'Lavender': The AI machine directing Israel's bombing in Gaza

Should have the ability to turn off comments for these.

I'm betting the author of this comment knows full well the Dang has been doing this for all US and Israel critical topics for years and this comment has zero chance of affecting anything. The very definition of lip service. I hope you feel better in your continuous support for HN.

Also the topic is locked and no one at this point gives a damn what is said in here.. I'm actually curious if I'm getting "disciplined" for this. No one besides the people that already posted and maybe one or two other people will ever see this.

So the entire family and neighbors family.

Sure would be convenient if Hamas is 6% of the population

Definitely Palestinians are not going to forget this.

With 37,000 Palestinians marked as suspected militants, it would mean they expected up to 555,000 - 740,000 civilian casualties.

The law of armed conflict acknowledges that civilian deaths are inevitable, and only prohibits attacks that are directed at civilians; rather than those which are directed at combatants with expected civilian casualties as collateral damage.

The legal question is whether the civilian casualties are proportional to the concrete military value of the target.

A question that's worth considering is whether, when considering proportionality, all civilians (as defined by law) are made equal in a moral sense.

For example, the category "civilian" includes munitions workers or those otherwise offering support to combatants on the one hand, and young children on the other. It also includes members of the civil population who are actually involved in hostilities without being a formal part of an armed force.

The law of armed conflict doesn't distinguish these; albeit that I think people might well distinguish, on a moral level, between casualties amongst young children, munitions workers, and informal combatants.

"Our system is 90% accurate if you don't count the 15-20 innocent people taken out for each hit". I know they're measuring the accuracy of target identification but that's laughable when used in this context.

For 100 targets, 90 are 'correct', plus 20x civs per-target is 90/2100 or 4% real accuracy.

Say you use a model that's only 50% accurate and limit yourself to 10 civs per-target, you're at 50/1100 or 4.5% accuracy!

As someone who sees both sides of this, and as someone who didn't understand this for some time, it's important to understand that one reason a story is likely to get flagged is because users think it's highly unlikely to lead to productive discussion. It doesn't mean it's a bad story, or even unworthy of discussion, but many types of stories seem to, pretty predictably, lead to a cesspool of comments where it's clear most folks have no desire to listen to opposing points of view.

FWIW, I found this to be a really interesting story that I didn't previously know about, so I hope it stays up, and this is a story I'd be willing to vouch for.

I don't understand why it was flagged, obviously it is a sensitive topic but AI being used to kill people is very clearly a HN-worthy topic

Let's see how long it takes this time! I'd give it 50% odds of lasting 12 minutes.

Edit: Flagged after less than 9 minutes, I overestimated!

I don't take any issue with people flagging a post, so long as an actual person makes the ultimate decision on whether to keep it up.

This is in contrast to how I feel about a statistical model flagging people to be murdered. That's not even remotely OK, even if the decision to actually carry out the murder ultimately goes through a person. Using a statistical model to choose targets is incredibly naive, and practically guarantees that perverse incentives will drive decision-making.

This is a typical phenomenon when a topic is divisive, and the current topic is perhaps the most divisive HN has ever seen.

I don't know about kill lists, but AI weapons kinda make sense.

No weapons are nice, but if the good guys don't develop AI weapons, the bad guys will.

From what I gather, many US engineers are morally opposed to them. But if China develops them and gets into a war with the US, will Americans be happy to lose knowing that they have the moral high ground?

They're great because the accountability for fuckups goes on the system, not on the people using the system. "Oops, the system had a bug" doesn't kill careers like "Oops, I made a bad call."

Depending on your definition of AI they’ve probably been around for a while.

This does seem to be a big step more “AI” than previous systems I’ve heard described though.

How do you think people are chosen to visit a secret CIA prison, or chosen to get a 12 hour interrogation every time they enter the US?

Can't wait to be killed by drone strike when a GPT hallucinates my name.

Such things have been around for at least a decade. It didn't start with the same kind of AI that's being talked about recently, but there is a large automated scoring component: "Targets are often chosen based on metadata."

https://en.wikipedia.org/wiki/Disposition_Matrix

This is the now-flagged to death HN thread on the Guardian version [0]

I would hope they can be unflagged and merged, this appears to be an important story about a novel use of technology.

