Hacker News Reader

Comment by onepointsixC | Hacker News Reader

For those who had not watched the AlphaDogfight competition, the most remarkable thing was not that the AI beat the Human fighter pilot. The most remarkable thing was the extreme gap in performance between the best AI and second and third place AI's. It wasn't even close. An AI that is marginally worse could still be able to beat a human but will lose consistently over and over against a better AI.

This bodes very poorly for any nation that falls behind in the AI race. As well as any nation who's best and brightest minds refuse to contribute while their adversaries have no such qualms.

View on HN

This bodes very poorly for any nation that falls behind in the AI race.

Not really. The chance of any sort of aerial dogfighting happening in a war are next to zero. The cost of planes, and pilots, is so high that almost all the effort spent these days is on killing your enemy with over-the-horizon weapons fired from hundreds of miles away. AI-powered dogfighting in aircraft is military marketing.

A nation could develop world-class AI planes and they'd be absolutely useless against a hypersonic missile fired from mainland China or a Russian submarine, nor would they be any use against a guerilla army in Afghanistan or Iraq that doesn't have the capability to fight in the air or that has 30 year old planes that can be shot down with 50 year old tech. War has changed; making better versions of old weapons is a waste of time.

This is true. But the barrier to using over-the-horizon weapons like icbm / hypersonic is much greater than say drone attacks or fighter jet bombs. Have over-the-horizon weapons ever been used? No. But drone attacks and fighter jet activities are daily occurrences because they can be used with little repercussions.

Over-the-horizon weapons absolutely have been used. Frequently, repeatedly, and (allegedly) effectively. Tomahawks count, unless they're retrospectively reclassified as drones.

> A nation could develop world-class AI planes...

such a nation would probably already have nuclear weapons. Any nation capable of hypersonic skip glide weapons also probably has nukes. So this hypothetical attack wouldn't be happening. At least w/ non-nuclear weapons.

Can you extrapolate on this? In my understanding in recent conflicts (e.g. Armenia and Azerbaijan) and currently in Ethiopia, drones (okay, we're not talking about fighter jets here but still, autonomous or near autonomous aerial systems) have played critical roles in eliminating air defenses, heavy weaponry, etc. and in future conflicts now that parties on all sides know how important the use of such drones are, will be deploying their own drones as countermeasures.

Not the guy you are asking, and won't cite due to time constraints, but here is the tldr. In Vietnam, one nation had a dream. It dreamt of MIG-21s falling out of the sky before they even saw the US fighter. That fighter was the F-4 and it did not get a cannon. It did not get a cannon, because its missiles were so advanced that it did not need one - it would spot the MIG-21/19 and shoot it out of the sky before the lighter, more agile plane (with an inferior radar) got close. Except that didn't happen. Some blame lack of approval-to-fire on target that isn't visually identified, but the reality is that the tech wasn't there.

Now it is. Most fighter-to-fighter kills now are radar beyond-visual-range "BVR" kills. Heavier, less maneuverable fighters with more speed, altitude, and radar range can take out lighter ones. The SU-27 can take out the MIG-29. The F-15 can take out the F-16, etc.

Now drones. Drones are mainly a joke when the big boys come to play. The aforementioned SU-27 and F-15 CAN detect and CAN take out most drones. Things like Bayraktars are great against insurgents. Not so much against an advanced military. Note that most drones in service now are used as spotters and ground attack. They are VERY slow, but are high-altitude.

And, no one ruled out EW when it comes to large-scale conflict. Ukraine, 2008 Georgia, Arm/Azer don't count.

edit: For the guy below talking about ICBMs - no one is talking ICBMs. Google AIM-120 AMRAAM and AESA radar. Not trolling, just faster to tell you what to Google than try to summarize.

Except a country with shitload of drones will send 100 of them for a cost of 1 F35 from field runways no satellite can find, and bomb your airstrips.

Even if you react in time, and shoot down a dozen, result doesn't change.

Current F-35A unit cost is $78M. Which model of drone costs $780K and has the same range?

In reality cruise missiles will be used for most strikes against fixed targets like runways. Even those cost like $2M each.

> Which model of drone costs $780K and has the same range?

None, but you don't need range if you will creep within few hundred kms of the target, and launch from field.

Well we already have that. Manned aircraft have been used to deliver cruise missiles for decades. Cruise missiles are drones by another name.

It's not quite the same. Calling a cruise missile an unmanned bomber does not match the effect.

A cruise missile gives you one big bang, enough to ruin a runway. Drones on the other end can keep buzzing in the sky for hours, and destroying whomever still tries to take off, or repair the runway.

The F-4 story is a bit more nuanced than that.

The issue wasn't that the F-4 didn't have a cannon, but that it was assumed that homing missiles would make teaching the art of dogfighting obsolete. This was proven to be a disastrous mistake. However once SFTI program (AKA TOPGUN) was established, the now much better trained American pilots had established a clear lead in air to air victories.

We know that it wasn't the lack of a gun because a gun pod was developed for the F-4. The ratios of victories to losses of the gun pod equipped F-4's was worse than those without.

For anyone reading this later who doesn't want to dig into it, I just want to confirm that everything onepointsix said is correct. Thanks for the details!

I mostly agree, but aim-120s are hundreds of thousands of dollars each, and each plane only carries about ten. toe-to-toe, a drone has no chance against a modern fighter, but fighters are an extremely inefficient counter to a fleet of cheap drones.

of course, how to get the swarm of cheap drones to a distant target is also an unsolved problem.

I think you're taking the example of dogfighting a little too literal when it can be easily applicable in all sorts of different ways.

Let's take your Hypersonic missiles. The largest difficulty with intercepting such weapons is not their speed, they travel slower than traditional ballistic missiles, but their terminal maneuvering which tries to dodge any interceptors. AI powered interceptors could be a game changer in improving the likelihood of those extremely expensive and limited quantity stand off weapons from reaching their intended targets. And having a slight edge in your interceptor AI vs say the terminal guidance AI could produce overwhelming advantages.

