The writer Tyler Austin Harper (of The Atlantic, etc.) sent me a thread this morning, asking whether a mistargeting yesterday that killed nearly 150 school children in Iran could have been the result of AI.
Pathocrats love AI because, I believe, they see it as just like themselves. Move fast and break things is the goal. That some things actually do take time, or absolutely should not be broken, is not a concept they have the ability to factor in. They need to be under strong supervision and control, preferable inside a penitentiary.
Objectively, Gen AI is NOT ready for military use... But apparently we're doing it anyway... It's people using AI in ways it's not safe.. and now that Pandora's box is open, we have to be able to understand the consequences. And that's not been given enough energy.
There are many military uses, and AI may be perfectly fine for some of those uses, like planning menus for the mess decks (I was in the Navy, they call it something different in the other services), scheduling shipments of household goods for personnel being assigned to a new duty station, or deciding which recruits (and how many) to send through which branches of A-school training. I think what you mean is that Gen AI is not ready for *combat* use.
General Ulysses S. Grant spent time in his early military career assigned to the task of keeping troops well supplied. He later used that experience and expertise to help defeat the Confederate army by strategically sabotaging supply lines and starving out his opponents, much as Zelenskyy is doing in Russia today. I wouldn't trust any LLM with that job either.
There are other tools/techniques/approaches within AI that can be used to accomplish those tasks. The core problem (IMHO) is the current tendency to default to GenAI based solutions.
Omg. The ai being used in the military(TLP Targeted Location Prediction) is not the same grade as civilian(CGI Consumer Grade Intelligence)
TLP is not meant to be safe it’s meant to kill targets. Collateral damage is accepted by the military in those equations. They don’t fucking care about 150 dead kids.
TLP is absolutely ready for military use and has been since GWOT. Palantir is a 20 year old company.
CGI is about selling you products more efficiently so gdp number goes up.
Yeah, uh, so...if your bot can't figure out how to take out your real target without killing well over 100 civilians, it ain't ready for prime time. 20 years old or not, "military-grade" or not.
Our soldiers did this not a bot. The bot is a location detection algorithm. The soldiers fire the weapons knowing killing civilians. We killed more than 3 million civilians in the Vietnam wars without computers. We took flame throwers to living crying babies.
Accident? The very fact that the attack on Iran contravenes international law, means that no matter how people were/are killed by USIsrael, it's no accident but a war crime.
Look how many US allies are refusing to call it that, including the UK. Just this morning (Sunday, March 1st), the UK's Defence Minister refused to comment on whether this was legal or not, even when confronted with the information that both Canada and Australia condemned the attack as illegal and a possible war crime. Britain remains the US's "poodle" even though the current tariffs under the new rule cannot exceed 15% despite Trump's threats. Starmer needs to grow some balls and stop being so deferential, particularly given his background in international law.
I'm not sure how true but according to one report, the British base in Cyprus was targeted by Iranian missiles, so clearly the UK is, yet again, an active collaborator in USIsraeli crimes.
Britain has clearly stated, for the record, that it had nothing to do with the attack on Iran. If that proves to be a lie, that would put Starmer in a very bad light. It had already angered Trump for denying the use of its airbases to support the US attacks, especially Diego Garcia in the Chagos Islands.
It appears that the UK has joined the US in a "limited capacity" - i.e., the wishy-washy "defensive operations only". At least some European countries, like Spain, have responded with a resounding "NO!" to the US using its bases in Spain. If this war drags on, like Iraq/Afghanistan, then the UK will regret its involvement, spending treasure when it needed to improve its economy and welfare services. The UK PM is still bleating about the "Special Relationship" between the UK and the US. Unfortunately, Trump isn't into such relationships, just "what have you done for me lately?"
The degree of cynicism on the part of the US government is truly breathtaking. Item: No credible threat. Item: Trump encourages an uprising then stands aside while the regime murders thousands, reminiscent of Stalin's response to the Warsaw uprising of 1944. Item: No plan, just kill some people. With allies like ours in the Middle East, it's no wonder we'll have blood on our hands.
AI already caused a lot of the destruction of Gaza, albeit not by accident.
They used models to predict who's Bad and accepted a very wide classifier (maximising recall not precision),
and allowed bombing their houses with no regard for surrounding lives lost (because ethnic cleansing was the goal, as well as killing all the Bad guys).
