Revolution might not be televised, but apparently genocide will be mathwashed.
From the team behind "we kill people based on metadata", a new exciting project: "we kill people based on… we don't even know what, it's a 2bln parameters black box that we have no real way of actually understanding."
Tech had overly stretched credulity with "The Russians gamed our engagement algorithms" and "Republicans manipulated our algorithmic amplification controls for two election cycles".
They had to come up with something new.
"Our Saudi investors are funding our AI research & the shareholders are demanding layoffs in the content moderation teams."
@rysiek AI weapons have been autonomously deciding to kill people for quite some time now.
However, it's mostly been mundane & retail - with such things as the unappealable withdrawal of benefits and support from vulnerable people (decisions made by your cost-effective friendly neighbourhood ML engine).
@rysiek
So much cruelty has historically been done by putting a middle man between the victim and the person giving the order (usually either another victim or a violent psychopath). All for that sweet sweet moral distance.
I'm not sure if it is good or bad the military industrial complex is seemingly running out of human middle men.
@rysiek Combine autonomous lethal drones with authoritarian far right governments that are rising all over the place and these things can easily be used domestically perhaps even more effectively than they could be used in a warzone.
@rysiek I have a regular donation to the Campaign to Stop Killer Robots, but since this stuff seems to just keep inexorably advancing, I'm not sure it's worth anything... https://www.stopkillerrobots.org/
"Will I be killed?"
"I'm sorry, but my dataset goes only as far as 2021. According to the data I have, the answer is :yes, but there may be more resources on the Internet. Was my response helpful?
👍 👎
"
@edavies@rysiek I suspect that in the same way we've said "the driver still has to pay attention, it's just an assistive technology", there will be "moral crumple zones" that push the decision-making down on some poor enlisted person who has "rules of engagement", contradicted by all sorts of workplace social pressure.
@rysiek I spent the last few months working on dataset poisoning and other adversarial attacks and I am scared shitless of how cheap and easy might be to get something like this to malfunction
@rysiek super encouraging, given that initially they were testing this crap with their training data back in Snowden times. This will definitely go well. /s
What the hell for? AI drones are a weapon of mass destruction. Swarms of them can quickly kill everyone in a city. Deploying them is equivalent to dropping a nuke, and will be responded to with nukes. We already have nukes; we don't need yet another WMD.
At least nukes follow orders. AI drones could just as easily decide to kill you as your enemy.
The counter to small drones is jamming, and the counter to jamming is autonomy.
And TBT it is potentially far more ethical: an autonomous drone that mis-identifies the target kills a couple of unintended people. But in many cases a 1000lb JDAM that hits the target dead on will kill a lot more unintended people.
@rysiek this is exactly what you never want to happen. I even don't trust LLM as my accountant. How can anybody with a physical biological mind think this is a good idea?
@AlliFlowers@rysiek Voyager hasan episode where two sets of Cyborgs designed for war that had long exterminated their nation state humans found by Voyager. Belanna reactivated one
@rysiek If there's anything left on your bucket list, get a wiggle on and get it done.
Fry bread. It's high time I made fry bread.
Yeah, my ambitions run large.
Add comment