> War has changed; making better versions of old weapons is a waste of time.

This has been declared a great many times and yet has been also disproven just as many. Just because a M1A1 Abrams tanks and F-117 Nighthawks Stealth Bombers wouldn't have won the Vietnam War doesn't mean they weren't a significant contributors to winning the Gulf War.

For better or worse airpower will continue to be the deliver method of most warheads onto foreheads. The hypersonic's which you suggest that are making them obsolete are very costly and very limited in total payload capacity compared to the amount which can be delivered by air. There's a reason why Russian and China are not discontinuing investments into new fighter jets and bomber programs just because they have hypersonics.

How is winning a war against a decrepit nation's badly run military proof of any weapons system's superiority?

Your belief that the Iraqi armed forces of 1991 were decrepit and badly run is the proof of just how superior those weapons were, along with the accompanying tactics and doctrine. Before the conflict the overwhelming victory for the Coalition of Willing was not so clear.

The Iraqi army was battle hardened and tested after a near decade long war against Iran. They had the third largest army in the world, which was trained in Soviet doctrine, armed with potent and modern weapon systems. Their integrated air defense system was state of the art combination of the most modern Soviet and European systems that were redundantly layered and hardened.

America, along with its allies, fought them and won in spite of being geographically on the other side of the planet. The exceptional difficulty of achieving that is impossible over state.

> Their integrated air defense system was state of the art combination of the most modern Soviet and European systems that were redundantly layered and hardened.

Not quite: for example, see [0]:

"Though impressive on paper, the Iraqi Air Force's primary role was to act as a regional deterrent, with a secondary role of supporting the Iraqi Army, rather than attempt to gain air superiority in any conflict. Basic training was rigid, inflexible, and left pilots with extremely poor situational awareness.... Like its aircraft, much of Iraq's ground air defenses were also outdated: SA-2 and SA-3 systems were nearing the end of their operational lifespan and their countermeasures were well-known at this point, while its other SAM systems were not much younger."

[0] https://en.wikipedia.org/wiki/Gulf_War_air_campaign#Iraqi_Ar...

I'm not talking about their missiles or air force, but the Integrated Air Defense System.

The French built KARI IADS system was a very much modern at the time, having been first operationalized in 1987 and was linked with fiber optic cable, with wireless communication as a backup [0]. While the Iraqi's did not have the latest SAM systems themselves, their integrated air defense system allowed them to keep the radars of SAM batteries turned off, keeping them hidden to avoid detection. Their missiles were very potent, and it had a lot of them at 16,000 in total [1]:

"The U.S. Government estimated before the war that the most critical Iraqi targets were more heavily defended than any in Eastern Europe even during the height of the Cold War."

S-125 missiles had victories against coalition aircraft. They would have had many more victories were it not for the fact that coalition air strikes severed the centralized air command communications in the starting minutes of the war thanks to the Stealthy F-117's which were already above Baghdad at the time of H hour. The combination of stealth aircraft taking out key crucial infrastructure, severing of fiber optics communication, jamming of wireless communications, electronic warfare, and decoy drones proved to be too much for the Iraqi IAD's. But without those technologies and strategies, the Gulf War could have been a blood bath much more akin to Vietnam.

[0]: https://media.defense.gov/2010/Sep/27/2001329817/-1/-1/0/AFD... [1]: https://www.rand.org/content/dam/rand/pubs/papers/2009/P7837...

>> The Iraqi army was battle hardened and tested after a near decade long war against Iran.

Which means absolutely nothing if you cannot stop the airplane from dropping bombs on your critical logistics nodes. Those battle-hardened soldiers weren't so hard after a few days without water rations, without fuel in their tanks. In modern maneuver warfare between nation states "toughness" means far less than even a slight technological advantage.

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

>Which means absolutely nothing if you cannot stop the airplane from dropping bombs on your critical logistics nodes.

Which was only possible thanks to the superiority of American weapons system's which were better versions of old weapons. Thank you for reinforcing my point. Also the major coalition action was over after 3 days of fighting, that wasn't long enough for any Iraqi divisions to run out of supplies.

At the time of the first Gulf War, Iraq had the world's fourth largest army. Some details here:

https://nuke.fas.org/guide/iraq/agency/army.htm

Not decrepit, IMHO.

It was also widely believed that America's "volunteer force" was a disadvantage.

The USA did many firsts. It was the first deployment of M1 Abrams, no one knew if it'd be good (and IIRC, it was considered a waste of money back then, since it wasn't battle proven yet). It was the first time the US used GPS on the battlefield. It was the first time precision weapons were used. Etc. etc.

A lot of people at the time thought the USA could lose Desert Storm. It turns out that war happened completely differently than armchair generals / pundits believed.

I don't know if we'll necessarily win or lose our next war. But I think I can rather confidently say that none of us will be able to predict it.

The US won and then lost Desert Storm the same way it won and then lost Vietnam. Anyone who was saying that Saddam would be able to repel the US was engaging in WWE-style theatrics, but anyone who thought it'd turn into an unending mess would have been right (but was anyone saying that?)

The US initially won Vietnam?

The US was "winning," defined narrowly in terms of military statistics, throughout almost the entire war - casualty ratios were vastly in America's favor, and every military target that was selected was taken and destroyed. Bombers flew with nigh impunity.

That carried public opinion until 1968 when business interests started realizing that none of those things corresponded to anything good. Very similar to Iraq and Afganistan where unquestioned military dominance led to the destruction of the ruling order which led to ???.

