What a 20-year-lead in military tech might look like

by Daniel Kokotajlo16 min read29th Jul 202038 comments


WarForecasting & PredictionWorld ModelingAI

I've spent way too much time speculating about near-future military technology. Here is a list of technologies I think are likely to be important to a conventional war in, say, 2040.

If you think this list shouldn't be taken seriously because I don't have official expertise, you are probably right. I think that if I became an expert I’d probably end up deleting a third of the things on this list and adding a similar number of new things. (I did run this draft past three people with relevant expertise, though!) I look forward to critiques in the comments.

If you are wondering why these speculations are relevant to LessWrong… well, it was requested, and also it’s relevant to a point I want to make later about AI. You see, the technologies on this list could probably be quickly achieved with the help of advanced-but-not-godlike-AI. The gap between these technologies and present-day military tech is an example of a “small” gap; a “merely” 20-year-by-human-standards gap. So, for example, suppose some corporation or nation gets AI systems which are good but not godlike at designing new technologies. They vastly speed up the process but don’t do anything human engineers couldn’t do in 20 years. Or, suppose there are lots of AI systems in the world but one faction has systems which are the equivalent of 20-years-by-human-standards ahead. In either case, I think, the gap would translate to an extremely large advantage, possibly enough to enable a small group to take over the world, conquistador-style.

(To be clear, while I do think AI takeover is probable, I think persuasion/politics/ideology is by far the most likely method.)

Better command-and-control capabilities (incl. Cyber and intel)

Command & control technologies are very important. Wars are won or lost on the basis of how well each side is able to observe, predict, plan, organize, and outmaneuver the other. I think the advancements in command and control tech that are likely to happen in the next 20 years are more important than everything else on this list combined. In case this isn’t obvious, I’ll say a few things below by way of justification.

Long-ranged guided weaponry already makes it easy to blow up targets if you know where they will be when your attack arrives. Winning is all about finding the enemy before they find you. Now more than ever, being able to quickly gather loads of information and integrate it and communicate the resulting orders to your forces is super important, as is being able to disrupt your enemies’ attempts to do the same.

There’s lots of room for new technologies in this space, or much more advanced versions of already existing tech: Satellites, spy drones, advanced sensors of various kinds, more robust communication networks (e.g. cloud computing), AI to integrate all the data, draw some interesting conclusions, and instantly make some of the relevant decisions. Already the US military collects way too much data for teams of human analysts to examine, and anyhow human analysts take too much time. Better to have a computer say “Odds of imminent attack in sector 7 have spiked to 11%” as soon it notices some unusual patterns. (A version of this might be useful at the tactical level, e.g. “The sniper who shot that bullet is probably in this window; the mortar shell was probably fired from the alley at coordinates XY.”)

What would it look like to have a significant advantage in C&C+intel+cyber tech? The enemy would be swinging blind punches, firing missiles at locations where your troops aren’t, or marching forwards in the hopes of making contact while your forces quietly dodge around them or set up an ambush. Your aircraft would fly low over their country, slipping through gaps in air defenses exposed by your satellites, destroying key nodes of communication and transport. In regions you occupy, insurgency would be impossible, as everything that moves (at least in important regions) would be tracked by your cameras. In regions the enemy occupies, your insurgents (probably local allies supplied with weapons and guidance) would be a real threat. Meanwhile, you might have all sorts of electronic spies in their network, due to superior cyber capabilities. You might even be able to shut down some of their systems at crucial moments.

Persuasion tools might also be possible. Propaganda is already a thing, but personalized propaganda powered by big data and machine learning will be much more potent. Moreover, it is useful in peacetime also, and useful to many non-military actors, so it will be heavily invested in and rapidly developed. This is another reason why having an advantage in this sort of tech will help suppress enemy insurgencies (and supercharge your own). It might also help with “turning” key enemy personnel (such as a captured leader, or even a leader you have communication access to via a spy or hack or parlay). At the very least it will help keep your own troops' morale and loyalty high, which has always been and still is extremely important.


Aimbots are guns that aim and fire automatically. A simple version would be a camera and computer hooked up to the trigger of an infantry rifle, such that the rifle fires when the computer calculates that the bullet would hit the target. The software involved would be a bit of image recognition, maybe a bit of physics, etc. The user would point the gun at the target and the bullet would fire at precisely the right moment, increasing accuracy while decreasing aiming time. These aimbots already exist and are being tested by the US military.