[0] https://news.ycombinator.com/item?id=39917727

guardian references 972 as source for report. it's not like it's "the guardian" article

Pretty disappointing that Guardian article also got quickly flagged after HN submission

The other option here is carpet bomb a la Drezden, that would have resulted in >400,000 casualties at best.

Why is it barbarism? If it makes the war more efficient and more targeted, it is preferred.

It sure would be nice if this industry had the tiniest shred of collective consciousness and realized our capacity to exert some level of control over what gets built and what doesn't.

The people working on these understand the alternative looks like a WWII bombing campaign with greater loss of life

Not everyone sees the world as you do. Given this article and other information I know about this system, I would be honoured to work on it and take a significant pay cut, as it actively makes the world a better and safer place.

The use of AI and the authorisation to kill civilians are unrelated parts of this story. Nowhere does it mention that the AI is being used to justify killing of civilians.

This is not what the article is about, and not what AI was being used for.

There is no justification for killing.

Wild that this is still a controversial statement on HN, which is otherwise rather forward thinking.

The issue as I see it is that the tools available don't just determine how a given war is fought, they also determine whether it is fought at all.

If Israel wasn't able to use tools like this, then it probably wouldn't be viable for them to identify much of Hamas (that's kind of the point of guerilla warfare). Since that would make it difficult to fight a war efficiently, they would be more likely to engage in diplomacy.

Why not both? Taking a side does not mean you are clouded in judgement on this point.

I'm not sure that is possible. The nature and limitations of current AI technology means that it is almost impossible to talk about it without coming to certain conclusions about the party using it.

To put it bluntly, useing AI to decide on targets for lethal operations in unconsiounable given the current and forseable state of technology.

Come back to me when it can be trusted to make mortgage eligability questions without engaging in what would be blatantly illegal discrimination if not laundered by a computer algorithm.

No, probably not. When the topic at hand is the selection criteria used to justify the killing of tens of thousands of civilians, your stance on whether the ones killing tens of thousands of civilians are justified in doing so is rather intrinsic.

I am going to bet the answer to your question is "No"

It's also interesting (and I guess typical for end-users of software) how quickly and easily something like this goes from "Here's a tool you can use as an information input when deciding who to target" to "I dunno, computer says these are the people we need to kill, let's get to it".

In the Guardian article, an IDF spokesperson says it exists and is only used as the former, and I'm sure that's what was intended and maybe even what the higher-ups think, but I suspect it's become the latter.

It always starts with making a list of targets that meet given criteria. Once you have the list its use changes from categorisation to demonisation -> surveillance -> denial of rights -> deportations -> killing. Early use of computers by Germans during WW2 included making and processing of lists of people who ought to be sent to concentration camps. The only difference today is that we are able to capture more data and process it faster at scale.

War is messy, I don’t think there was ever one with the surgical precision these courts expect. Ai flagging targets sure beats the cold impunity of the strategic air campaign during wwii and vietnam by far.

What part of this upsets you vs a baseline understanding of reality?

There's often a criticism of the US military doctrine that our weapons are great but are often way more expensive than the thing we shoot them at (as exemplified in our engagement with the Houthis in the Red Sea.)

If anything, the quote you pulled sounds like its talking about highly precise weaponry, and it seems to me that the way to minimize the overall death in a war is to use your precise weapons to take out the most impactful enemy.

Which part of this is different than how you see the world so that reading this quote threw you?

Its rich when the argument for the system is that the targeting is the bottleneck.

This is also reported on by The Guardian https://www.theguardian.com/world/2024/apr/03/israel-gaza-ai...

okay ... sure: https://mediabiasfactcheck.com/972-magazine/

Where is YOUR evidence for such accusations?

ISIS ways of operation sure influenced other terrorist organizations. Hamas atrocities do look like copycats of ISIS ( burning alive, beheadings) with deliberate usage of social media to satisfy their Muslim supporters and Western left/woke narcissists.