>> AI powered interceptors could be a game changer

Nope. AI wouldn't make any difference. The problem in shooting down an inbound maneuvering hypersonic isn't reaction time. It doesn't matter how quickly you see that the target has changed course. Hypersonics are a matter of commitment. Once the interceptor is launched it is committed to very narrow cone of potential trajectories. If the inbound hypersonic makes a slight turn, it is then outside that cone. The interceptor cannot physically alter its course enough to hit the new trajectory. It lacks the energy (delta-V) to make such a change. AI cannot create delta-V out of thin air.

A computer can spot a track, project its likely path and guide an interceptor missile. But that isn't AI. Computers in the 1950s were capable of such things before any notion of what we now call AI.

I don't understand, why can't an interceptor missile change its course but a hypersonic missile can?

Relative speeds. A slight turn by the inbound hypersonic will could move the impact/intercept point several hundred miles left or right. If you are the ground-launch interceptor that shift left or right requires a much larger turn, a turn requiring much greater energy on your part than the hypersonic spent making the smaller turn. Say the inbound hypersonic just turned a few degrees east. The interceptor fired towards the old western trajectory might now have to turn around and head east, meaning it no longer has enough energy to get to the new intercept point.

This doesn’t make intuitive sense.

It’s all in how fast the interceptor updates to reflect the new trajectory. If it were magically simultaneous and their mass were the same, the energy would be the same, no?

Or is this something I am lacking findamentL knowledge on.

The missile is faster and travels a greater distance, and the interceptor is smaller and slower. That's because every target needs to have interceptors nearby, but you only need one missile.

Why are interceptors always slower

Could an algorithm determine the possible evasion changes and provide full coverage with n interceptors per missle?

What if the missile just turns away and comes back?

Sure, if you have unlimited budget to build those interceptors and the associated launch platforms. These are extremely complex and expensive weapons which no country is able to mass produce.

Laser on the interceptor?...

edit: might not matter against a thing designed to fly in intense heat

Because, in this case, the incoming weapon dictates the fight. If you are chasing something down (think air-to-air missile) then every turn by the target allows the missile to catch up faster via a shorter path and smaller turn (a2+b2=c2, the missile follows c). A 45 degree turn by a fighter jet might mean only a few degrees turned by the sidewinder missile fired from behind.

But with a hypersonic coming towards and the interceptor being slower, the opposite of a classic dogfight, the opposite becomes true. A small turn by the hypersonic means a larger turn by the intercepting missile to get to the new intercept point. If you draw the triangle then the intercept path is C, but the shorter of A or B. The hypersonic is following the longer of A or B.

Are interceptors always slower and why?

For hypersonic weapons, yes. Firstly, the interceptor has to climb up to the altitude of it's target. Then, it has to fly through very thick atmosphere. It also has to do so rapidly enough for it not to be too late. It needs to be much cheaper than the weapon it's intercepting for it to make economic sense, also. Because of this, they won't be airbreathing, and even if they were they would be slower.

I can imagine interceptors fired off near the enemy launch site that run down the hypersonic while it's still accelerating. Or, interceptors fired down from space either at the incoming hypersonic or along an expected path intending to converge.

There's two problems with your first idea. The primary issue is that such an interceptor would need to be more expensive and more capable than the missile it's gonna hit, not even taking into account all of the supporting infrastructure necessary. The second issue being that if you're going to be operating near (horizon distance, so let's say 200km) of a missile launch site, you're going to be bombed or missile'd to smithereens. If you can prevent that, you could just bomb the launch site to begin with.

As for interceptors fired down from space, you'd need to expend all the energy to get them into orbit, and then expend the energy to get them back into atmospheric flight, which itself is going to be 10x more expensive than it's target. There's also the issue that while in suborbital flight, the hypersonic missile can just adjust it's trajectory a bit and the whole trajectory is now useless.

>It needs to be much cheaper than the weapon it's intercepting for it to make economic sense, also.

Certainly it would be nice if it were cheaper than the incoming missile, but doesn't it really just have to be cheaper than whatever the missile is targeting for it to make economic sense?

One is falling down from space, the other climbing up. Imagine if I threw a brick off a building. You are on the street and see that the brick is going to hit one of two babies. You might be able to jump in front of the brick in time to collide/intercept the brick, but there is no world where you would be moving upwards anywhere as fast as an object falling from great height. Now imagine the brick can move to avoid you and instead hit the other baby. The brick changes course after you jump. That's what is happening with a hypersonic diving onto a target.

> A 45 degree turn by a fighter jet might mean only a few degrees turned by the sidewinder missile fired from behind.

This is true for some "classic" guidance laws (such as proportional guidance), but is it true for optimal guidance? Especially for point defense? Your "small vs. large turn" becomes an issue of transverse acceleration, and that makes the advantage of the hypersonic vehicle way lower, if not non-existent (since to my knowledge hypersonic lifting bodies don't exactly shine in the lift-to-drag ratio department, so it's far from clear that an interceptor would be unable to match transverse acceleration, and in any case it might be able to force enough velocity loss on the hypersonic vehicle that it may have to work way harder to hit the target even if it's not actually hit by the interceptor at all).

Because the hypersonic missile has an air breathing engine and/or goes much faster, so it has a lot of energy to expend.

Also, a very slight change in trajectory by a missile far away requires a much larger change in trajectory for the interceptor, because the interceptor has to go to where the missile is projected to be, meaning that the interceptor is always at a disadvantage.

Many of the hypersonic weapons currently under development use a boost-glide design and won't have an air breathing engine running during the terminal phase.

https://www.scientificamerican.com/article/the-physics-and-h...

It seems that some think it’s possible:

https://breakingdefense.com/2021/11/raytheon-northrop-lockhe...

Intercept is very possible, it just doesn't depend on AI. It depends on rocket equations and potential energy. The calculations/decisions are very simple. It is a rocketry problem rather than an information problem.

I’m guessing the ai helps optimize chance of interceptions, coordinating with other missiles/aegis to zone off areas or be at the highest likelihood place at the right time.