A more complex version would have some way of pointing the gun. For example, a canister of compressed gas as an under-barrel attachment, with multiple nozzles pointing in different directions. The computer selectively opens the nozzles for split-seconds, emitting jets of gas that jerk the barrel in such a way as to line it up with the target, at least momentarily, during which the computer fires the bullet. Perhaps the camera would also be connected to a cell phone screen, so that the soldier holding the rifle can stick it out from behind a wall while his buddy looks through the camera screen and clicks on targets.

Anyone who has played a first-person shooter with an auto-aim hack or ability will already think this is powerful. A rifle attachment that quickens your reaction times and makes almost every shot hit? Amazing! However on a real battlefield the effect would be much more powerful than it is in video games. Here is why:

The vast majority of bullets fired in war are not aimed, but sprayed in the general direction of the enemy. This is called "suppressing fire;" it keeps the enemy pinned in place and prevents them from shooting back at you. Suppressing fire is mostly not a phenomenon in video games because players don’t die in real life if they get hit by a single video game bullet. In fact in most games a single bullet doesn’t even kill your video-game character, and if your character does die he simply respawns. In real war, however, suppressing fire is the primary way to deal with enemy infantry. They are usually in cover, so you can’t just shoot them; however, you can suppress them, so that your own forces have freedom to maneuver safely. Then, your own forces can flank the enemy position, or close with it and lob grenades, or whatever. Of course, they’ll be trying to prevent you from doing this by suppressing you; you “win the firefight” when they are suppressed and you are not.

Now suppose that one side has aimbots on their rifles. They can now shoot their rifles without poking their heads out of cover (the point-and-click method mentioned above). This means they can provide suppressing fire without endangering themselves. This means you can’t win the firefight against them, at least not in the normal sense. Meanwhile, their shots are way more accurate, and their reaction speed is instantaneous--meaning they don’t have to spray bullets in your general direction, they can just wait, silently, and shoot a single bullet at you when you poke your head around the corner. The bullet will hit you before you can react to whatever it is you’ve seen. What this means is that (1) your forces will be suppressed extremely quickly, even if you outnumber the enemy and start shooting at them before they start shooting at you, (2) you’ll lose a few men before you realize what is happening and hunker down, (3) their ammunition will last much longer because they make every bullet count, and (4) after the enemy leaves you might still be suppressed because you won’t know when they are gone because you are afraid to look. (With ordinary human-caused suppressing fire, the shots are being sprayed at you and are unlikely to hit, so if you are brave you can peek for a second at relatively small risk. Problem #4 can be solved to some extent by countermeasures like using your own cell phone or a mirror to peek around the corner, assuming their AI isn’t good enough to shoot it too.)

What this means is that an infantry force equipped with aimbots would be not only much more deadly, but also much harder to kill or stop. It would move through the battlefield like a basilisk, paralyzing and/or destroying anyone who comes within line of sight. Of course, if both sides have aimbots, the fight would be much more fair.

Aimbots would help enormously against drones. See next section. Aimbots might also lower the amount of training needed before a recruit becomes better than useless.

Instead of an attachment to an infantry rifle, an aimbot could be implemented as a portable autoturret. It’s on a tripod or something, you plop it on the ground and it can aim and fire automatically, autonomous or remote-controlled. I think this would be less efficient than the rifle implementation but it would have its advantages too. For example, you could attach them to your vehicles. This already exists and has been quite effective, see e.g. Trophy.

Battle bots

These are machines which propel themselves around the battlefield, identifying and attacking targets. Early versions are mostly remote-controlled by humans, later versions are more and more autonomous. There are many varieties, and related concepts. The core idea is that battle bots are cheaper than human soldiers; for the price, hassle, and transportation cost of one human soldier you can deliver dozens if not hundreds or even thousands of battle bots to the field. There are other advantages too. For example, they don’t have morale problems, and they always obey orders. On the other hand, they lack initiative and creativity and lose more of their effectiveness when communication with command is cut off. But these disadvantages will lessen as the technology improves.

Drone swarms are perhaps the most important kind of battle bot. Kamikaze drones fly into the enemy and explode. Gun drones shoot bullets. Utility drones can transport cargo, open doors, recharge other drones, drag huge nets through the sky to catch other drones, scout with bigger cameras and computers, drop bombs, and many more things, depending on how they are equipped. (All of the above would come as modules that can be attached or detached easily by a human handler). I’m not sure whether kamikaze or gun drones would be more prevalent. Currently kamikaze drones are. (Note: I the previous links I found after writing this. It’s nice to see my predictions confirmed.)

