There’s obviously a limit to how much detail you can fit into a headline but just to be clear this is using pattern matching across various sensors to develop a list of things matching a particular signature which is then sent to a human for review long before anyone even thinks about pulling and triggers.
If anyone is looking for a serious dive into this topic I’d highly recommend starting here https://youtu.be/1-0L5Wv86fQ?
How much less care will be taken by the human reviewers knowing that an AI picked the target? There is a bias there in the back of a mind and when you get into a stressful situation you are more likely to trust that the "system" didn't make a mistake.
How many times have you pushed to production skipping the final review because all tests passed anyway only for it to fail?
Yeah different industries need different approaches. For services which serve North America, you’d preferably have a capable team in Australia, or similar, so you can push changes during their biz hours but our off hours.
> How much less care will be taken by the human reviewers knowing that an AI picked the target?
it would depend on how the liability flows when the target is wrong. Will the AI cover the ass of the previous person who would've had to vet it? Or will the ultimate responsibility and therefore, liability, lay with a person?
I agree that it’s not a great analogy but would caution that some of the same concerns about complacency apply. If this tool produces some decent looking output, it’s not hard to imagine a mistake being okayed because it looked good similar to someone hitting “LGTM” on Friday afternoon because it’s been fine the last 500 times OR miss something else because they were assuming that the AI would flag the important areas.
I know that people in the military are very familiar with problems like that so I think the main concern here is making sure that vendors don’t successfully oversell their products’ capabilities.
At the very least, it would increase the risk by expanding usage. I think there’s an additional angle, however: the better an automated system like this gets, the more people assume it’ll always work. Tesla drivers get in FSD crashes they’d never get in with a basic cruise control because it’s good enough to handle enough road situations that they assume it can handle all of them, despite periodic fatal reminders that it isn’t.
I don't know the technical details behind this software. But I strongly suspect this is firmly in the camp of "machine learning" rather than "AI." Big Tech's intentional confusion between those terms has caused an enormous amount of mischief over the last 10 years.
If the military had said "statistical inference using modern Big Data methods" that would have the following advantages:
- honesty
- officers' expectations would be more tempered and reasonable
- public outrage would be more tempered and reasonable
Though to be fair this terminology would have the following disadvantages:
- military tech grifters have their BS more easily called out
- AI tech grifters have to look in the mirror and accept this is what their algorithm is
- AI journalism grifters get less clicks
- Hacker News conversations indulge less navel-gazing speculation about Skynet
>If the military had said "statistical inference using modern Big Data methods"
If this is what I saw the military use I would be upset at the blatant attempt at being misleading. In my mind I would convert it back to AI. It's equivalent to saying "I didn't drive drunk, I got here using carnot-like cycles after injesting social communication enhancing drugs".
If I were reviewing AI written code that passes all tests vs reviewing human written code that passes all tests, I think I’d trust the human written code more.
I knew a guy who was in charge of approving strikes in iraq. His job was assessing the second order impacts of a strike beyond whether or not it was targeting what they believed.
I'm not saying iraq wasn't a shit show, but for what it's worth, this stuff is very carefully reviewed and is not done quickly.
This is tragic but it's noteworthy because it is NOT the norm. Unlike, say, what's going on in Ukraine with frequent missile strikes into arbitrary civilian areas.
This is a serious failure of controls that warranted direct attention and the apology of the president. imo this is moreso evidence that the US takes its controls seriously rather than frivolously.
Both of those things are lies though. Per the British medical journal Lancet, approximately 600000 civilians were killed in Iraq. Per the UN estimates fewer than 11k civilians were killed so far in Ukraine.
I don't really care enough to get into this, so I won't respond further but that study has largely been debunked. It uses both flawed methodology and appears to have outright fabricated data regarding death certificates. Further, it is grossly misleading to imply that the 600k killed were the result of reckless US military strikes.