Advanced algorithms can be helpful in allocating interceptors and launching platforms, and then deciding on optimal salvo size and launch time. But what we think of as "AI" doesn't necessarily add anything there. Simpler deterministic algorithms are usually superior because they're easier to test and less likely to fail in unexpected ways.

But you are assuming an AI fighter plane is the same as a human-powered plane. If there is no need for a pilot, you can probably design the fighter to be something closer to a missle. It will pull harder Gs, have smaller radar profiles, etc.

> Not really. The chance of any sort of aerial dogfighting happening in a war are next to zero. The cost of planes, and pilots, is so high that almost all the effort spent these days is on killing your enemy with over-the-horizon weapons fired from hundreds of miles away. AI-powered dogfighting in aircraft is military marketing.

AI would dramatically reduce the cost of planes and it would eliminate the cost of pilots altogether. If you don't need a pilot then the economics of building planes completely change--you can throw lots of cheaper planes at the enemy and accept that a certain number of them won't come back.

I think that will also change the psychology of AI sorties. If a cheap Russian drone downs a cheap US drone, it will probably feel more like "Russians believed to be behind US government hack" than "Russians shot down a US fighter plane". The latter is a lot more of a casus belli than the former.

No the construction economics are still mostly the same. UCAVs with the range and capacity to replace manned tactical aircraft are only marginally cheaper. MQ-25 unit cost is in the $150M range. No country can afford to build a lot of them.

The small, cheap "swarming" drones have very limited range and endurance. Useful in some limited scenarios but only if you can first find a way to deliver them to the target area

I posit that the costs will continue to decrease as mindshifts change--right now (or more precisely, around the time these drones were designed and purchased) we think about drones as being "unmanned planes" and planes are supposed to be big, reliable, and expensive and the military procedures and culture is calibrated to minimize losses of these planes. As expectations change, I think costs will decline as well which will feed back into a change of expectations (time will tell).

Regarding "delivering them to the target area", it's pretty easy to conceive of a larger transport craft managing delivery.

What's your point? We already have that. Large bombers are used to deliver cruise missiles close enough to the target area. Cruise missiles are drones by another name. But they're hardly cheap. In fact they're getting more expensive.

- build small, short range but very smart swarm drones

- package them up (cluster bomb style) into a "dumb" ballistic missile that only needs to get to the general area (e.g. near a city)

- airborne deploy the swarm once near the target

Combines long range "cheap" deployment with short range smart capabilities.

There have been proposals to build conventionally armed ballistic missiles to fulfill the prompt global strike mission. The problem is that if Russia / China / North Korea detect a ballistic missile headed in their general direction they might misinterpret it as an incoming nuclear first strike and retaliate accordingly. A very dangerous game.

Also it's rather pointless to deploy any sort of "drone" from a long range ballistic missile. It would need a complex and heavy braking system to slow down enough during re-entry to allow for controlled flight. Easier to just hit the target directly with a re-entry vehicle.

Have some stupid CGI: https://youtu.be/n8u0l_c2_NA

The point is that AI has the power to change the ecology of war. Cruise missiles occupy a niche, and yet they haven't obsoleted combat aircraft, so clearly they aren't "drones by another name"--drones serve a purpose today and will continue to diversify (including a blurring of lines between "drone" and "cruise missile"). Anyway, this conversation has taken a defensive tone, so I'll duck out.

You would need sufficient hypersonic missiles to knock out the offensive capability of the target.

People have been claiming that over-the-horizon missiles will obsolete dog-fighting since before the Vietnam war:

https://en.wikipedia.org/wiki/Douglas_F6D_Missileer

> Through the later part of the 1950s and into the 1960s, military air planners increasingly believed that future air combat would be carried out almost entirely by long-range missile fire. This changed the basic requirements for a fighter design considerably. The pilots would be expected to fight primarily through their radar and fire control systems, hopefully never even seeing their opponent. Because of this, the emphasis was on "head down" combat and an all-round view was considered unimportant....

Not saying that won't eventually be true, but it doesn't mean it will be true any time soon. The lesson is to preserve the capability until it's actually been proven to no longer be needed.

If you look at the actual data from the past couple decades, far more aircraft were destroyed by SAMs or cruise missiles (on the ground) than in BVR or WVR combat. Dogfights haven't been a significant factor for a long time.

> If you look at the actual data from the past couple decades, far more aircraft were destroyed by SAMs or cruise missiles (on the ground) than in BVR or WVR combat. Dogfights haven't been a significant factor for a long time.

Data can only tell you so much. For instance, according to the data, nuclear weapons haven't destroyed any enemy equipment in the last 80 years, but you'd be going too far to infer from that conventional weapons have made them militarily obsolete.

The difference is that nobody is avoiding WVR dogfights just so they can get shot down by a missile in BVR or a SAM. Nobody is getting within range of a dogfight nowadays because these other weapons are so extraordinarily dominant.

> The lesson is to preserve the capability until it's actually been proven to no longer be needed.

And yet when faced with evidence of this, your immediate response is "data can only tell you so much".

>> The lesson is to preserve the capability until it's actually been proven to no longer be needed.

> And yet when faced with evidence of this, your immediate response is "data can only tell you so much".

That evidence probably isn't sufficient. The kind of evidence I'm looking for is: fight an actual war against an enemy with an air force, and find out your pilots never actually used dog-fighting skills or capability. Until then, maybe hold off on plans to replace all the fighters with missile trucks or something.

> fight an actual war against an enemy with an air force

Like Iraq? Or Afghanistan? Or do you mean an air force that's at least somewhat comparable to ours, like Russia or China?

Case one is that we're up against an enemy with a drastically inferior air force, in which case WVR is never going to happen. Their air capability will be devastated whether on the ground or in the air.