I can only think of how the AI world was all about using the trolley problem to work out the least problematic choice for a self-driving car to make in a situation where any course of action would kill at least one person, as if the trolley problem were entirely about how many and what type of person were killed and completely ignoring what I had taken to be its most important insight when we covered it in undergraduate psychology - that people assign more moral culpability for a negative outcome the more direct the connection is between the person's actions and the consequences. E.g. letting someone die by doing nothing is less bad than doing an action that causes their death; pushing a person into the path of the trolley is worse than changing the points to send the trolley in their direction; the latter is worse if it is done by pulling a lever at the trackside than if it is done by pressing a button in a control room far away, etc.
Using AI in war is like the apotheosis of that mindset that the more layers of mediation and the smaller the action taken by the human at the start of the chain, the less bad the killing. It's honestly terrifying.
Well I’m sorry you’re out of the loop. Palantir has been at this for 20 years. This is somewhere around their 20,000th deployment of software in war. You must have missed the priors. Not to mention Maduro…
Israel is the U.S. Israel became property of the U.S. at Bretton Woods through the lend/lease act where we dismembered Britains colonies and seized global mode of production. Israel began serving its purpose as a U.S. military installation after the petrodollar formation in 1974. It belongs to the U.S.
All of Israeli technology is Americas property. As is all of Israel’s weapons which they receive for free because they are a U.S. military base.
The people in charge are not interested in accountability; they are only interested in profit and blaming their enemies, real and perceived. Generative AI is their technological soul mate.
I'm reminded that the West (UK and UK) both opted for strategic bombing from WWII onwards. Reading the late Freeman Dyson's scientific memoirs "Disturbing the Universe" (1979), he noted that his analyses, while working for the RAF in WWII, showed that strategic bombing was not effective in destroying Germany's ability to defend against the Allies, but may have increased it. Vietnam showed that such bombing was a failure, too. While advanced technology powers have moved towards targeted bombing (with smart bombs and drones), humans in the loop have made mistakes in identifying targets, e.g., wedding parties. Making such decisions more autonomous with AI will unlikely improve matters, and may make it even worse, as there is no longer guilt by pilots making errors, it is back to the old "Humans make mistakes, but it takes a computer to really f**k up." The more machines make decisions, the more "collateral damage" will occur. (At least the AIs won't laugh with glee at killing medical responders to casualties, or play the Israeli "double tap" executions by aerial arms.)
Illegal collateral damage should have the buck stop at those in charge, like Hegseth at DOD [DOW?] and POTUS. We have already seen how Hegseth tried to distance himself when the circumstances of the first "drug boat" executions of survivors came to light. Trump will do the same if any mistargeted killings in Iran come to light.
I'll emphasize the point made using the dice analogy: anyone who uses generative "AI" to make a recommendation (targeting or otherwise) is 100% personally responsible for that recommendation.
I put the "AI" in "Generative AI" in quotes because it's really a misnomer. It shouldn't be called "AI" because it's not really intelligent. It's just a statistical micro-plagiarizer, mashing together so many things written by real humans that it's not possible to identify everyone it's stealing from.
Certainly Generative "AI" can be useful for searching and brainstorming, provided everything it comes up with is scrupulously double and triple checked.
Suspicious that Hegseth was so adamant that Anthropic had to get on board before Friday end of day. Perhaps he wanted to a non-military scapegoat if/when a targeting error occurred.
"Nearly 150." It started out as 50. Here it's 150. I've seen it as high as 200, recently. Here. Let me help. I say that *ten thousand* tiny, innocent babies were slaughtered by (checks list) Evil Drumpf, Nazi Jews, Demonic AI, White Supremacist Nationalists - pick at least one. Feel free, and don't worry about being wrong. My number has precisely as much confirmatory proof as does 50, 100, 150, 200, whatever.
Pathocrats love AI because, I believe, they see it as just like themselves. Move fast and break things is the goal. That some things actually do take time, or absolutely should not be broken, is not a concept they have the ability to factor in. They need to be under strong supervision and control, preferable inside a penitentiary.
Well said, Joy!
Objectively, Gen AI is NOT ready for military use... But apparently we're doing it anyway... It's people using AI in ways it's not safe.. and now that Pandora's box is open, we have to be able to understand the consequences. And that's not been given enough energy.
"...Gen AI is NOT ready for military use."
There are many military uses, and AI may be perfectly fine for some of those uses, like planning menus for the mess decks (I was in the Navy, they call it something different in the other services), scheduling shipments of household goods for personnel being assigned to a new duty station, or deciding which recruits (and how many) to send through which branches of A-school training. I think what you mean is that Gen AI is not ready for *combat* use.
General Ulysses S. Grant spent time in his early military career assigned to the task of keeping troops well supplied. He later used that experience and expertise to help defeat the Confederate army by strategically sabotaging supply lines and starving out his opponents, much as Zelenskyy is doing in Russia today. I wouldn't trust any LLM with that job either.