Drone swarms, since they can fly, can close with the enemy very quickly. Since they are cheap and small, you can have very large swarms and transport them to the battlefield easily. (You can even just drop them out the back of an airplane!) Say a drone carrying an explosive charge the size of a grenade costs $10,000 — a conservative estimate considering that the computer on the drone is probably the most expensive component, and wouldn’t be more than $2,000. Then you can drop 1,000 of them on the battlefield for about the price of a tank (crew not included). Or 200 for the price of a cruise missile. Swarms this big would be very hard to kill. They are just a bunch of dots in the sky moving very fast and maybe wiggling erratically. With aimbots you could shoot them down, but even an autoturret would probably only be able to take out 10 or so before they closed in on it and blew it up. And depending on how dense the terrain was, maybe that number would be more like 1. So the swarm would just steamroll over everything in its path until it ran out of battery, or was stopped by a force of aimbots or battlebots of similar size. (Ordinary human infantry would be much less effective. Even if you have good aim and good reflexes, you probably won’t be able to shoot more than 1 kamikaze drone out of the sky before the swarm is upon you.) Hiding in buildings or under cars or under camoflauge would be an effective tactic against drone swarms, but not super effective — after all, while you are hiding you can’t contribute to winning the battle, and moreover the drones can just land like locusts, conserving battery and waiting for targets to come out of hiding. (This is what makes drone swarms better than artillery barrages or massive airstrikes. That, and the ability to distinguish friend from foe from civilian. And the ability to redirect to somewhere else, or even come home, if the enemy turns out to no longer be in the target location. And the ability to sweep a broad area looking for enemies.) Moreover, some kinds of drones (quadcopters) might be able to follow you indoors. See the Slaughterbots video, which has some large kamikaze drones for taking out doors and windows, and then small kamikaze drones for killing humans. You could hang beads and cloth in doorways and hallways to impede drone movement, but utility drones with attachments could get around that.

So far we’ve been discussing swarms of small drones. But swarms of big drones will happen too. Air-to-air combat in the age of autonomous vehicles is very much a numbers thing. How many missiles can you bring to the battle? How many distinct aircraft do you have, so that the enemy needs to bring at least that many missiles? How much range do your missiles have? We already have big Predator and Reaper drones, fitted with various kinds of missiles. Just make loads more of them. In world war 2 the biggest air raids had thousands of planes. Nowadays we could probably make at least that many big drones, and probably many more. (Spot check: Predator costs $4M, so the Pentagon could easily afford to buy 10,000 of them per year, even without accounting for economies of scale.) “Air raid” will probably be a misnomer; “Air occupation” more like it. A stormfront of drones smashes into enemy territory, exchanging missiles with enemy aircraft and anti-air defenses, losing many drones in the process but not nearly enough. Drones that get low on fuel or ammo go home, and are replaced by reinforcements. Enemy territory is paralyzed due to constant circling drones overhead, ready to drop missiles on any suspicious activity.

Minitanks are autoturrets with wheels. Maybe they have some armor too, maybe not. They are much less mobile than drones, even less mobile than human infantry. Why build them? Well, if you need lots of auto-aimed guns in a certain location, maybe you don’t have enough human legs to carry them. Or maybe your humans are too valuable and fragile. Why not just build drones instead? Well, maybe one minitank can shoot fast enough and is armored enough to be worth 10 drones in most fights, but costs only as much as 3. Ground travel is more energy-efficient than air, and it can afford to have armor. Multiple militaries are already experimenting with these.

Drone submarines are to naval warfare what large drones in the sky are to aerial warfare. They’ll be much smaller and cheaper than ordinary submarines; maybe they won’t even launch torpedos, maybe they’ll be kamikazes instead. Swarms of them will blockade enemy coastlines, ruin enemy trade routes, etc. The US military is already experimenting with this.

Balloon bombs are a special kind of drone swarm. In World War Two, Japan made about 10,000 balloons that carried bombs, and sent them up into the jet stream to drift over the USA and drop their bombs. At the time they didn’t have any sort of ability to aim the bombs or the balloons, so very little damage was done. However, project Loon has demonstrated that modern technology can steer balloons pretty much anywhere you want them to go. It just takes a while for them to get there. Once there, the bombs they drop can be smart bombs, that identify targets using cameras and GPS and glide right to them. A glide bomb could hit a target within about 70 miles of the balloon it drops from. They could also carry air-to-air missiles. Balloon bombs are like large drone swarms, except that they are much much cheaper. It might even cost more to shoot them down than to make them! If Japan could make 10,000 eighty years ago with a tiny fraction of their war effort, a modern nation with a bigger economy could be motivated to make millions.

