Hacker News Reader

Comment by dgudkov | Hacker News Reader

The moral dilemma of automated killings is not new. To remind you, lethal devices that kill people autonomously without operator's direct control have long existed. They have killed probably millions of people and keep killing every day. They are called landmines. Landmines are not as cool as flying drones for sure, but they are way more dangerous because they are dirt cheap, produced en masse, and widely available to unsophisticated combatants. If you're terrified by autonomous drones, maybe you should be even more terrified by landmines despite they are not as intellectually stimulating as drones.

View on HN

Landmines have a simple binary "decision" process: is the target on them or not. So in itself that's not a decision, it's a physical phenomenon: you close a switch, you trigger an event.

Drones have several layers of added "decision making" that are out of human control making them far more dangerous. They get to autonomously decide, target, and kill based on many factors, using algorithms that we cannot trust or predict 100%. They also does it from afar. I know you can't trust a human 100% either but it's still the best we've got.

It's the difference between walking in a bog and encountering a lion. Both might kill you but they're in no way similar even in principle.

Autonomous killer drones are just automating murder and terrorism. Taking horrific weapons and making them worse by giving them better range, better targeting, and a brain that's smart enough to kill but stupid enough to do it as dangerously as possible.

Landmines are worse, because they don't even know, the war has ended.

Wouldn't it be even worse if a landmine could just come to you at any time and "decide" to take out it's target and everything that surrounds it? Maybe the algorithm was tripped and you're a "good enough" match, wherever you may be, even in your safe zone. There's no "keep out" sign, no safety because someone else already checked or if you keep to the main road, no removing the threat. They need no preparation, they will actively seek their target from as far away as possible. They can cover more ground than mines ever could and they won't just take out the target but also the ones around it. And all this under the control of an algorithm.

This is the terror element. People will have to worry about getting blown up anytime, anywhere. That's why we keep making weapons with more range and better targeting, because waiting for the target to come to the weapon just wasn't horrific or effective enough.

Sure, landmines are really bad but we're inching closer to things that surpass landmine horrors in every way and makes them seem like the lesser evil. Because if there's one thing that's bad, it's taking an already bad thing and augmenting it. And that should really scare you.

If autonomous drones really become that much of a threat then counter-measures will quickly be developed. We already have radar or even just auditory systems that can target drones, and something like a laser might be enough to take them out.

That's fine for the military, how does that protect joe random citizens?

While Joe random citizen may have illusion of grandeur, they probably are not important enough to be targeted by an advanced drone attack.

If someone really wants a hit on joe random citizen, why would they choose one of these drones over the traditional method? Much harder for a weaponized drone hit to look accidental.

Joe random citizen has a much harder life in places at the wrong end of USA's, Russia's, Turkey's, Israel's, China's, and others' interests, countries who can afford to deploy such technology at scale or have a propensity for projecting their power.

> why would they choose one of these drones over the traditional method

Why does anyone use automation or AI for anything? To have dozens or even hundreds of drones constantly in the air with close to no supervision from drone operators. When you're going after an enemy false positives and collateral damage matter very little, if at all.

> Much harder for a weaponized drone hit to look accidental

Nobody bothers making deaths look accidental in the places these drones will be used. If reality showed anything it's that they barely bother to argue when they bomb schools and hospitals beyond the boilerplate "we're fighting terrorism".

This kind of tech makes a bad situation even worse for many people in some parts of the world. But you'll start worrying about it only if/when someone decides to use it somewhere in a civilized country, closer to home. I'm sure so many people will suddenly find plenty of arguments against the autonomous killing machines then.

USS The Latest will have all the countermeasures it needs. What about $random_middle_eastern_wedding_guests? Countermeasures aren't available to anyone but the military. But in some parts of the word everybody can be a target, especially if some superpower needs to maintain its sphere of influence, or they happen to be at war with some genocidal regime.

Drones have been used quite successfully for many decades now and clearly haven't developed any countermeasures for what most of their targets are concerned, "quickly" or otherwise. This is why we still read about drones bombing people out of existence, whether the correct target or not.

An autonomous drone just outsources the usual killing from the human operator to a script, making it even more likely to make a mistake and also apply the desired effect: terror.

Or to flip the (pithy) argument: "Autonomous drones are worse, because they can declare war."

historically yea, but not for long

What about guided missiles? Once the instruction to kill is given, they hone into the target autonomously.

I guess missiles are given an order for a specific aircraft.

The unspoken issue with these drones is the fact that they’re widely available and made up of commercial-off-the-shelf components. A guided missile requires expertise in chemistry, materials science, and a lot of expensive experimentation. You can’t fly under a government’s radar with them. A drone however, is accessible to every citizen, and the software that runs them is widely available and accessible.

In nuclear weapons it’s the difficulty in sourcing the nuclear material that prevents widespread proliferation. Imagine if the core could be made of something like iron, everyone would have nukes and our civilization would not survive.

Automated targeted killing devices have not existed before. That distinction is very important. The difference between "randomly kill some person in the future who goes through this area" and "upload picture to kill target" is HUGE in practice.

That still isn't true. Fire and forget ranged weapons with heuristic targeting capabilities have also long existed. Heat-seeking missiles are the most obvious example, but target signatures have been more sophisticated than "kill the hottest thing reasonably close to a straight line from initial launch direction" for a while.

The line that has never yet been crossed is the invention of weapons that decide when to launch themselves, at least in ranged indirect fire. Things like CIWS are arguably both autonomous and self-initiate firing.

That's not nearly the same thing. The capability to kill based on the identity of the target is transformative.

A slightly separate issue for landmines is the tendency to main but not necessarily kill. It might seem morbid, a lot of the logic revolved around whether a weapon caused unnecessary pain compared to quick death. Dum Dum bullets for example banned for military use for their tendency to cause more pain than needed (sadly only applies to military). I believe before that there was similar debate on a sword type used by cavalry.

Coupled with its indiscriminate nature, and denial of swaths of lands, and it's an cruel atrocity of a weapon.

Automated weapons I feel fall under something else.

>they are dirt cheap, produced en masse, and widely available to unsophisticated combatants.

Quadcopters are not far from having all of these qualities either.

A lot of people are terrified of landmines. That's why there are treaties around the use of landmines, such as the Ottawa Treaty [0], which ban anti-personnel landmines. 164 countries have signed that treaty.

I don't like this "whataboutism".

In addition, while landmines are terrifying and deadly, they stay in one place. People generally don't need to fear them when going about their day. Drones on the other hand are mobile, they could fly into a city and kill people. They could fly on your property and kill you.

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

Unless, you know, they are on your property, ie your field, on the side of the road to school for your children. Interesting to see 1st world perspective vs 3rd world, where the former looks at it as strictly 'other's problem'.

This is still whataboutism. I don't see anyone arguing that landmines are trivial or harmless.

The point is that they are entirely different concepts, and since this a discussion about drones, landmines have about much place in this thread as Tse-Tse flies, SUVs and low-head dams.

That's a reasonable take when the comparison is between a landmine and a present-day autonomous drone, but the latter technology is in its infancy. A large part of the fear involves the future. As scary as inadvertently stepping on a landmine is, it's a whole different level of terror than running for your life from some relentless Terminator-style robot.

There are laws or conventions about landmines.

The concept of killing ai is not well defined.

Reply on news.ycombinator.com