But that only looks more brutal to those who get their news from Facebook. It was always brutal being attacked by Arab/Muslim mobs. Farhud in Iraq(1941) and Arab riots of the 1920s in Palestine and other places. Same hallmarks with beheadings, rapes and burnings.

Nothing new.

The weaponisation of online media for manipulating the perception of global audiences about the conflict, has definitely ramped up recently. For example, the official Twitter account of Israel’s Ministry of Foreign Affairs has posted videos of muslim preachers appearing to denounce lgbt culture during public service in Palestinian mosques. Hamas themselves are denying their involvement in the 2023 massacre and accusing Israel of staging the graphic footage that was disseminated. This greatly polarises the debates on social media and it’s much more common now to see people who are deeply invested emotionally in the narrative of either side.

When the US dropped napalm indecriminately over the vietnamese jungle or absolutely leveled dresden in one bombing run or unleashed nuclear hellfire over japan, they probably killed a lot of journalists and doctors and food workers as well. Interestingly, western media did not beat itself into a frenzy over it at the time. Its easy to get cynical about it all seeing how easily narratives are manufactured and controlled to serve political ends.

Most of the mainstream media has historically glossed over the atrocities, but it is impossible to ignore them today because of what we see live on the scene thanks to smaller outlets having a broader reach and social media.

It's mostly business as usual. The technology makes the brutality more efficient, though:

Describing human personnel as a “bottleneck” that limits the army’s capacity during a military operation, the commander laments: “We [humans] cannot process so much information. It doesn’t matter how many people you have tasked to produce targets during the war — you still cannot produce enough targets per day.”

...

By adding a name from the Lavender-generated lists to the Where’s Daddy? home tracking system, A. explained, the marked person would be placed under ongoing surveillance, and could be attacked as soon as they set foot in their home, collapsing the house on everyone inside.

“Let’s say you calculate [that there is one] Hamas [operative] plus 10 [civilians in the house],” A. said. “Usually, these 10 will be women and children. So absurdly, it turns out that most of the people you killed were women and children.”

Using Google search, you can search new articles in previous years. You'll find older articles about Israel killing aid workers, for example. This is from 2018: https://www.theguardian.com/global-development/2018/aug/24/i...

The interesting thing about how this conflict is developing is that this story is full of quotes from Israeli intelligence. Most plainly say what they're doing. Western outlets may put a positive spin on it (because our governments generally support Israel), but the Israeli military themselves are making their intentions clear: https://news.yahoo.com/israeli-minister-admits-military-carr...

It’s much easier to bomb a family’s home.

Okay, how is this not a war crime?

There are ~2M civilians who live in Gaza, and many of them don't have access to food, water, medicine, or safe shelter. Some of those unfortunates live above, or below, Hamas operatives and their families.

"Oh, sorry, lol." "It was unintentional, lmao, seriously." "Our doctrine states that we can kill X civilians for every hostile operative, so don't worry about it."

The war in Gaza is unlike Ukraine -- where Ukrainian and Russian villagers can move away from the front, either towards Russia or westwards into Galicia -- and where nobody's flattening major population centers. In Gaza, anybody can evidently be killed at any time, for any reason or for no reason at all. The Israeli "strategy" makes the Ukrainians and Russians look like paragons of restraint and civility.

Watching i24 news is a little unsettling. They run bits with interrogators announcing how productive torture has been, and make jokes about how it would be much easier if lemons just gave up their juice without being squeezed.

AI based killing technology is actually more ethical and safer than warfare that doesn’t use AI

No. It's just a tool. People still configure the parameters and ultimately make decisions. Likewise modern missile do not make conflicts more or less ethical just because they require advanced physics.

No the AI was the scapegoat for IDF deciding to "target" low-level enemies, then bombing them with bunker-buster 2000lb bombs that leveled entire buildings and city blocks around those targets.

The AI did something, but the IDF used it to justify effectively committing a genocide.

I think the concern is that the AI is making life or death judgements against people. Some may of course be lawful combatants under the rules that govern such things, but the fact that an AI is drawing these conclusions that humans act on is the shocking part.