Wikipedia, really? Ok, let’s set that one aside for the moment. What are your thoughts on this here gem from the first Iraq war: https://m.youtube.com/watch?v=KP1OAD9jSaI Is that “humane” or “proportional”?
> This is tragic but it's noteworthy because it is NOT the norm. Unlike, say, what's going on in Ukraine with frequent missile strikes into arbitrary civilian areas.
Let's not dismiss the propaganda factor here.
No one caring about their livelihood in the west will ever be digging too deep into US/NATO totally-not-war-crimes-just-collateral-shit-happens-you-know, as well as no one will be arguing Russia's case the same way Ukraine's attacks over Belgorod Region and office buildings in Moscow are excused with "this missile/drone was on its way to a military target, but was shot down over civilians heads by air defense".
Assange's case is a very illustrative example what happens when something gets loud enough to actually matter and disrupt the image of righteousness.
Everyone knows the US and NATO have done horrible shit, but unless you have a child's view of the world it's incredibly obvious that the US and NATO take incredible steps to minimize civilian casualties. Much more than non-NATO aligned military powers.
It's good to learn about the past, but it doesn't actually tell you anything about how modern western military operations actually work. You'll always be able to cherry pick random platoons doing heinous shit because humans are still humans. Just because you are from the west does not mean you are better than anyone else.
Is this how people excuse away the genocide currently being committed by Israel with the support of the United States? The US is the only reason this is still going and children are currently starving to death.
As long it's for the greater good its ok right? /s
When the US is the only one holding up their hand at the UN against an immediate ceasefire the last 3 times it came up, it is quit difficult to make this claim.
> When the US is the only one holding up their hand at the UN against an immediate ceasefire the last 3 times it came up
The only one on the SC. Lots of countries have violated UN resolutions, including practically every member of the SC. This is far from evidence of being the sole reason for anything.
In addition to what JumpCrisscross said, you seem to assume that, if the UN passed a resolution for an immediate ceasefire, that fire would actually cease. I see absolutely no reason for that assumption.
If the resolution has no effect why would the US risk being seen as a genocide supporting nation by vetoing every resolution? The US is also risking loosing the veto as many are now considering changing how the SC works.
The SCs job is to secure global security and peace. The US is doing the absolute opposite.
Then when Israel continues their genocide you strip them of the munitions we supplied. If they continue then a global force can move into Israel and enforce a no fly zone and prevent further bombings.
If we don’t even take the first step of trying to stop it we are complicit and the primary reason it is ongoing.
But you're getting off track here. We were discussing the total package of US support for Israel as a driving factor in the current, grotesque horror we are all witnessing.
And it is the main driving factor. If that support were to vanish, Israel's expansionist adventures in Gaza and the West Bank would grind to a halt in very short order indeed.
> it's incredibly obvious that the US and NATO take incredible steps to minimize civilian casualties
And to substantiate that, look at the ratio of military to civilian deaths. In modern warfare, America’s numbers are incredibly low. Our problem isn’t that we fight brutally, but that we get pulled into fights a lot.
Bullshit, the Unites States just legalize/justify their brutality. Torture becomes enhanced interrogation technique, Assassination List is the Disposition Matrix etc. etc.
Which you know about because it became subject to protests and litigation in America and open Congressional scrutiny. (It remains a travesty there has been no legal action in its respect.)
I'm not saying America doesn't do terrible things. Just that the rate at which it does so is dramatically lower than others. If you had to face military action from a foreign power, America is the best of the lot. Of course, that doesn't change that you'd rather not face military action from any foreign power.
We know it happened because of leaks that would get you thrown into prison now, and no one in charge was held accountable. So not sure what you’re trying to get at there.
As for your second point, your last statement is what matters. The us has engaged in more foreign military actions than any other nation since WW2. Who cares if they mask the brutality behind terms like precision strikes and collateral damage. We do terrible things and we do them frequently, usually for no benefit to anyone other than a select few. Everything you said is just trying to rationalize the fact we are the most aggressive country on the earth.
