@graydon The Geneva convention is pretty damn firm on the requirement to positively identify a target as a combatant, notwithstanding the massive criminal negligence we've seen from the US and Israel there, and even the Israelis are still meticulously faking a paper trail. I think there will be a lot of hesitancy to implement something that dumb, and even if you did, you'd need to implement serious IFF systems in them, which adds cost and attack surface. IFF hardware is usually pretty heavily protected with self-destroy-on-tamper devices etc, which is hard to do when you need it on munitions you're buying by the 100k, not to mention key distribution, etc. It's not that this is impossible, but it's neither cheap nor easy. Yes, folks outside of conventional militaries may do this stuff anyway, but it's not a straightforward set of choices even there — human in the loop solves a ton. @dr2chase@cstross@isaackuo
It's why I don't expect we're going to see a "recognise faces" approach. Think "biting insect"; it's got several sorts of toxin and enough circuitry to go "is it warm?" and "does it have a heartbeat?" and some way to prefer stinging skin to armour.
Add in a few simple eusocial rules and some return-for-reload capacity if you can, but even as a one-way munition, the cost per corpse is likely much lower this way.
@cstross@isaackuo@graydon and also, image recognition is nothing like the LLM stuff, it trains fast and cheap and runs on tiny hardware (source: niece's friend's boyfriend who worked in that industry, also looking for a job; and TinyGo-powered flying drone demos, that recognize human faces, now).
@isaackuo@graydon Longer term, the speed with which GANs for image recognition are advancing (see also the AI bubble) suggests fully autonomous attack drones—or drones which go autonomous if their command channels are interrupted—are likely to show up within a very few years.
Yeah, we don't know where measure/countermeasure is going with drones, yet. The article mentions body armor, but as you note there are other possibilities.
Jammers have some inherent limitations - they're completely ineffective against wire guidance, which is already in use, and wire guidance could be enhanced with towed kites and laser comms (the kites provide altitude for direct line of sight at long range).
I think hard kill air defense has more potential 1/2
@cstross Nothing says you can't build or power a mobile jammer that puts out several MW; nothing says you can't build defensive drones, nothing says the sensor quality contest has even happened yet (can the attacker spot the target before the target spots the attacker and responds?), nothing says the trend to precision fire over volume of fire can't produce precise volume of fire. (the move to precision is economic.)
@cstross What we're seeing with drones is the collapse of the loop between "see" and "hit"; the diffusion of fire which increases specific lethality (individual weapons are much closer to just enough); and the beginnings of the ability to do reconnaissance-by-fire in a literal way. ("seek until found", then kill it.)
Thing is, this is an asymmetric mass contest; Ukraine has cultural machinery Russia doesn't. If it's analogous to 1915, it doesn't reflect what 1918-equivalent will look like.
From 1860 to 1914, the decisive force is rifle regiments; you see things like the reconquest of Sudan where enemy forces are shot to pieces at ranges above 1000 m.
1915 and subsequent, the continuous front and concentration means anything seen gets ground up by artillery from beyond rifle range. It's a pure mass contest. (and "mechanised warfare" dilutes by making the continuous front move. 1939-1945 is an historical anomaly.)
@graydon@dr2chase@isaackuo Yup. Another thing: I think we're going to see the end of standardized uniforms/rank insignia, sooner rather than later. We've already seen the end of officers in braid and distinctive uniforms on the battlefield (snipers) but this is going to be orders of magnitude worse.
@wonka Not having effective iff or visual human confirmation means you turn your semi-autonomous munitions into fratecide machines.
Shockingly, there are reasons for many of these laws and reasons why states signed them that don't have anything to do with human rights, too. @graydon@dr2chase@cstross@isaackuo
@pettter@dymaxion@cstross@graydon@dr2chase@isaackuo I don't think either party is fully adhering. I also do think there are glaring violation to them by Russia that are orders of magnitude worse than Ukraine's. Then there's also the Genocide Convention that is also clearly being violated by Russia and not by Ukraine. So this is a "Both sides are not equally bad" situation.