I doubt an artillery system using machine learning to correct its trajectory and get better accuracy would be controversial, since the AI in that case is just controlling the path of a shell that an operator has determined needs to hit a target decided upon by humans.

I think that depends on what the alternative is. It seems to me that the problem is that there's no way for Israel to wipe out Hamas without massive collateral damage. However, instead of giving up on wiping out Hamas, they just decided that they are OK with the collateral damage.

How far does the AI system go… is it behind the AI decision to starve the population of Gaza?

No, the point of this program seems to be to find targets for assassination, removing the human bottleneck. I don't think bigger strategic decisions like starving the population of Gaza was bottlenecked in the same way as finding/deciding on bombing targets is.

is it also behind the decision to kill the aid workers who are trying to feed the starving?

It would seem like this program gives whoever is responsible for the actual bombing a list of targets to chose from, so supposedly a human was behind that decision but aided by a computer. Then it turns out (according to the article at least) that the responsible parties mostly rubberstamped those lists without further verification.

can an AI commit a war crime?

No, war crimes are about making individuals responsible for their choices, not about making programs responsible for their output. At least currently.

The users/makers of the AI surely could be held in violation of laws of war though, depending on what they are doing/did.

Also, can an AI commit a war crime?

"An AI" doesn't exist. What is being labeled "AI" here is a statistical model. A model can't do anything; it can only be used to sift data.

No matter where in the chain of actions you put a model, you can't offset human responsibility to that model. If you try, reasonable people will (hopefully) call you out on your bullshit.

There’s so much about this death machine AI I would like to know.

The death machine here is Israel's military. That's a group of people who don't get to hide behind the facade of "an AI told me". It's a group of people who need to be held responsible for naively using a statistical model to choose who they murder next.

It's unconfirmed who authorized it but the recent food charity workers killed by Israeli bombing had a security person (death confirmed by family in UK) who is unarmed but by job description clears the way by telling Israeli authorities where the charity team is going to be so the chain of command knew who they were, so one is naturally lead to ask - who would authorize a targeted killing in this situation? The after photos show the missile went right through the roof of the car, ironically next to the food charity's visible logo on top of the car. Israeli defense minister now claims it was a mistake, although if they had hit a real target it might have been acceptable in terms of their rules of engagement with 15-100 unrelated collateral deaths according to the investigation.

I think optics of being advanced aren't the main goal. Some form of "justification", no matter how flimsy, especially if it's hard to audit how the "AI" came to it's conclusions, is the goal. Now anyone is a target. Similar to cops in the US "smelling weed" or dogs "signaling". It provides the means to justify any search, or in this case, any kill. The machine grinds away..

I'm not sure why its such a shock to many to see the censorship on HN. This isn't a public square.

We are privy to the whims of whatever political views of those that aligned/run/manage/stake in YC and their policies and values.

This is not new and dang and the others are absolutely fine with posts getting gang-flagged in a matter of minutes. Just shows how impartial they are.

That's a valid point, but a terrible example because AI cars are legal in many places.

In all fairness, driving a car is a lot more complicated and full of dangerous edge cases than dropping objects or shooting anyone within a geofence.

"AI" in this case is probably mostly Oct 6 cell phone locations.

It is obvious that Israel has loosened their targeting requirements, this story points to their internal justifications. The first step in ending this conflict must be to reimpose these standards of self restraint.

what do you mean "bribed and blackmailed"?

It's been pretty clear to me for a while now that Israel's long term plan for the Palestinians is to expel them all. Starvation isnt a requirement for that, but it is probably the path of least resistance. I will say that its happening a lot faster than I expected though, Israel definitely taking advantage of the situation here.

Bad UX, it's hiding under the hamburger menu in the top left https://web.archive.org/web/20240401100849/https://www.972ma...

https://mediabiasfactcheck.com/972-magazine/

https://www.972mag.com/about/

Another site covering this story:

https://www.theguardian.com/world/2024/apr/03/israel-gaza-ai...

The purpose of a system is what it does. :(

Endless scrolling feed

Continuously throw enough plot twists and general stimulation at people and they'll never have the time to consider whether they're living in a simulation.

Or in the words of Shaggy…

“Saw you blowing up the children…”

“It wasn’t me.”