Case two is that we're up against a country with a powerful air force. Assuming we're somehow tiptoeing around the issue of full-blown nuclear war, we're now talking about a situation where combat is somehow happening hundreds of miles away from either side's radar-guided mobile SAM systems (Russia's S-400 has an operational range of at least 400km), both sides have managed to evade multiple BVR missile launches per airframe, both sides have an equal number of jets in the engagement, and the area objective is so important that it's worth gambling multiple $50+ million jets on.

Do you realize how comically implausible this scenario is?

If you can think of a more realistic situation in which the US will actually engage their fighter jets in dogfighting, I’m all ears.

> People have been claiming that over-the-horizon missiles will obsolete dog-fighting since before the Vietnam war:

This is exactly what this article from yesterday is claiming. I have to agree with it TBH.

https://www.bbc.co.uk/news/world-40327934

> People have been claiming that over-the-horizon missiles will obsolete dog-fighting since before the Vietnam war:

If you're interested in this line of thought, the book "Boyd: The Fighter Pilot Who Changed the Art of War" is a really great read.

> The chance of any sort of aerial dogfighting happening in a war are next to zero.

I am not so sure. Fundamentally, bullets are cheaper than missiles. A single AIM-9X Sidewinder missile comes at $400k, about the price of a brand new Cessna. A fixed wing drone under $100k is not inconceivable (drones don't have to support pilots, so all sorts of failure modes are acceptable, as opposed to what's ok for a Cessna). It would be within the capabilities of many nations' armed forces to send swarms of thousands of such drones. Shooting them down with missiles will be both expensive and impractical. The only realistic way to defend against swarms of cheap drones is to use your own cheap(ish) drones that can out-dogfigth them.

I'm not sure that these drone 'swarms' are going to have enough of a payload to take out anything significant like strong armour or heavy emplacements.

But against small units of well dug in, or hidden enemies - mortars, snipers or troops in trenches, I'm sure they'd be great. I'm thinking basically a grenade with wings that can hover, spot a target and be guided by an operator ~1km away.

Still have to fear the EMP though.

The only way to generate a significant EMP is with a nuclear weapon, at which point you have bigger problems than drone swarms. Non-nuclear EMP weapons do exist but the effects don't extend very far.

>But against small units of well dug in, or hidden enemies - mortars, snipers or troops in trenches, I'm sure they'd be great. I'm thinking basically a grenade with wings that can hover, spot a target and be guided by an operator ~1km away.

I don't know why people insist on flying grenades with lots of electronics. That is very expensive. It would be more logical to use the drones for physical target acquisition and surveillance and then use an actual grenade launcher or other conventional weapons to do the killing blow.

You're right. But if the target is on the fourth floor of an eight story building, a drone (or human) might be the only way to get the weapon there.

Slaughterbots provide better leverage of air superiority than current missile-based options.

> I'm not sure that these drone 'swarms' are going to have enough of a payload to take out anything significant like strong armour or heavy emplacements.

Maybe you're thinking of sub-$1000 quadcopter-type drones. I'm thinking of fixed wing drones, something like the Houthi Samad [1] or the Iranian Shahed [2].

[1] https://en.wikipedia.org/wiki/Samad_(UAV)

[2] https://en.wikipedia.org/wiki/Shahed_149_Gaza

Where are all the bullet shooting missiles?

There aren't. My point is, if you want to take down a cheap drone that comes at you with a 500 pound gravity bomb (and could make your day quite unpleasant), you can't afford to spend a missile on that, but you can afford to send another drone with a 50 cal machine gun or a 20mm cannon. But that means dogfight.

How about if instead of a machine gun it just explodes when it gets close? Looks like we're back to missiles.

Yes, but then it's not cheap anymore.

Are guns cheaper than explosives?

No, guns are more expensive, but they are reusable. You can fire north of 10000 rounds with a gun. The individual rounds are very cheap in the dollar range.

A drone that carries and explosive and meets another drone and blows up is quite an expensive solution. You'd prefer a gun any day.

Seems like then you're at cross purposes where you want an expensive gun on an inexpensive, vulnerable/disposable vehicle.

Guns are nowhere near as expensive as fixed wing drones. Here, for a sense of scale are some costs:

- bullet: cents

- Kalashnikov rifle: hundreds of dollars

- 500 pound unguided gravity bomb: a few thousand dollars

- 50 cal machine gun: 20k dollars

- 500 pound guided gravity bomb: 40k dollars

- Cessna 172 civilian plane: 400k dollars

- AIM-9X air-to-air missile: 400k dollars

- Houthi Samad drone: unknown, but probably hundreds of thousands of dollars

- Turkish Bayraktar TB2 drone: depending on client and delivery package, between $1 million and $5 million

- Tomahawk missile: $9 million (wikipedia says $2 mil, but that's out of date; the latest DoD budget [1] shows the Pentagon will buy 60 Tomahawks in 2022 for $551 MM, see page 23)

[1] https://comptroller.defense.gov/Portals/45/Documents/defbudg...

And what are those cheap drones supposed to do? Fire even cheaper missiles? Fire bullets at aircraft far faster, higher and more maneuverable than them?

Persistent-loitering ground support would be one idea. Keeping air support hanging around for ground forces is a big problem, and "more is better" is essentially the order of the day.

So your enemy gets to have a problem: if they do nothing, then you rain close-air down with impunity for your ground forces. But if you start shooting at drones with all your high-tech missiles, you're both spending a lot more per kill then they are per loss, and giving away the location and position of your ground-based anti-air assets (as well as forcing them to commit their high tech anti-air assets to the offensive - and they have pretty limited AA payloads).

People need to stop thinking in terms of single-branch military, because the order of the day for any modern military is combined arms - air power is in support of some objective, not "just because".

>air power is in support of some objective, not "just because".

Thats called "air superiority" which then trickles down to "ground support."

In a conflict between equivalent militaries, the idea of establishing air superiority doesn't exist - otherwise you've already won.

>> making better versions of old weapons is a waste of time.