@isaackuo Yeah. I feel like the ground environment is a lot more complex, though — like, yes, if you're using this in the context of an initial push on a trench line or defense against the same, sure, but once you're in the middle of breaking through or reacting to contact in a disordered environment, or in basically any urban context at all, it's going to be a lot messier, especially if you're taking advantage of them as light standoff weapons and running them a couple km out @wonka@graydon@dr2chase@cstross
Homing torpedoes, in contrast, need navigation capabilities to be able to use a kill box. But the idea is quintessentially the same as a mine field - the torpedo will try to kill anything it finds in the kill box (possibly with additional sensor profile parameters), but it will NOT try to kill something outside the kill box.
This system isn't perfect, but it's a starting point.
Yeah, I didn't want to get into the legal and ethical considerations, since there are practical reasons to consider that I felt like diving into.
Now, the thing is ... we don't have to just speculate on autonomous munition fratricide machines. We've had them for some time in the forms of mines and homing torpedoes. And these are still relevant as the heavy use in Ukraine shows.
Basically, it's about defining a kill box. Mines are, of course
@isaackuo Definitely, with sufficient limitations. It feels like a lot of the lethality is down to the precision of fully-intelligent terminal guidance. As a jamming backup it's an on obvious win, but as a primary, it's less clear. @wonka@graydon@dr2chase@cstross
Yeah, there's definitely limitations. It's like using artillery laid mines, although these could probably be reused so that alters the logistical calculations.
I'm just pointing out that it IS possible to employ fully autonomous killbots, in a manner that is already familiar to military users, even without IFF systems.
@isaackuo See up thread — an absolute minimum hardware cost terminal guidance package that e.g. might not even be able to handle a bounding box in a gps-denied environment, because good inertials are expensive and video terrain guidance is expensive at the resolution needed. @graydon@wonka@dr2chase@cstross
I don't really know precisely what you mean by an "artificial biting insect", but Ukraine is already using pretty much the least expensive FPV drones practical. If you want something smaller, it'll be more expensive and have much less range and endurance.
@isaackuo@dymaxion@wonka@dr2chase@cstross The problem is incredibly hard. Rested, trained humans aren't good at it. (and in a conflict like Ukraine, with similar troops, uniforms, and equipment, it's even worse.)
The capability to do "artificial biting insect" is near-term, if it's not poorly-distributed-present.
I expect someone is going to go for what they can build. It's how we got chemical warfare in the Great War; it's at least a chunk of how we got napalm and cluster munitions.
There's this vision that many are fascinated by, of locust swarms of drones sweeping the enemy off the battlefield.
I can understand the appeal to Raytheons and Raytheon wannabes. Selling millions of expensive drones to the US military sounds like a pretty sweet way to rake in megabucks, right?
But I'm more puzzled by how much this idea dazzles ordinary folks. Without full autonomy, massive drone swarms are a C3 non-starter. With it ... ehh ...
@isaackuo I think with r&d time and volume to get the quirks out, tricopters might win — slightly more complex, but less battery (and thus more payload) for the same weight and perf envelope. I don't see much going further afield than that winning soon, though. @graydon@wonka@dr2chase@cstross
How do you know it's not a quadcopter? A quadcopter is extremely simple and cheap. The monospinner is even simpler and cheaper, but it's much slower and less maneuverable.
There are a lot of cheap RC toy drones, including drones with only two props and motors. But these aren't maneuverable enough to attack a target. The quadcopter seems to be the cheapest option that's also maneuverable enough to be used as a guided weapon.
@isaackuo Ukraine's adapting commercial hardware, like early Great War aircraft using rotary engines originally designed for motorcycles. A state-equivalent actor starting today and setting out to make an invader's costs unbearable isn't going to start there.
There's work going into "looks like a bird", "smallest practical flying robot", and so on. "How small and cheap can something be and have a 5% PK against human targets for a day?" isn't a quadcopter.