Comment by metalsiliconYT
2 hours ago
Respect for sticking to your beliefs. Just out of curiosity though why do people not want smart AI weapons? I would much rather have an onboard AI that can discriminate between unarmed civilians and military assets, seems irresponsible to not... Is a dumb sea mine that blows up everything somehow better than a smart sea mine that knows to not blow up sometimes?
"You're absolutely right, that wasn't a military target—it was actually a girls school. It won't happen again!"
I realize that's not a great argument and was definitely tongue-in-cheek, but given there's still a lot of debate about the accuracy of AI for far more mundane tasks, my personal perspective is that until we have LLMs and such that are truly, demonstrably far more accurate than humans, with true reasoning and judgement capabilities, they don't belong where lives are at stake.
I wouldn't want an LLM-underpinned machine running anesthesia during a surgery; why would I want an LLM-underpinned military apparatus that is deciding the lives of far more? I wouldn't, not in their current state.
In a hypothetical future where we truly trust incredibly smart AIs or LLMs or whatever "smart" technology it is for driving weaponry, okay - if it's truly necessary; I abhor war and the death and destruction wrought by it.
In my mind, though - even if we get to that future where there's some vastly superior technology to the LLMs we have today, which can judge and reason, then I'll have a bunch of other questions, like understanding the motivations of said technology, because I suspect it'll be something much closer to AGI, and that opens a whole separate can of philosophical worms.
Not only that but an AI could follow any orders it is given (whether it hallucinates that's a different story) whereas humans may oppose. The friction of orders having to go through a chain of command through its completion is a feature not a bug and that adds responsibility. AI can take no responsibility by itself and with 'moving fast break things smart weapons' that responsibility can be further shoved under the rug.
> "You're absolutely right, that wasn't a military target—it was actually a girls school. It won't happen again!"
What if it was the target though? AI may be more capable than we're giving it credit for (especially the AI accessible to the US and other governments). The attack on the girls' school coincided with Purim and I don't believe it was a mistake. I think it was the opening salvo by a radically religious Zionists (Christian and Jewish).
I don't think the biggest problem with AI weapons is that they make "mistakes", I think the biggest problem is that they allow people who want to kill civilians the ability to accurately do just that.
So you would prefer weapons that cannot reason at all?
> "You're absolutely right, that wasn't a military target—it was actually a girls school. It won't happen again!"
Most likely this event happened due to a bad targeting system that wasn't smart enough, if it had a better llm underpinning it (assuming it had any) maybe those lives would have been saved. More reason for more smart people like the author to work on these systems.
The bottom line is a human should have to fire the gun and answer for it
> So you would prefer weapons that cannot reason at all?
Um, yes? It's bad enough humans are murdering each other. At least a human can in theory be held accountable for pulling the trigger. The last thing we need is an unaccountable ralph loop reasoning about which schools and churches to bomb every time it wakes up.
> if it had a better llm underpinning it
Ah yes, the "LLMs are intelligent, you're just not using the newest fanciest model" fallacy. This time with innocent lives on the line. If only we used ChatGPT-8.9 instead of 8.8, those poor kids would still be alive today.
President Trump reveals the US Space Force has CAMERAS that can read the NAME TAG "Mohammad" trying to reach any nuclear dust in Iran
"We say: 'Mohammad something is there with shovels.'"
"We have cameras that can read the badge of the person. 'Mohammed Something.'"
"WE'RE WATCHING. If anybody goes there, THEY GET BLOWN UP."
"Eventually, we'll take it."
They have also admitted they saw the children’s flower chalk drawings too. And they double tapped.
Mainly I don't want them because I don't want a machine instantly making life or death judgements. It might be fine if it mostly informs a human who ultimately pulls the trigger, but having that failsafe is important in keeping machines from going on killing sprees.
I also want people to be held accountable when they do unjustified killings. AI weapons make it FAR too easy to simply pass off a killing as a "woopsie doodle." It's just not acceptable to say "The algorithm made a mistake, version 23 will do better".
I don't have a problem with the AI providing additional information to it's user, but when that's incorporated into a weapon it's a short distance from that to completely automating the killing.
That's why I'm completely against AI weapons.
Responsibility will always have to lie with the user, but if the weapons are more advanced and safer then why is that bad? I like that autonomous vehicles like Waymo are 10x safer then human drivers, even if a 'Machine' is making decisions.
Because they aren't "safer" for the folks getting killed.
Historically, the folks doing the murdering just don't care about the folks they kill, so "safer for the killers" isn't a win for most of us.
"More precision" isn't about killing less folks, it's about making it safer and easier for folks who kill folks to do that work.
So those of us who dislike killing don't like these tools because we consider them to be immoral.
1 reply →
> Responsibility will always have to lie with the user
Being fully autonomous makes it hard to identify exactly who that user is and is easy to dilute responsibility. Perhaps someone was added to a kill list by mistake. Maybe some internet hi-jinks tricked an AI into falsely identifying someone as a kill target. Perhaps it's the case that someone was in the 5% of a 95% confidence of identification. I'm not a fan of putting killing into the hands of something known to get things wrong 1 in 20 (95%), or 1 in 100 (99%), or 1 in 1000 (99.9%).
> but if the weapons are more advanced and safer then why is that bad?
It's yet to be proven that they are "safer" as they become more advanced. There's also a question of "safer to who". It's technically "safer" for a soldier to shoot first and ask questions later, it's obviously not safer for the villagers.
False identification, which is an absolute part of AI, doesn't make these weapons safer for anyone.
> I like that autonomous vehicles like Waymo are 10x safer then human drivers, even if a 'Machine' is making decisions.
Waymo has the reverse bias. If anyone dies as a result of waymo it's gone horribly wrong.
AI weapons are designed exactly to kill, if they don't kill when they should something has gone wrong.
>Just out of curiosity though why do people not want smart AI weapons?
When the decision to kill another human being is made that should be in the hands of a directly accountable other human being, not an unaccountable machine developed in the basement of a private corporation.
And mines, both dumb and smart, in particular anti-personell mines are banned by the Ottawa treaty ratified by 162 countries. It's exactly the autonomous and fundamentally uncontrollable nature of mines, not just that they're dumb, that has produced countless of casualties long after wars were over. Can you tell me that millions of autonomous loitering munitions are not going to end up exactly like those mines still blowing legs off people decades after conflicts are over? And who is responsible then?
The mines got banned because they were dumb...
no they were banned because they're persistent. That's also the US position despite not being signee to the treaty. Making autonomous weapons intelligent only makes the issue worse because at least expertise was a limiting factor for most rogue actors. Do you know what will happen to millions of autonomous drones when the Russia Ukraine war draws down? Which hands they will end up in? You've created the capacity to kill people 2k kilometers away for a thousand bucks on a chip.
And when those inevitably backfire and start blowing innocent people up, are the CEOs going to prison, the politicians who ordered their production originally? No, nobody is going to claim they're responsible, and that's why it's insane to build a weapon that does not have a name, time, place and direct order associated with it.
Even the US military is already incapable of taking responsibility of blowing a school with 150 kids up. You want to hand these people systems that are unaccountable by definition?
They aren’t training smart AI weapons to discriminate between civilians and soldiers. Even so, what happens when the soldiers are disguised as civilians? Or when the enemy forces civilians to serve or be a decoy?
Because it's going to be used in any possible way, and certainly not for the good of the people, as imply in your "naive" false dichotomy.