> Omnilert later admitted the incident was a “false positive” but claimed the system “functioned as intended,” saying its purpose is to “prioritize safety and awareness through rapid human verification.”
It just created a situation in which a bunch of people with guns were told that some teen had a gun. That's a very unsafe situation that the system created, out of nothing.
And some teen may be traumatized. Again, unsafe.
Incidentally, the article's quotes make this teen sound more adult than anyone who sold or purchased this technology product.
Another false positive by one of these leading content filters schools use - the kid said something stupid in a group chat and an AI reported it to the school, and the school contacted the police. The kid was arrested, stripped searched, and held for 24 hours without access to their parents or counsel. They ultimately had to spend time in probation, a full mental health evaluation, and go to an alternative school for a period of time. They are suing Gaggle, who claims they never intended their system to be used that way.
These kinds of false positives are incredibly common. I interviewed at one of their competitors (Lightspeed), and they actually provide a paid service where they have humans review all the alerts before being forwarded to the school or authorities. This is a paid addon, though.
> Gaggle’s CEO, Jeff Patterson, said in an interview that the school system did not use Gaggle the way it is intended. The purpose is to find early warning signs and intervene before problems escalate to law enforcement, he said.
“I wish that was treated as a teachable moment, not a law enforcement moment,” said Patterson.
It's entirely predictable that schools will call law enforcement for many of these detections. You can't sell to schools that have "zero tolerance" policies and pretend that your product won't trigger those policies.
The kid was arrested, stripped searched, and held for 24 hours without access to their parents or counsel. They ultimately had to spend time in probation, a full mental health evaluation, and go to an alternative school for a period of time.
All he wanted was a Pepsi. Just one Pepsi. And they wouldn't give it to him.
Holy shitballs. In my experience such pain addons have very cheap labor attached to them, certainly not what you would expect based on the sales pitch.
Engineer: hey I made this cool thing that can help people in public safety roles process information and make decisions more efficiently! It gives false positives but you save more time than it takes less time to weed through them.
Someone nearby: well what if they use it to replace human thinking instead of augment it?
Engineer: well they would be ridiculous. Nobody would ever think that’s a good idea.
Marketing Team: it seems like this lands best when positioning it as a decision-making tool. Let’s get some metrics on how much faster it is at making decisions than people are.
Sales Rep: ok, Captain, let’s dive into our flagship product, DecisionMaker Pro, the totally automated security monitoring agent…
::6 months later—some kid is being held at gunpoint over snacks.::
Nice fantasy, but the reality is that the "people in public safety roles" love using flimsy pretenses to harass and abuse vulnerable populations. I wish it was just overeager sales and marketing, but you're view of humanity is way too naive especially as masked thugs are disappearing people in the street as we type.
delegating decision to AI, excluding human from the "human in the loop" is kind of unexpected as a first step, as in general it was expected that exclusion will start from the other end. Sideway i wonder how is that going to happen on the battlefield.
for this civilian use case the next step - AR googles worn by police with that AI will be projecting onto the googles where that teenager has his gun (kind of Black Mirror style), and the next step after that is obviously excluding the humans even from the execution step.
In any system, there are false positives and false negatives. In some situations (like a high recall disease detection) false negatives are much worse than false positives, because the cost of a false positive is a more rigorous screening.
But in this case both are bad. If it was a false negative students might need therapy for a more tragic reason.
Aside from improving the quality of the detection model, we should try to reduce the “cost” of both failure modes as much as possible. Putting a human in the loop or having secondary checks are ways to do that.
In this case false positives are far, far worse than false negatives. A false negative in this system does not mean a tragedy will occur, because there are many other preventative measures in place. And never mind the fact that this country refuses to even address the primary cause of gun violence in the first place: the ubiquity of guns in our society. So systems like this is what we end up with when we ignore to address the problem of guns and choose to deal the downstream effects of that instead.
> But in this case both are bad. If it was a false negative students might need therapy for a more tragic reason.
Given the probability of police officers in the USA taking any action as hostile and then ending up shooting him a false positive here is the same as swatting someone.
The system here sent the police off to kill someone.
I was swatted once. Girlfriend's house. Someone called 911 and said they'd seen me kill a neighbor, drag their body into the house, and was now holding my gf's family hostage.
We answered the screams at the door to guns pointed at our faces, and countless cops.
It was explained to us that this was the restrained version. We got a knock.
Unfortunately, I understand why these responses can't be neutered too much. You just never know.