Except for hypersonics being a basic step up for ballistic missile tech, a better warhead/payload atop a rather standard missile. Those submarines are incremental improvements on very old sub designs. Drones have been around for decades too. Even autonomous "suicide drones", but we used to call those cruise missiles with loiter capacity. Despite the constant rhetoric there is actually very little revolutionary in the modern battlespace.

There are no hypersonic air-to-air weapons, and they're not even being considered.

Hypersonic missiles are only useful against ground-based, slow moving targets. They also look indistinguishable to strategic ballistic missile launches, and cost about the same.

They get a lot of US military "give us funding" excitement, but China's various "anti-aircraft carrier" weapons are almost completely useless in war time because they look exactly like launching a nuke and you can't know in advance they're not.

The recent scmp report (i.e. info PRC wants to leak) on IR sensor development for hypersonic implies they're considering hypersonic AA to directly engage 5th gen fighters. Previously it was hypersonic drones to intercept 5th gen and engaging with normal AA missiles.

>indistinguishable to strategic ballistic missile ... >completely useless in war time because they look exactly like launching a nuke and you can't know in advance they're not.

They're useful and will be used because US will necessarily be compelled to adopt second strike posture, knowing PRC will use conventional IR/CBMs, if such capabilities are demonstrated to be credible. Especially if PRC backs up with massive nuclear retaliation (which she's building as we speak).

Reality right now is that every standoff US weapon from tomhawks to other conventional ordnance can all be nuclear tipped and are already indistinguishable from nuclear attacks for US adversaries. There's no reason PRC or any adversary capable of nuclear retaliation should give any major US platforms a benefit of a doubt - Arleigh Burkes, Ohio SSGN, B2/Lancers from AF global strike etc - they're all capable of nuclear delivery. But benefit of a doubt is necessarily extended by merely knowing US has conventional options, that will be used per doctrine, with massive nuclear retaliatory capabilities. They must wait and posture for second strike.

Hence why PRC is now conducting nuke buildup to force US into adopting second strike in response to rapid PRC advanced in conventional capabilities. Current PRC rocket force doctrine is absolutely to saturate US carrier groups with IRBMs. And it looks like eventually this will apply to conventional ICBM/hypersonics, i.e. prompt global strike capabilities once accuracy is sufficient, as per SCMP article on being able to hit "all" critical countries / strike with "strategic depth" so traditional warfare no longer exists. That means hitting CONUS as an option if US ever thinks about hitting PRC soil.

If you use a weapon which is launched from a land base like a ballistic missile and is indomitable against a bunch of targets with conventional weapons, then your adversary is not going to sit back and lose aircraft carriers to it - they're going to target the launch sites.

They may not be able to hit those sites with non-ballistic weapons. Which means, you've created a situation where they face conventional defeat because you're implicitly trying to use your nuclear deterrent to protect a conventional asset - but how are you supposed to know the ballistic response headed for your mainland isn't nuclear and isn't targeting your cities?

>implicitly trying to use your nuclear deterrent to protect a conventional asset

That's the point and what US does already.

Flip situation around, PRC conventional and nuclear infra is deliberately entangled to deter conventional attacks, but US planning may still attempt to hit these targets anyway. There's enough think tank papers out there that presumes US will attack targets on PRC soil with the unspoken assumption that PRC will be a rational actor and not launch on warning to indomitable standoff attacks launched from US delivery vehicles indistinguishable from strategic platforms. So why wouldn't PRC launch on warning from these type of attacks (apart from alleged no first use)? Because US has overwhelming second strike capability and even if PRC can rationalize US strike as conventional, ultimately they don't know for sure and is resigned to wait for determination regardless. Maybe they'll have ABM to intercept or try to shoot a few AA or conduct ASW in the meantime in anticipation that attack was conventional, but those are the conventional actions the escalation ladder affords.

The point is precisely to calibrate costs and resign decision making to second strike anyway. Structure deterrence and nuclear doctrine weapons so launch on warning becomes unthinkable to create escalation environment that allows conventional ir/cbm use. Hence PRC no first strike policy, and nuclear build up for credible second strike, and rocket force doctrine that they _will_ use conventional ir/icbms in force on force scenarios. All developments to spell out that PRC will definitely be using these hypersonic platforms, so work that into the strategic game theory. Otherwise why have such conventional capabilities in the first place? They're not for show just like US conventional capabilities aren't.

Because conventional capabilities don't matter if they're indistinguishable from nuclear capabilities till they impact.

Or put it another way: China launching a hypersonic kill vehicle at a US carrier is pretty much indistinguishable to China launching a tactical nuke at one. Their launch and flight profiles are identical, and while fallout is an issue, the use of a battlefield nuclear weapon wouldn't actually warrant a strategic nuclear response. From a US perspective, the issue is they lost a carrier because the nuke is going off in Chinese waters in your scenario.

But the problem is that launch is a ballistic missile. The only actual defense is that launching 1 doesn't look like a strategic nuclear strike. But let's pretend for a moment you're somehow using these in air defense: you'd be launching hundreds and there's no way to know while you have eyes that they're not a strategic strike.

The PRC won't launch - or hopefully doesn't launch - in a conventional war because the counterstrike would definitely kill them. Same reason the US doesn't. You can de-escalate a conventional war, but you can't de-escalate strategic missile launches at cities.

You can quite happily today roll out onto the battlefield and start slinging artillery fired nukes and it would be a big deal but not warrant strategic nuclear response because everyone is only in danger of losing their deployed forces, not their entire industrial base, population and environment.

>Because conventional capabilities don't matter if they're indistinguishable from nuclear capabilities till they impact.

Then wait for impact and assess. That's the entire point of having flexibility of second strike, to give options other than launch on warning. It's why SSNs exists, to make space for conventional warfare. It's the posture major powers have adopted to reduce or mitigate unknowns in case of peer warfare where both sides are going to throw everything at each other. Because they'd rather wait to be sure than end the world on a maybe.