There are other tools/techniques/approaches within AI that can be used to accomplish those tasks. The core problem (IMHO) is the current tendency to default to GenAI based solutions.
Good distinction.
Omg. The ai being used in the military(TLP Targeted Location Prediction) is not the same grade as civilian(CGI Consumer Grade Intelligence)
TLP is not meant to be safe it’s meant to kill targets. Collateral damage is accepted by the military in those equations. They don’t fucking care about 150 dead kids.
TLP is absolutely ready for military use and has been since GWOT. Palantir is a 20 year old company.
CGI is about selling you products more efficiently so gdp number goes up.
Now pull the other finger.
Yeah, uh, so...if your bot can't figure out how to take out your real target without killing well over 100 civilians, it ain't ready for prime time. 20 years old or not, "military-grade" or not.
Our soldiers did this not a bot. The bot is a location detection algorithm. The soldiers fire the weapons knowing killing civilians. We killed more than 3 million civilians in the Vietnam wars without computers. We took flame throwers to living crying babies.
Those sound like more reasons to be against it, not to minimize it.
It was Anthropic’s Claude that Palantir’s Maven System used in Iran.
What’s the point of talking about collateral damage as though the tech did it? Is that how stupid we’ve become!?!
How many times can the same people pull the wool over your eyes?
Accident? The very fact that the attack on Iran contravenes international law, means that no matter how people were/are killed by USIsrael, it's no accident but a war crime.
Look how many US allies are refusing to call it that, including the UK. Just this morning (Sunday, March 1st), the UK's Defence Minister refused to comment on whether this was legal or not, even when confronted with the information that both Canada and Australia condemned the attack as illegal and a possible war crime. Britain remains the US's "poodle" even though the current tariffs under the new rule cannot exceed 15% despite Trump's threats. Starmer needs to grow some balls and stop being so deferential, particularly given his background in international law.
I'm not sure how true but according to one report, the British base in Cyprus was targeted by Iranian missiles, so clearly the UK is, yet again, an active collaborator in USIsraeli crimes.
Britain has clearly stated, for the record, that it had nothing to do with the attack on Iran. If that proves to be a lie, that would put Starmer in a very bad light. It had already angered Trump for denying the use of its airbases to support the US attacks, especially Diego Garcia in the Chagos Islands.
It appears that the UK has joined the US in a "limited capacity" - i.e., the wishy-washy "defensive operations only". At least some European countries, like Spain, have responded with a resounding "NO!" to the US using its bases in Spain. If this war drags on, like Iraq/Afghanistan, then the UK will regret its involvement, spending treasure when it needed to improve its economy and welfare services. The UK PM is still bleating about the "Special Relationship" between the UK and the US. Unfortunately, Trump isn't into such relationships, just "what have you done for me lately?"
The degree of cynicism on the part of the US government is truly breathtaking. Item: No credible threat. Item: Trump encourages an uprising then stands aside while the regime murders thousands, reminiscent of Stalin's response to the Warsaw uprising of 1944. Item: No plan, just kill some people. With allies like ours in the Middle East, it's no wonder we'll have blood on our hands.
International law does not exist except in the minds of certain academics. Keep on deluding youselves.
💯🎯
The regime murders thousands? No, CIA/Zionist terrorists murdered thousands in a (failed) attempt at regime change.
AI already caused a lot of the destruction of Gaza, albeit not by accident.
They used models to predict who's Bad and accepted a very wide classifier (maximising recall not precision),
and allowed bombing their houses with no regard for surrounding lives lost (because ethnic cleansing was the goal, as well as killing all the Bad guys).
The humans are always responsible.
Agreed. Deploying classifiers to make predictions makes it easier for a person making an uninformed decision though.
"Computer says he's a terrorist, let's go with it".
I can only think of how the AI world was all about using the trolley problem to work out the least problematic choice for a self-driving car to make in a situation where any course of action would kill at least one person, as if the trolley problem were entirely about how many and what type of person were killed and completely ignoring what I had taken to be its most important insight when we covered it in undergraduate psychology - that people assign more moral culpability for a negative outcome the more direct the connection is between the person's actions and the consequences. E.g. letting someone die by doing nothing is less bad than doing an action that causes their death; pushing a person into the path of the trolley is worse than changing the points to send the trolley in their direction; the latter is worse if it is done by pulling a lever at the trackside than if it is done by pressing a button in a control room far away, etc.
Using AI in war is like the apotheosis of that mindset that the more layers of mediation and the smaller the action taken by the human at the start of the chain, the less bad the killing. It's honestly terrifying.
