and empathy. The warriors coming back from WW2 brought with them a deep desire to not do that anymore. The gamification of killing eachother. Calling ourselves "meat". In every other context, for thousands of years, we have been referring to ourselves as human beings, people, folks, neighbors, etc. Now at the very beginning of this technology we are throwing all that out and going with "meat". Contempt for the amazing gift that humanity is, thats what using "meat" to describe ourselves indicates.

They're made out of meat https://www.youtube.com/watch?v=7tScAyNaRdQ

By the way, I agree with you, it is unsettling to hear human beings reduced to such terms.

I'm pretty sure humans will always value human life - on their own team. The greater danger is dehumanizing enemy humans which is a common technique to motivate soldiers to kill each other.

> [...The AI exhibited “superhuman aiming ability” during the simulation, Mock said...]

> [...Somewhere, the infamous Red Baron is no doubt laughing in amazement...]

The infamous Red Baron would likely ask what does AI pilot use for eyes?

In a classic dogfight, spotting your enemy before he sees you is an enormous advantage. Loosing the sight while in pursuit often equates to loosing the fight (thus going to sun or clouds for cover).

If AI was strapped into simulator's digital feed for exact positioning (basically as if manning a game's AI craft), then it had "God's eyes".

This kind of superadvantage may be warranted in a game to compensate for usually otherwise inferior in-game AI abilities.

However, when plugging a whole computer rack for AI, this sight-advantage would definitely create a biased environment (for the purposes of meaningful comparison).

I wonder how would the AI fare if it relied on CV for its positioning?

Air Combat will be very different in the future. I don't believe dog fighting will even be a thing.

AI, Stealth Technology, Hyper Sonic platforms, Drone Swarms, Satellite Warfare, etc seems to be the way things are going.

This is a slippery slope. For all the failings of humans, most air-to-air conflict has been within visual range. That's because we don't trust sensors/intel/decision-making/airspace beyond that range to not blow up the wrong thing. Airspace is filled with lots of non-fighters filled with passengers.

Once we (and they) go to unmanned, then the latency of decisions will require less up-close fighting and much longer range engagements. Acro performance won't even be a thing. Cheaper, more efficient platforms will result in more platforms carrying more missiles shooting in much bigger hairballs.

I'll take old-style few elite knights on horses swinging at each other over the hordes of drone infantry decimating a country-side.

We continue to rush headlong into autonomous robot warfare without any consideration of the consequences. Humanity continues to disappoint.

352 comments and counting. So sad to see so many talented and smart people burn their energy discussing war machines.

On the other hand... if the AI can't even out perform in a simulation, you kind of have to worry about the model.

Meatbags are G-limited. There is an immediate advantage to maneuverability when you remove the human.

I'm ready for a remake of that 2005 film "Stealth" that features an AI pilot.

Macross Plus would be better

"The Feeling of Power" by Asimov feels like an appropriate mention here.

Or perhaps "Malak" by Peter Watts.

And how likely is this to be needed?

Perhaps it would be better to spend the money elsewhere.

The USAF needs this to get out of the F-35 cost problem. The F-35 costs so much they can't afford enough. So they have to figure out ways to do more with fewer manned aircraft. (Trying to get rid of the A-10 was the last bad idea in that direction.)

Did they use Microsoft Flight Simulator?

Military robots like drones are ironic because they are created essentially to force humans to work like robots in an industrialized social order. Why not just create industrial robots to do the work instead?

Nuclear weapons are ironic because they are about using space age systems to fight over oil and land. Why not just use advanced materials as found in nuclear missiles to make renewable energy sources (like windmills or solar panels) to replace oil, or why not use rocketry to move into space by building space habitats for more land?

Biological weapons like genetically-engineered plagues are ironic because they are about using advanced life-altering biotechnology to fight over which old-fashioned humans get to occupy the planet. Why not just use advanced biotech to let people pick their skin color, or to create living arkologies and agricultural abundance for everyone everywhere?

These militaristic socio-economic ironies would be hilarious if they were not so deadly serious. ...

Likewise, even United States three-letter agencies like the NSA and the CIA, as well as their foreign counterparts, are becoming ironic institutions in many ways. Despite probably having more computing power per square foot than any other place in the world, they seem not to have thought much about the implications of all that computer power and organized information to transform the world into a place of abundance for all. Cheap computing makes possible just about cheap everything else, as does the ability to make better designs through shared computing. ...

There is a fundamental mismatch between 21st century reality and 20th century security thinking. Those "security" agencies are using those tools of abundance, cooperation, and sharing mainly from a mindset of scarcity, competition, and secrecy. Given the power of 21st century technology as an amplifier (including as weapons of mass destruction), a scarcity-based approach to using such technology ultimately is just making us all insecure. Such powerful technologies of abundance, designed, organized, and used from a mindset of scarcity could well ironically doom us all whether through military robots, nukes, plagues, propaganda, or whatever else... Or alternatively, as Bucky Fuller and others have suggested, we could use such technologies to build a world that is abundant and secure for all.

So, while in the past, we had "nothing to fear but fear itself", the thing to fear these days is ironically ... irony. :-) ...

====

The above is from: "Recognizing irony is key to transcending militarism" -- an essay I wrote in 2010: https://pdfernhout.net/recognizing-irony-is-a-key-to-transce...

If Hollywood has taught us anything, its that this is not a good idea.

But, its interesting to think about the future of war when all combatants are automated. Is it purely a matter of resource-superiority? Or will algorithms be the main differentiator? And what about all the countries who are not even in the automated race.

Sometimes people need to be reminded about the existence of CBRN. Maybe you can add drones to that, but it doesn’t change the fact that a single attack has been able to completely anhillate an opponent with no possibility of defense for decades now.

If drones are fighting each other, it’s called entertainment, maybe propaganda, or maybe just imperial business, but not war. This sort of sporting match as a proxy for total war is common throughout history.

Reply on news.ycombinator.com