False positives can effectively lead to false negatives too. If too many alarms end in teens getting swatted (or worse) for eating chips, people might ignore the alarm if an actual school shooter triggers it. Might assume the AI is just screaming about a bag of chips again.
I think a “true positive” is an issue as well if the protocol to manage it isn’t appropriate. If the kid was armed with something other than nacho cheese, the provocative reaction could have easily set off a tragic chain of events.
Reality is there are guns in schools every day. “Solutions” like this aren’t making anyone safer. School shooters don’t fit this profile - they are planners, not impulsive people hanging out at the social event.
More disturbing is the meh attitude of both the company and the school administration. They almost engineered a tragedy through incompetence, and learned nothing.
The danger is that it's as clear as day that in the future someone is gonna be killed. That's not just a possibility. It's a certainty given the way the system is set up.
This tech is not supposed to be used in this fashion. It's not ready.
Did you want to emphasize or clarify the first danger I mentioned?
My read of the "Um" and the quoting, was that you thought I missed that first danger, and so were disagreeing in a dismissive way.
When actually we're largely in agreement about the first danger. But people trying to follow a flood of online dialogue might miss that.
I mentioned the second danger because it's also significant. Many people don't understand how safety works, and will think "nobody got shot, so the system must have worked, nothing to be concerned about". But it's harder for even those people to dismiss the situation entirely, when the second danger is pointed out.
Americans are killed by police all the time, and by other Americans. We've already decided as a society that we don't care enough to take the problem seriously. Gun violence, both public and from the state, is accepted as unavoidable and defended as a necessary price to pay to live in a free society[0]. Having a computer call the shots wouldn't actually make much of a difference.
Hell, it wouldn't even move the needle on racial bias much because LLMs have already shown themselves to be prejudiced against minorities due to the stereotypes in their training data.
[0]Even though no other free society has to pay that price but whatever.
> The danger is that it's as clear as day that in the future someone is gonna be killed.
This argument can be made about almost every technology, including fire, electricity, and even the base "technology" of just letting a person watch a camera feed.
I'm not downplaying the risks. I'm saying that we should remember that almost everything has risks and benefits, and as a society we decide for or against using/doing them based mostly on the ratio between those two things.
So we need some data on the rates of false vs. true detections here. (A value judgment is also required.)
"“We understand how upsetting this was for the individual that was searched as well as the other students who witnessed the incident,” the principal wrote. “Our counselors will provide direct support to the students who were involved.”"
Make them pay money for false positives instead of direct support and counselling. This technology is not ready for production, it should be in a lab not in public buildings such as schools.
This assumes no human verification of the flagged video. Maybe the bag DID look like a gun. We'll never know, because modern journalism has no interest in such things. They obtained the required emotional quotes and moved on.
The taxpayers collectively pay the money, the officers involved don't (except for that small fraction of their income they pay in taxes that increase as a result).
The question is whether that Doritos carrying kid is still alive, instead of being shot at by the violent cops (who typically do nothing when an actual shooter is roaming a school on a killing spree; apropos the Uvalve school shooting, when hundreds of cops milled around the school in full body armor, refusing to engage the shooter on killing spree inside the school, and they even prevented the parents from going inside to rescue their kids) based on a false positive about a gun (and the cops must have figured that it's likely a false positive, because it is info from AI surveillance), only because he is white?
Before clicking on the article, I kinda assumed the student was black. I wouldn't be surprised if the AI model they're using has race-related biases. On the contrary, I would be surprised if it didn't.
> Make them pay money for false positives instead of direct support and counselling.
Agreed.
> This technology is not ready for production
No one wants stuff like this to happen, but nearly all technologies have risks. I don't consider that a single false positive outweighs all of its benefits; it would depend on the rates of false and true positives, and the (subjective) value of each (both high in this case, though I'd say preventing 1 shooting is unequivocally of more value than preventing 1 innocent person being unnecessarily held at gunpoint).
I think I’ve said this too many times already but the core problem here and with the “AI craze” is that nobody realy wants to solve problems, what they want is a marketable product and AI seems to be the magic wrench that fits in all the nuts and since most of the people doesn’t really knows how it works and what are its limitations the happily buy the “magic dust”.
Stuff like this feels like some company has managed to monetize an open source object detection model like YOLO [1], creating something that could be cobbled together relatively easily, and then sold it as advance AI capabilities. (You'd hope they've have at least fine-tuned it / have a good training dataset.)