I served as a strike cell attorney in Baghdad during the Daesh fight (Operation Inherent Resolve).
Specific criteria are classified, but the U.S. military systematizes the four LOAC principles: military necessity, distinction, proportionality, no unnecessary suffering.
Positive identification goes only to the first two principles, and—while I don't know specifics—the DoD's "human in the loop not on the loop" AI philosophy makes me think that AI won't satisfy PID by itself. In addition to that, all munitions targets would have to meet CDE in accordance with CJCSI 3160.01. In my experience, there is typically a CDE expert as well as a lawyer involved in every targeting decision.
If you critically examine the source that provided you this number and follow the rabbit hole, you will probably learn that it is wildly pessimistic AND not telling you what you think it’s telling you. E.g. these estimates often include estimated casualties from sectarian violence that was merely uncorked by the Iraq war. You are clearly not taking pains to prevent the casual reader from assuming that 600k (already a specious number) were all directly killed at the hands of US soldiers.
> they're literally worse than Stalin in this regard.
> and accurate body counts.
Ironically, not really, the numbers people typically refer to were calculated by extrapolating birthrate dynamics, not an actual accurate body count.
Applying the same method to calculate the victims of the Great Depression will yield about 7 mil dead. I remember some Americans being real pissed when some Russian historian did this kind trolling, mirroring Stalin-related western propaganda.
Yes as your comment and lawyers comment below show there was a veneer of “we are war professionals, our war is legal and just” and I was listening to every patriotic song that came out celebrating every bomb. My highschool had some highlights of iraq being bombed to the tune of “let the bodies hit the floor.”
At the end of the day war is a racket, the prescence of the lawyer detailing what he did below should only raise everyone’s BS meter to “overload.”
I’m not sure if it was a million we killed or 600k or maybe less but the defense of “many of those weren’t us, it’s was subsequent totally unrelated non us involved” is also a very lawyerly statement. Like, we only lit this one tree on fire maaaannn, we didn’t light the whole damn neighborhood!
Let us not forget the Wikileaks either as a self purported hacker community. 12 unarmed men gleefully murdered by an Apache helicopter (including 2 ap reporters). [1]
Let’s also not forget Madeleine Albright defending the death of 1/2million Iraq children in 1996. When asked by 60 Minutes reporter Lesley Stahl "We have heard that half a million children have died. I mean, that's more children than died in Hiroshima. And, you know, is the price worth it?", Albright replied "I think this is a very hard choice, but the price - we think the price is worth it." [2]
For what it’s worth, I am a US patriot, I just don’t excuse our BS behavior as I’ve been sold the whole “we are very careful, we have lawyers and 2ndary explosion analysts” line one too many times.
The price is always worth it to certain people and industries. I’m not ragging on the lawyer below, I hope he was able to do some good. But it’s all bullshit, from start/middle/end (Afghanistan withdrawal, you should have seen the look on vets faces as they realized every sacrifice, buddy lost, person killed was completely and totally pointless, but don’t worry we have a national suicide call line)
This is so self aggrandizing for the AI community. The military has tanks, guns, ships, cruse missiles, satellites, and nuclear weapons capable of ending life on earth but now we're supposed to tremble at the might of AI that is doing the equivalent of "hotdog or not"?
If the ai can get hotdog wrong sometimes, shouldn't you be extremely concerned that the military is using to kill people and destroy things. I feel like "hospital or not" is a much more dangerous game.
The military already plays hot dog or not except they use a 19 year old with a grainy FLIR display to make that judgement today, with the end result being a ridiculous amount of civilian casualties in all modern conflicts. The AI model here can’t come soon enough imo.
Rafael has been demoing SPICE (Smart, Precise Impact, Cost-Effective) bombs since 2019^1
> The newly-unveiled ATR feature is a technological breakthrough, enabling SPICE-250 to effectively learn the specific target characteristics ahead of the strike, using advanced AI and deep-learning technologies.