Again from PRC perspective, almost every conventional US capability are already indistinguishable from nuclear. Especially stand off / beyond visual ordnance that will be used against PRC soil in initial SEAD wave or whatever used to penetrate A2D2. At minimum they threaten the entangled PRC conventional/nuclear infra, which is direct attack on PRC nuclear forces. So do we expect US never to attack PRC mainland? We can hope not, but it's mentioned in enough wargames / policy / think tank papers that it can't be ruled out. So a proportional conventional PRC counterstrike can't be either. Even if it elevates chance of strategic retaliation.

The best that can be done is set expectations that retaliation will be proportionally conventional before moving up the escalation ladder, i.e. US Lancers hits Dalian shipyard building new PRC carriers, PRC notifies US via hotline that 10 conventional hypersonics is going to sink CVN being retrofitted at Puget Sound within 15 minutes. It's not risk free, but everything becomes risky when things go hot.

I remember I read a fighter pilot's analysis on the events:

- Bullet drop and travel time were not modelled, the guns were literally lasers

- Most of the kills the AI did were high aspect shots, meaning the aircraft flew perpendicularly to each other. In real life, these shots are almost impossible to hit, due to the aforementioned ignored factors

- A human pilot in a simulator doesn't have the situational awareness of flying in a physical plane

Yes.

>> The pilot, dressed in an olive-green flight suit, sat in a high-backed gaming chair, his face obscured by a virtual-reality headset.

So nothing like the real cockpit they would have trained in, with low-resolution, high-latency visuals. This alone could have removed a lot of the expert advantage of a real pilot in a real plane.

If they didn't model bullet drop, did they model effects of high-G on the pilots?

Just as advances in war tech in the past have led to the more "advanced" war machine "fighting the last war," these advances only serve to change the behaviors, rules, and context of any future engagement.

Keeping in mind that war is ultimately a "battle" of resources and expenditures, no one in a position of great disadvantage will invest in that particular effort, but seek a way to circumvent it. Whereas the "winner" has already sunk that cost. (Even without AI, how common are dogfights or air-to-air confrontation in the last 50 years?)

There are ways for the more advanced participant to defeat this game, but so far it has required violating human norms that are held sacrosanct even in war.

AI's advantage might end up being the ability to navigate this evolving game theory, but only if it's able to properly value what's acceptable to human nature and what isn't.

A "kill all humans as a solution to end spam" doesn't work here (unless the AI has taken ultimate charge, which becomes an entirely different issue).

On the other hand, certain near-peer potential adversaries have signaled a willingness to violate whatever norms need be violated to secure victory.

Arthur C Clarke has a good story on the dangers of warfare at the bleeding edge

https://en.m.wikipedia.org/wiki/Superiority_(short_story)

>> As well as any nation who's best and brightest minds refuse to contribute while their adversaries have no such qualms.

Wait, what? Are you saying that if a scientist refuses to contribute to AI development that will be used to kill human beings then he plays into the hands of nation's adversaries ?

Not to put words in their mouth but isn't that pretty clear? If you don't develop advanced weaponry, you are going to be killed by people who _do_ develop advanced weaponry.

Unless you manage to stop the advanced weaponry from being developed at all, which is admirable, but incredibly difficult, and with huge consequences for being wrong.

Or unless you have a significant enough deterrent and the credibility to use it in basically any sort of conflict (N. Korea may fit into this category)

That depends on your model of international relations. Most states have recognized that international relationships are far from zero sum, and that wars between peer competitors are almost never worth it. Indeed, the success stories since WW2 have been those countries that have spent very little on weapons, and engaged in few wars (with the US being the exception).

Even if you take WW2 and WW1 into account, the biggest winner of those conflicts (the US) spent the least on armaments prior, and had probably the least aggressive foreign policy of any major power.

> have been those countries that have spent very little on weapons, and engaged in few wars (with the US being the exception).

Perhaps, because the US is an exception.

Maybe? China has historically spent very little on its military, and it's not like they benefited from US protection. Quite the opposite, in fact.

China's military spending is now effectively near parity with the USA and rising rapidly.

https://www.heritage.org/defense/commentary/chinas-defense-s...

Dunno about the methodology there, but, that's why I said historically.

Great, so maybe it's time to revise the rest of the thinking as well.

>>but isn't that pretty clear?

No, for me it's counterintuitive. Personally I'd never contribute to any military technology.

OK, but that means someone with good weapons can kill you and take your stuff.

killing me is easy enough as-is, without the need for advanced technology. Literally any cop can bash my head in. Further arms race when nuclear weapons are present is pointless, as any conflict with nuclear power is likely to end in nuclear exchange, and is acceptable only if you are a psycho that thinks megadeaths is just a number upon which strategy can be built.

Although the idea that MAD saved us from a nuclear holocaust is prevalent, people tend to forget how a strong anti-war effort was required to dissuade the crazy people from ideas like megadeaths being an ok strategic decision and stopping plans like the "Star Wars Program" on their tracks (which has unfortunately resumed with potentially tragic consequences).

These weapons and technologies are developed by nation-states only, and a collaboration, however unlikely might seem to you, is the only way to stop us from pointing guns at each others heads with AI controlling the trigger.

The technologies will be developed, and have to be developed for the sake of balancing powers, but they need to be put in a drawer and locked there after being designed, not widely deployed and flaunted at each other.

Yes, that's exactly what it means.

Consider for a moment that you believe that Nuclear weapons are abominations which should be all eliminated.

How do you convince North Korea to eliminate theirs? If you eliminate your nuclear weapons, you now open yourself to being threatened with nuclear destruction by North Korea with no mutual assured destructive response to deter them.

The genie is out of the bottle and you cannot stop it. But you can deter its use against you.

Surely, this is not a serious take.

For one, NK hardly has the capacity to "destroy" the US - they barely have working ICBMs.

