Death by algorithm: the age of killer robots is closer than you think | Hacker News Reader

Death by algorithm: the age of killer robots is closer than you think

vox.com

11 pointssaravana856 comments

All this hoopla about state level actors and un bans and scavenging the tech, but realistically speaking of any of us really really needed to, we could probably bolt a gun to a drone and use pytorch/tensor flow/whatever to make a reasonably effective LAWS. Sure a country could do it better than me, but you don't need Manhattan project resources to do this. The cats out of the bag. Pandora's box is open. The horse has already bolted

>Experts in machine learning and military technology say it would be technologically straightforward to build robots that make decisions about whom to target and kill without a “human in the loop”

I'd like to know who those experts are because like almost always in these ethical debates about autonomous systems, the first thing I notice is the grandiose claims being made about the technology itself.

It is highly non trivial for humans to make calls about civilians, targets, journalists and so on with high fidelity in war zones.

And even if the classification was possible, what stops bad actors from the trivial countermeasures of handing civilians weapons or using disguise, or even adversarial methods to fool autonomous systems? That's not something any existing technology can cope with, it requires extraordinary human judgement.

Just like with autonomous driving technology I'm almost more afraid of the notorious overselling of dumb machines than I am of ethical debates.

Giving the quote its most charitable reading, it sounds like they mean it's possible technologically to create a machine that:

* identifies a potential target * chooses to kill them * kills them

And so the problem isn't whether those are possible but entirely on the ethical component of choosing who to kill (or building such a system at all in the first place.)

And obviously a portion of that ethics is the adversarial component you mentioned which makes the closed loop choosing more difficult.

Bad actors already use adversarial patterns by mixing up within the civilian population.

This is been to core concept of guerrilla warfare since it was conceived and likely well before it was named.

In fact even outside of war zones bad actors constantly hide in plain sight we just call them criminals rather than terrorists when they aren’t a foreign entity with a political goal (although organized crime can easily fall under the definition of terrorism under certain conditions).

As far as dumb machines go then we already have plenty of systems which already automate much of the killing including target identification and tracking.

We’ve already been over selling dumb machines that make killing easier a human in a conflict zone overloaded with sensor data that makes it easy to hit a target isn’t more likely to make the right decision.

Given the limited window of opportunity and the level of cognitive dissonance any one engage in combat has to develop as coping mechanism the idea that humans are inherently capable of making a moral decision under these conditions is pretty laughable to me.

Reminds me of the somewhat famous Slaughterbots video from the Stop Autonomous Weapons initiative: https://youtu.be/9CO6M2HsoIA

It's a pretty crappy video that is doing nothing but fearmongering.

Did you know that there is a knife in every kitchen? So why isn't there a knife murder epidemic? Because murder has to be "profitable" in some way. As long as there are better ways to obtain that "profit" violence will generally not happen on a large scale.

Nowadays the bigger problem is that countries like the USA are willing to kill journalists and other civilians in foreign countries Drone or rifle does not matter. They were innocent either way.

Reply on news.ycombinator.com