Drone carriers are to tanks and artillery what aircraft carriers are to battleships. A drone carrier could be a civilian pickup truck or trailer truck, with crates of drones, maybe a generator, loads of charging cables and spare batteries, and a few people in the back. Someone drives, maybe someone else is on their laptop giving commands to the drones, and the people in the back unpack drones from crates and launch them on strikes and then grab returning drones out of the air and swap out their batteries. By military standards civilian trucks are practically free, so a drone carrier is basically as expensive as the people and drones it carries. And the people need not be highly skilled. And the drones are cheap. Just as battleships became obsolete, so too will tanks and maybe even artillery become obsolete. Drone carriers will be just as mobile, but longer-ranged and hit with more striking power. And they’ll be much much cheaper as well. They’ll also be easier to hide since they can pretend to be civilian trucks and easier to deploy because they aren’t as heavy or bulky and also the truck can potentially be stolen from the local population after you arrive.

Finally, this technically doesn’t count as a battle bot because it doesn’t fight, but… currently advanced militaries use a small portion of their soldiers to shoot at the enemy; most have some sort of job repairing and maintaining vehicles, driving transport vehicles, loading and unloading things, etc. Many of those jobs could be automated away by self-driving vehicles or boston-dynamics-style robots, making military logistics faster and cheaper and freeing up more troops for other tasks. This would be especially useful if you need to maintain a swarm of thousands of big drones.


Starships are vehicles like the SpaceX Starship. Access to space will be very important in a future war. Being able to view in real-time every inch of their territory, while they are stuck guessing what is happening in yours, is a humongous advantage. If you can launch things into space more cheaply than they can, you can destroy their satellites and put up swarms of your own. Starships will plausibly be at least one order of magnitude cheaper for delivering stuff into orbit than legacy rockets, possibly two or more. (Caveat: In space, it is generally easier to destroy than to create. So even if one nation can put up 10x as many kilograms than another, the other nation might be able to render space unusable, e.g. by creating a sort of man-made Kessler Syndrome. However, with Starships it would become so cheap to put things into space that this difficulty might be overcome, e.g. by putting up swarms of armored satellites that can last half an hour before all being destroyed, but during that time give you the valuable information you need. I’m not sure about this and would love to see an analysis of the effectiveness of numbers and/or armor vs. tons of tiny bits of metal polluting LEO)

More excitingly, Starship is supposedly going to be capable of transporting cargo and humans across large fractions of earth’s surface in mere minutes. SpaceX is in talks with the US military about this; they claim that eventually their costs will be low enough that a Starship trip will cost about as much as a Globemaster (military cargo plane) trip. It can carry about as much cargo, but does so much faster and without the need of a runway on the end. If this is true, it might make cargo planes obsolete, or at least substantially improve logistical capabilities, especially for nations who don’t have very powerful navies or air forces.

The most interesting (and also most speculative) potential use is for delivering troops and other payloads into battle. Because the starship would probably not be reusable after getting shot at, and would not have access to refueling facilities in battle anyway, this would be an expensive strategy. However, the benefits might be worth it--I’ll try to make the case for that below.

(You could use Starships that normally serve as cargo ships, thus you still get to amortize the cost over several flights before you throw them away in the attack. Or, you could custom-build single-use Starships that use a two-stage design to save all the expensive bits for reuse.)

A SpaceX Starship can carry 100 tons. 50 tons of drones is plausibly 50,000 drones of varying sizes. That’s enough to paralyze a small city; drones everywhere, blaring on tinny speakers for everyone to lie on the ground with their hands on their heads. You’d have 50 tons of cargo space remaining for troops and miscellaneous equipment.

Or, you might use 99 tons to carry extra fuel, so that when you land you could immediately take off again, perhaps after having picked up or dropped off a few very important passengers. You wouldn’t be able to fly all the way home, but maybe you could fly somewhere safe.

You could strike anywhere in the enemy country, at any time, and they’d only have twenty minutes of warning at best. So, you could e.g. land at all major power stations simultaneously, or all major cities, or all major airports. (Seizing airports is especially important because you can use them to land reinforcements; see the airborn invasion of Crete) You could land wherever their forces are not, if you have good enough intel, which you probably would since you have starships. Without actually attacking, but merely by threatening to, you could force them to keep much of their military at home, spread out over the country, defending various important installations. This particular advantage scales with the size of the country you are fighting; larger nations have higher area-to-perimeter ratios, so forcing them to defend their whole area rather than just their perimeter is a big deal.

