Uh, pretty sure it’s already happened, if by “robot” you mean programmed (but not necessarily independently mobile) automaton / machine, by “autonomous” you mean not under the direct and realtime control of a human operator, and by “deliberately” you mean ML came up with a > 0.5 certainty that the target met its targeting criteria.
Doubt a robot will ever actually deliberate, but that’s more of a philosophical issue.
I took the title to mean that human commanders will deliberately choose to deploy an autonomous robot configured to kill a person or persons (as opposed to a robot's killing a person, but the death was unwanted and unforeseen by the commanders).
Yeah, my point is that’s kinda already been done, if we include modern “smart” landmines and loitering munitions. Honestly, I think this threshold is arguably behind us, but I’d also say if it isn’t it will be in a lot less than a decade.
My money is on automated AA systems near some inland border where commercial flights just so happen to never go nabbing a crop duster, surveying aircraft or SAR/med chopper because some comedy of errors resulted in the device getting put on way too hair trigger of a setting.
You'd know exactly how it was going to happen if you could review every line of code, every comment, and every bit (byte) of data involved, and make sure it was meaningful.
So you could precisely pinpoint the exact data path that would carry out such a deed, and how it got that way. And be able to follow the trail of bits throughout the entire chain-of-command and arrive at the root cause quite logically.
Oh wait a minute . . . I was thinking about an accidental killing, my bad.
For a deliberate killing you don't need any of that.
For military drones, yes, this will certainly happen if it hasn't already.
I'd love to see some predictions on manufacturing robots intentionally killing someone for the greater good in a sort of Trolley Problem [1]. The theoretical potential of AI safety protocols getting misaligned and a robot deciding to sacrifice a human worker to save multiple lives.
In 2001 an S200 missile fired at a drone during a military exercise missed the drone, saw an airliner 160 miles further away and went and took that out. Not sure if that counts?
23 comments
[ 12.7 ms ] story [ 1296 ms ] threadDoubt a robot will ever actually deliberate, but that’s more of a philosophical issue.
I'm not so sure, with the technology they're "shooting for".
I thought this said it all:
>They can differentiate between friendly, civilian and enemy, decide to engage or alert based on target type, and even vary their effects.
Obviouly they're only going to be designing, building, and deploying "nice bombs".
So you could precisely pinpoint the exact data path that would carry out such a deed, and how it got that way. And be able to follow the trail of bits throughout the entire chain-of-command and arrive at the root cause quite logically.
Oh wait a minute . . . I was thinking about an accidental killing, my bad.
For a deliberate killing you don't need any of that.
I'd love to see some predictions on manufacturing robots intentionally killing someone for the greater good in a sort of Trolley Problem [1]. The theoretical potential of AI safety protocols getting misaligned and a robot deciding to sacrifice a human worker to save multiple lives.
[1] https://en.wikipedia.org/wiki/Trolley_problem
https://asimov.fandom.com/wiki/Robotic_Civil_Wars
No one actually told it to go for the airliner but it kind of took the initiative. (https://en.wikipedia.org/wiki/Siberia_Airlines_Flight_1812)
Just a question.