Bad actors will always look for excuses to evade culpability.
Humans caused that not the tech. What an insane perspective you have.
It's of course the people behind the decisions that are responsible.
That was the first mass targeting of people for assassination using AI that I know of, so I wanted to share it.
Well I’m sorry you’re out of the loop. Palantir has been at this for 20 years. This is somewhere around their 20,000th deployment of software in war. You must have missed the priors. Not to mention Maduro…
In Iraq and Afghanistan?
Was it at a similar scale as Israel's Lavender system? (ten of thousands of targets decided by the computer)
Israel is the U.S. Israel became property of the U.S. at Bretton Woods through the lend/lease act where we dismembered Britains colonies and seized global mode of production. Israel began serving its purpose as a U.S. military installation after the petrodollar formation in 1974. It belongs to the U.S.
All of Israeli technology is Americas property. As is all of Israel’s weapons which they receive for free because they are a U.S. military base.
GWOT is 69 countries
https://www.ojp.gov/ncjrs/virtual-library/abstracts/international-contributions-war-against-terrorism#:~:text=The%20list%20includes%20the%20countries,PDF
You're a bit of a cretin arn't you.
The people in charge are not interested in accountability; they are only interested in profit and blaming their enemies, real and perceived. Generative AI is their technological soul mate.
Well said, Amy!
Well, they managed to hit the Chinese Embassy in Belgrade without any help from "AI".
Will you update the post? It was an Iranian missile that exploded on their own embedded launchpad. Officially corroborated.
I'm reminded that the West (UK and UK) both opted for strategic bombing from WWII onwards. Reading the late Freeman Dyson's scientific memoirs "Disturbing the Universe" (1979), he noted that his analyses, while working for the RAF in WWII, showed that strategic bombing was not effective in destroying Germany's ability to defend against the Allies, but may have increased it. Vietnam showed that such bombing was a failure, too. While advanced technology powers have moved towards targeted bombing (with smart bombs and drones), humans in the loop have made mistakes in identifying targets, e.g., wedding parties. Making such decisions more autonomous with AI will unlikely improve matters, and may make it even worse, as there is no longer guilt by pilots making errors, it is back to the old "Humans make mistakes, but it takes a computer to really f**k up." The more machines make decisions, the more "collateral damage" will occur. (At least the AIs won't laugh with glee at killing medical responders to casualties, or play the Israeli "double tap" executions by aerial arms.)
Illegal collateral damage should have the buck stop at those in charge, like Hegseth at DOD [DOW?] and POTUS. We have already seen how Hegseth tried to distance himself when the circumstances of the first "drug boat" executions of survivors came to light. Trump will do the same if any mistargeted killings in Iran come to light.
I'll emphasize the point made using the dice analogy: anyone who uses generative "AI" to make a recommendation (targeting or otherwise) is 100% personally responsible for that recommendation.
I put the "AI" in "Generative AI" in quotes because it's really a misnomer. It shouldn't be called "AI" because it's not really intelligent. It's just a statistical micro-plagiarizer, mashing together so many things written by real humans that it's not possible to identify everyone it's stealing from.
Certainly Generative "AI" can be useful for searching and brainstorming, provided everything it comes up with is scrupulously double and triple checked.
Morons using morons to decide who to kill in an illegal war. You get what you vote for.
Suspicious that Hegseth was so adamant that Anthropic had to get on board before Friday end of day. Perhaps he wanted to a non-military scapegoat if/when a targeting error occurred.
Or did Amodei know what was imminent and want to dissociate Anthropic from any blame for the [mis]use of their AI technology?
From this morning's WSJ -- "For Operation Epic Fury, Claude helped with intelligence assessments, target identification and simulating battle scenarios." https://www.wsj.com/wsjplus/dashboard/articles/anthropics-claude-joins-operation-epic-furys-extensive-arms-list-134169616506270110
Unfortunately, WSJ is behind a pay wall.
"Nearly 150." It started out as 50. Here it's 150. I've seen it as high as 200, recently. Here. Let me help. I say that *ten thousand* tiny, innocent babies were slaughtered by (checks list) Evil Drumpf, Nazi Jews, Demonic AI, White Supremacist Nationalists - pick at least one. Feel free, and don't worry about being wrong. My number has precisely as much confirmatory proof as does 50, 100, 150, 200, whatever.
Indeed. AI is going to be a very effective ‘accountability sink’.
Just wait until ‘target identification’ is let loose at the public at large.
Oh wait, that has already been happening in several countries.
A country that accepts active mass shooter drills in its elementary schools is probably not all that concerned about killing 100+ Iranian children.