"Moore said. The US Central Command has also tried to run an AI recommendation engine to see if it could suggest the best weapons to use in operations and create attack plans. The technology, however, "frequently fell short.""
Are they using LLM for this as well? Then of coruse it's going to be hard. I don't understand why many working with AI refuses to use an AI that's good for this. CSP solvers! There are some papers about using them for drone making flight paths.
71 comments
[ 5.4 ms ] story [ 136 ms ] threadIf anyone is looking for a serious dive into this topic I’d highly recommend starting here https://youtu.be/1-0L5Wv86fQ?
How many times have you pushed to production skipping the final review because all tests passed anyway only for it to fail?
edit: the point is you choose when to deploy, eg shoot.
If it breaks on my work hours, np, i will revert the deployment. Good luck getting me to revert anything on a weekend.
it would depend on how the liability flows when the target is wrong. Will the AI cover the ass of the previous person who would've had to vet it? Or will the ultimate responsibility and therefore, liability, lay with a person?
This is the US.
For example the Captain USS Vincennes got promoted after he ordered the shooting down of a commercial airliner by mistake.
What rank was Captain Rogers promoted to after the incident?
I know that people in the military are very familiar with problems like that so I think the main concern here is making sure that vendors don’t successfully oversell their products’ capabilities.
If the military had said "statistical inference using modern Big Data methods" that would have the following advantages:
Though to be fair this terminology would have the following disadvantages:>If the military had said "statistical inference using modern Big Data methods"
If this is what I saw the military use I would be upset at the blatant attempt at being misleading. In my mind I would convert it back to AI. It's equivalent to saying "I didn't drive drunk, I got here using carnot-like cycles after injesting social communication enhancing drugs".
My general experience is that AI is just beyond what somebody understands and when they understand how it works they just use a more specific wording.
I'm not saying iraq wasn't a shit show, but for what it's worth, this stuff is very carefully reviewed and is not done quickly.
[1] https://en.wikipedia.org/wiki/Kunduz_hospital_airstrike
This is a serious failure of controls that warranted direct attention and the apology of the president. imo this is moreso evidence that the US takes its controls seriously rather than frivolously.
https://en.wikipedia.org/wiki/Lancet_surveys_of_Iraq_War_cas...
I'm not interested in following your link.
Let's not dismiss the propaganda factor here.
No one caring about their livelihood in the west will ever be digging too deep into US/NATO totally-not-war-crimes-just-collateral-shit-happens-you-know, as well as no one will be arguing Russia's case the same way Ukraine's attacks over Belgorod Region and office buildings in Moscow are excused with "this missile/drone was on its way to a military target, but was shot down over civilians heads by air defense".
Assange's case is a very illustrative example what happens when something gets loud enough to actually matter and disrupt the image of righteousness.
It's good to learn about the past, but it doesn't actually tell you anything about how modern western military operations actually work. You'll always be able to cherry pick random platoons doing heinous shit because humans are still humans. Just because you are from the west does not mean you are better than anyone else.
As long it's for the greater good its ok right? /s
America is far from the only reason for the War is Gaza. It’s a major contributor to Israel’s war machine, but it’s also one to Palestinian aid.
The only one on the SC. Lots of countries have violated UN resolutions, including practically every member of the SC. This is far from evidence of being the sole reason for anything.
The SCs job is to secure global security and peace. The US is doing the absolute opposite.
If we don’t even take the first step of trying to stop it we are complicit and the primary reason it is ongoing.
But you're getting off track here. We were discussing the total package of US support for Israel as a driving factor in the current, grotesque horror we are all witnessing.
And it is the main driving factor. If that support were to vanish, Israel's expansionist adventures in Gaza and the West Bank would grind to a halt in very short order indeed.
But it's clearly the most important. To an extent that it dominates all other factors.
And to substantiate that, look at the ratio of military to civilian deaths. In modern warfare, America’s numbers are incredibly low. Our problem isn’t that we fight brutally, but that we get pulled into fights a lot.