It shows Israel in a bad light.

HN, both its community and the moderators, flag posts that generate a lot of conflict in the comments. The comments on this are especially bad by HN standards and therefore the flagging is inline with how the site is openly operated.

I am pro Palestine and not simping for Israel. I think visibility on Israel's actions matter, but HN is also very clearly not the appropriate website for a lot of politically involved news.

War has always been scary. We are busy inventing new ways of killing each other and there is no sign of stopping.

I'm sorry, you think this is new?

War is terrible. War has always been terrible. It was almost certainly worse in the past, but it still sucks now. Most of the things you mention were way worse 100 years ago.

Sure, AI didn't write the propaganda, instead humans did. The affect was the same.

What can one do when Hamas has embedded them self in the civilian population? Why don't they get out and meet the Israeli army on the battle field? This is no different than chemotherapy, in order got the body to survive some healthy cells will die together with cancerous one. It is much better than the carpet bombing used by other nations.

how the world was working in 2001

By the world you mean the US, but yes you are correct.

"NSA targets SIM cards for drone strikes, ‘Death by unreliable metadata’"

https://www.computerworld.com/article/2475921/whistleblower-...

I think the algorithm, in this case, makes a suggestion and then a human evaluates it. The article claims they've only looked at the sex of the target (kill if male) but also claims 90% effectiveness. I'm curious if 90% is a good number or not? War will always have collateral damages but if technology can help limit that beyond what only a human could do then I'd say it's a net positive. I think the massive efficiencies the algorithm brings to picking targets is a bit frightening (nowhere to run or hide now) but there's no real turning back.

People thought this way about the machine gun, the armored tank, the atom bomb. But once the genie is out there's no putting it back in.

As an aside, I think this is a good example of how humans and AI will work together to bring efficiency to whatever tasks need to be accomplished. There's a lot of fear of AI taking jobs, but I think it was Peter Thiel who said years ago that future AI would work side by side humans to accomplish tasks. Here we are.

I really want to support this, but the website is pretty bad. Blinding colors, poor and sparse information, and a links to shop/donate without a notion as to what or who the org is.

That's just everyday citizens

https://edition.cnn.com/2024/03/08/middleeast/gaza-israelis-...

Yea this really seems like more of a weapon of propaganda directed at Israelis. If they didn't want people to know about it, we probably wouldn't know about it. The fact that we're talking about it is probably not an accident, and I guess the play here would be to convince Israelis that the army is technologically advanced and they know what they're doing, so don't question it. But AI or not they were going to commit genocide and violate every international humanitarian law on the book. But for the people that still believe the genocide is justified I think this probably improves the optics.

It is and they are.

The link between targeting - whether for advertisements or for death - is quite disturbing.

Both use personal metadata, and both can horribly get it wrong.

Shades of https://www.nytimes.com/2012/05/29/world/obamas-leadership-i....

It is also because Mr. Obama embraced a disputed method for counting civilian casualties that did little to box him in. It in effect counts all military-age males in a strike zone as combatants, according to several administration officials, unless there is explicit intelligence posthumously proving them innocent.

Counterterrorism officials insist this approach is one of simple logic: people in an area of known terrorist activity, or found with a top Qaeda operative, are probably up to no good. “Al Qaeda is an insular, paranoid organization — innocent neighbors don’t hitchhike rides in the back of trucks headed for the border with guns and bombs,” said one official, who requested anonymity to speak about what is still a classified program.

So frustrating how easy it is for those of a certain zeal to wipe off mention of that which they find inconvenient.

There could hardly be a more pertinent issue for tech right now. Just sweepingly wild shit that we should be grappling with.

As a backer on the original Oculus kickstarter, I have such a sinking feeling in my stomach every time this comes up. My money went to enable Luckey to achieve this and I hate myself for it.

This literally looks like any aborrhent ai "predicting" system such as the ones we've heard a ton about in the past, with the same mistakes (I wonder if they're really mistakes, bugs, or ahem... Features)

It’s slightly more complicated a.) looks like male b.) lives here c.) send unguided munition if less than 15 or 100 other non targets depending upon value of target.