Secondly, whatever the amount of nuclear warheads necessary to deter this is, it's surely less than 3,750[0].

We cannot know whether it is impossible to stop the "genie," as no serious attempt has been made. The U.S. and Russia together hold 90% of the world's nuclear weapons. The Cold War being over, are you seriously proposing they need this entire arsenal in order to defend themselves against the very serious and dangerous threats of checks notes Pakistan and North Korea?

If so, then how come the US still reserves the right to a pre-emptive nuclear strike? How come it has not adopted a "no first use" policy like China and India?

[0]: https://en.wikipedia.org/wiki/Nuclear_weapons_of_the_United_...

I chose North Korea for the thought experiment because of it's belligerence, not because of it's quantity of nuclear weapons.

>The Cold War being over, are you seriously proposing they need this entire arsenal in order to defend themselves against the very serious and dangerous threats of checks notes Pakistan and North Korea?

Why did you choose to ignore the Chinese who are currently rapidly expanding their nuclear arsenal. The world is on the precipice of a new cold war between Beijing and Washington.

>If so, then how come the US still reserves the right to a pre-emptive nuclear strike? How come it has not adopted a "no first use" policy like China and India?

A "no first use" policy is worth less than the paper it would be printed on. No one serious actually believes in them.

Well, is it to defend against small, irrelevant third-world countries with a handful of nukes with barely functioning delivery systems, or is it to defend against large, serious places with serious arsenals? Because these are two very different goals.

If North Korea is a bad example for a thought experiment, then you should pick a better example.

As for China - surely the U.S. and China could come to a mutual agreement to reduce their nuclear arsenals?

It might be worth noting here that China's "rapidly expanding" arsenal clocks in at 350 warheads, while the U.S. has only a meagre 6,185 (barely 17x the size). If America is concerned about large nuclear arsenals, surely they could take the first step or three?

> A "no first use" policy is worth less than the paper it would be printed on.

In that case, its decision seems even stranger.

> As well as any nation who's best and brightest minds refuse to contribute

Sickening pro robo war jingoism!

> As well as any nation who's best and brightest minds refuse to contribute while their adversaries have no such qualms.

I am worried about this somewhat. I know a few too many great engineers reluctant to work on defense work because they are concerned about what their friends will think. It's not particularly tasteful in progressive circles.

Those fucking defense companies can start paying more and maybe I’ll consider building weapons to be sold to the Saudis.

> This bodes very poorly for any nation that falls behind in the AI race. As well as any nation who's best and brightest minds refuse to contribute while their adversaries have no such qualms.

Curious how you think such an arms race avoids complete annihilation of both sides.

There is a great course by Prof. William Spaniel about Game Theory in the context of Nuclear Proliferation. Highly recommend it for anyone who wants to know about the nuts and bolts of how nuclear policy is shaped: https://www.youtube.com/playlist?list=PLKI1h_nAkaQqivWh2_XEt...

Basically, you have to choose between MAD or asymmetrical destruction. With exponentially improving AI, the situation is more like an inverted pendulum. Any small perturbation (say resources or breakthroughs in AI) would lead to catastrophic asymmetrical advantage for one side.

That doesn’t make much sense.

You’re implying that because this AI beat another AI by a lot that all skill gaps between AI must be significant. What if the second and third place versions just weren’t as good and over a relatively short term it all converges?

This is an extremified version of the general progress of technology in warfighting though: while we have the stories of automotive factories being converted to produce tanks during World War 2, that was something that was only possible in the environment of the time.

More then ever, you go to war with the military you have, the outcome is going to be decided very quickly because the rate of equipment attrition is enormous: military victory will be decided in weeks, not years (you can of course still lose the occupation, but that's a different matter).

An AI system will consistently beat an opposing AI system if it has been trained to beat the opposing AI system, as it will only engage if it can predict a favorable outcome. Fully autonomous systems are going to have to be very, very secret because one leak will destroy their utility. This alone might be a good reason to keep them as pilot assist systems rather than pilot replacement.

Nothing a proper EMP weapon will not sort out.

https://www.doh.wa.gov/portals/1/documents/pubs/320-090_elec...

From your link:

> There are two basic ways to protect or harden items against EMP effects. The first method is metallic shielding. Shields are made of a continuous piece of metal such as steel or copper. A metal enclosure generally does not fully shield the interior because of the small holes that are likely to exist. Therefore, this type of shielding often contains additional elements to create the barrier. Commonly, only a fraction of a millimeter of a metal is needed to supply adequate protection. This shield must completely surround the item to be hardened.

> The second method, tailored hardening, is a more cost-effective way of hardening. In this method, only the most vulnerable elements and circuits are redesigned to be more rugged. The more rugged elements will be able to withstand much higher currents. This method has shown unpredictable failures in testing, though it is thought it may be useful to make existing systems less vulnerable.

Seems reasonable to provide EMP shielding to a drone?

Since we are capable of EMP-hardening very large airframes (see Nightwatch, Looking Glass, SAM) I assume shielding crucial systems on a drone would be an initial design challenge but not impossible.

Do we actually have non nuclear emp weapons

Also from the reference above:

"For weapons purposes EMP producing sources other than nuclear detonations have been successfully developed. Several nations, with United States at the forefront, are reported to have developed non-nuclear bombs capable of generating EMPs. Electromagnetic bombs (E-bombs) are specialized, non-nuclear tools designed to destroy information systems. These devices are primarily intended for battlefield application..."

Yes, but they really only work on civilian targets because military hardware is EMP hardened and has been since a few weeks after the advent of nuclear weapons. The scary stuff is electronic warfare where you instruments seem to be working but they're not.

> This bodes very poorly for any nation that falls behind in the AI race.

How much do flight characteristics matter? A better AI might still lose in a worse plane.

I could say the same thing about lots of obvious economic policies... The world doesn't work like that.

Reply on news.ycombinator.com