Which you know about because it became subject to protests and litigation in America and open Congressional scrutiny. (It remains a travesty there has been no legal action in its respect.)
I'm not saying America doesn't do terrible things. Just that the rate at which it does so is dramatically lower than others. If you had to face military action from a foreign power, America is the best of the lot. Of course, that doesn't change that you'd rather not face military action from any foreign power.
As for your second point, your last statement is what matters. The us has engaged in more foreign military actions than any other nation since WW2. Who cares if they mask the brutality behind terms like precision strikes and collateral damage. We do terrible things and we do them frequently, usually for no benefit to anyone other than a select few. Everything you said is just trying to rationalize the fact we are the most aggressive country on the earth.
Specific criteria are classified, but the U.S. military systematizes the four LOAC principles: military necessity, distinction, proportionality, no unnecessary suffering.
Positive identification goes only to the first two principles, and—while I don't know specifics—the DoD's "human in the loop not on the loop" AI philosophy makes me think that AI won't satisfy PID by itself. In addition to that, all munitions targets would have to meet CDE in accordance with CJCSI 3160.01. In my experience, there is typically a CDE expert as well as a lawyer involved in every targeting decision.
At least he had an appreciation for statistics, and accurate body counts.
I have no idea what you think I'm defending, but whatever.
Ironically, not really, the numbers people typically refer to were calculated by extrapolating birthrate dynamics, not an actual accurate body count.
Applying the same method to calculate the victims of the Great Depression will yield about 7 mil dead. I remember some Americans being real pissed when some Russian historian did this kind trolling, mirroring Stalin-related western propaganda.
At the end of the day war is a racket, the prescence of the lawyer detailing what he did below should only raise everyone’s BS meter to “overload.”
I’m not sure if it was a million we killed or 600k or maybe less but the defense of “many of those weren’t us, it’s was subsequent totally unrelated non us involved” is also a very lawyerly statement. Like, we only lit this one tree on fire maaaannn, we didn’t light the whole damn neighborhood!
Let us not forget the Wikileaks either as a self purported hacker community. 12 unarmed men gleefully murdered by an Apache helicopter (including 2 ap reporters). [1]
Let’s also not forget Madeleine Albright defending the death of 1/2million Iraq children in 1996. When asked by 60 Minutes reporter Lesley Stahl "We have heard that half a million children have died. I mean, that's more children than died in Hiroshima. And, you know, is the price worth it?", Albright replied "I think this is a very hard choice, but the price - we think the price is worth it." [2]
For what it’s worth, I am a US patriot, I just don’t excuse our BS behavior as I’ve been sold the whole “we are very careful, we have lawyers and 2ndary explosion analysts” line one too many times.
The price is always worth it to certain people and industries. I’m not ragging on the lawyer below, I hope he was able to do some good. But it’s all bullshit, from start/middle/end (Afghanistan withdrawal, you should have seen the look on vets faces as they realized every sacrifice, buddy lost, person killed was completely and totally pointless, but don’t worry we have a national suicide call line)
1 https://hackcur.io/collateral-murder-wikileaks-iraq/
2 https://x.com/theserfstv/status/1506706179178725379?s=46
We are saying the US military, as an organization, takes targeting decisions seriously and slowly.
Never. Because like the gate guard in "Zero Dark Thirty" I understand that process only works if you do it all the time.
Alternative: “US military pulls the trigger, implements AI-assisted pattern matching for military target acquisition”
I even kept their “pulls trigger” pun.
More discussion here: https://news.ycombinator.com/item?id=39531348
> The newly-unveiled ATR feature is a technological breakthrough, enabling SPICE-250 to effectively learn the specific target characteristics ahead of the strike, using advanced AI and deep-learning technologies.
1. https://www.rafael.co.il/press/elementor-4174/
Wired: ...
Expired: ...
Are they using LLM for this as well? Then of coruse it's going to be hard. I don't understand why many working with AI refuses to use an AI that's good for this. CSP solvers! There are some papers about using them for drone making flight paths.