Being able to strike anywhere at any time on very short notice is more valuable than it sounds. It’s easy to underestimate the importance of speed, surprise, and reaction time in military conflict. For example, say a country friendly to you is surprised by a coup. If you don’t act quickly, it will switch over to your enemy’s camp. Fortunately, you can load and fuel a squadron of Starships in a few hours, and then it takes only twenty minutes to arrive on scene. Or, maybe you are launching a big attack at dawn. Simultaneously with the attack you can assault the airbases, anti-air facilities, and transportation hubs behind enemy lines with drone swarms packed into starships. The drones could potentially be launched while the starship is still high in the sky, making mission success still possible even if the starship is shot down.

One problem with attack starships is that they might be hard to distinguish from nukes, and thus might trigger MAD. This problem could potentially be solved by using them against non-nuclear enemies, or by using them one by one instead of all at once. (Launch one starship every fifteen minutes; that way if they are secretly nukes the enemy will still have enough nukes left to massively retaliate; that way they won’t massively retaliate until they see whether they are nukes or not.)

3D printers

I am of the opinion that 3D printers will be a big deal in ten years or so. Currently 3D printers are expensive to buy, expensive to operate, and the parts they build are crappy. So they are great for rapid prototyping, but not good for being part of a factory that makes a finished product. However, the parts they build are rapidly getting better--in some ways they are better than any other manufacturing method. Meanwhile the cost of 3D printers is falling pretty fast. Already I know of some products (cars, rockets) that have some important 3D-printed parts. As the price falls and quality improves, more and more parts will be made with 3D printers.

Anyhow, having lots of 3D printers lying around is valuable militarily, I think. Three reasons. First, military vehicles break down constantly and being able to print replacement parts is nice. Especially since some vehicles will no longer be in production so parts will be scarce, and other vehicles may be captured from the enemy and thus have no spare parts at all.

Second, insurgencies have supply chain problems. They need to buy weapons abroad and smuggle them in somehow. Being able to print what they need is a game-changer. Even major nations may have supply chain problems if their enemies have destroyed your most important factories and supply depots with long-range missiles or airstrikes; it sure would be nice to have a “virtual factory” consisting of hundreds of thousands of 3D printers distributed throughout your nation.

Third, a future war will involve rapidly changing and evolving technologies and tactics. World War Two was already like this; the weapons and tactics used at the end of the war were substantially better than those used in the middle, which were substantially better than those used at the beginning. 3D printers speed up the design/prototyping process by, like, an order of magnitude or more. If you have enough of them, they can also speed up the mass-production process as well; instead of taking three months to build a factory that produces 1,000 of the new widgets per day, you can command 10,000 of your nation’s 3D printers to immediately produce one widget per day each. Example: Your enemies have countered your drones by putting up nets everywhere. Response: Design a net-cutter attachment for your drones, and patch their software so they know how to use it. Thanks to 3D printing you can be testing out 1,000 different designs within five hours of seeing the enemy nets. In a day you have settled on a design that works, having collected enough data to train/program your drones to use it. It’s a hacky solution, sure, but that’s OK since you’ll be working on a better solution soon. By the next morning you’ve printed a million net-clippers in various facilities around the country and are already shipping them off to the front to be affixed to your drones. Another example: Your eggheads come up with a complex 3D pattern that works as an adversarial example for the enemy aimbots, when viewed from a variety of angles and lighting conditions. You print out a bajillion of them and distribute them to your troops for the next assault, before the enemy realizes the problem and patches it. And when you find a new way to fool their aimbots, your printers will be ready again.

Laser weapons

Their main application is for defense against incoming drones and missiles (and balloons?). They are bulky and require a lot of power, but they might still turn out to be better than the CIWS and other autoturret systems available. They could also potentially be used as better sniper weapons, since light travels much faster than bullets. They also can and already are being used as “dazzlers” to temporarily (or, if you are willing to violate treaties, permanently) blind humans and cameras. However I am not sure laser technology will ever be good enough to outcompete more mundane alternatives, and even if they do, they probably won’t be that useful compared to conventional alternatives. I might be wrong though.


Exoskeletons are definitely a thing. Moreover I am fairly confident that given enough R&D, a version could be designed that includes enough armor to enable the wearer to smash through doors and storm an enemy-occupied building, invulnerable to ordinary bullets. However, by the time this is designed, it will be mostly obsolete, as battle bots can do the same stuff but better and cheaper and without risking your soldier’s life. At least, so I predict. Perhaps a lighter exoskeleton will be built that goes over the soldier’s legs, and allows him to walk longer and carry heavier loads without getting tired. Even if this happens, I doubt it will be a game-changer. Similarly, I expect exoskeletons might see some use for logistical purposes, helping troops load and unload cargo, or arm and repair aircraft. But not a game-changer, especially since robots might get good enough to do those tasks as well.