We've got a model out there now that we've just seen has put someone's life at risk... Does anyone apart from that company actually know how accurate it is? What it's been trained on? Its false positive rate? If we are going to start rolling out stuff like this, should it not be mandatory for stats / figures to be published? For us to know more about the model, and what it was trained on?
I expect that John Bryan -- who produces content as The Civil Rights Lawyer https://thecivilrightslawyer.com -- will have something to say about it.
He frequently criticizes the legality of police holding people at gunpoint based on flimsy information, because the law considers it a use of force, which needs to be justified. For instance, he's done several videos about "high risk" vehicle stops, where a car is mistakenly flagged as stolen, and the occupants are forced to the ground at gunpoint.
My take is that if both the AI automated system and a "human in the loop" police officer looked at the picture and reasonably believed that the object in the image was a gun, then the stop might have been justified.
But if the automated system just sent the officers out without having them review the image beforehand, that's much less reasonable justification.
Or they'll tell us police have started shooting because an acorn falls, so they shouldn't be expected to be held to higher standards and are possibly an improvement.
The article says the police later showed the student the photo that triggered the alert. He had a crumpled-up Doritos bag in his pocket. So there was no gun in the photo, just a pocket bulge that the AI thought was a gun... which sounds like a hallucination, not any actual reasonable pattern-matching going on.
But the fact that the police showed the photo does suggest that maybe they did manually review the photo before going out. If that's the case, do wonder how much the AI influenced their own judgment, though. That is, if there was no AI involved, and police were just looking at real-time surveillance footage, would they have made the same call on their own? Possibly not: it feels reasonable to assume that they let the fact of the AI flagging it override their own judgment to some degree.
Ah, the coming age of Palantir's all seeing platform; and Peter Thiel becoming the shadow Emperor. Too bad non-deterministic ml systems are prone to errors that risk lives when applied wrongly to crucial parts of society. But in an authoritarian state, those will be hidden away anyway, so there's nothing to see here: move along folks. Yes, surveillance and authoritarianism go hand in hand, ask China. It's important to protest these methods and push lawmakers to act against them; now, before it's too late.
American, please, wake up. The masked border police are on the streets arresting citizens, the military is being paid as a client of the president, corruption is legal, and a mass surveillance machine unfathomable to prior dictatorships is being/has been established. You're fucked. Listen to the soapbox. It is very, very relevant. Wake up.
I'm pretty sure that some people will continue to apply the term "soapbox ranting" to all opposition against the technofascism even when victims of its false positives will be in need of coroners, not psychologists.
Omnilert later admitted the incident was a “false positive” but claimed the system “functioned as intended,” saying its purpose is to “prioritize safety and awareness through rapid human verification.”
prioritize your own safety by not attending any location fitted with such a system, or deemed to be such a dangerous environment that such a system is desired.
Calling it today. This company is going to get innocent kids killed.
How many packs of Twizzlers, or Doritos, or Snickers bars are out there in our schools?
First time it happens, there will be an explosion of protests. Especially now that the public knows that the system isn't working but the authorities kept using it anyway.
This is a really bad idea right now. The technology is just not there yet.
And then there's plenty of bullies who might put a sticker of a picture of a gun on someone's back, knowing it will set off the image recognition. It's only a matter of time until they figure that out.
>First time it happens, there will be an explosion of protests.
Why do you believe this? In the US, cops will cower outside of a school with an armed gunman actively murdering children, forcibly detain parents who wish to go in if the cops wont, and then re-elect everyone involved
In the US, an entire segment of the population will send you death threats claiming you are part of some grand (democrat of course) conspiracy for the crime of being a victim of a school shooting. After every school shooting, republican lawmakers wear an AR-15 pin to work the next day to ensure you know who they care about.
Over 50% of the country blamed the protesting students at Kent state for daring to be murdered by the national guard.
Cops can shoot people in broad daylight, in the back, with no justification, or reasonable cause, or can even barge into entirely the wrong house and open fire on homeowners exercising their basic right to protect themselves from strangers invading their homes, and as long as the people who die are mostly black, half the country will spout crap like "They died from drugs" or "they once sold a cigarette" or "he stole skittles" or "they looked at my wife wrong" while the cops take selfies reenacting the murder for laughs and talk about how terrified they are by BS "training" that makes them treat every stranger as a wanted murderer armed to the teeth. The leading cause of death for cops is still like heart disease of course.
Trump sent unmarked forces to Portland and abducted people off the street into unmarked vans and I think we still don't really know what happened there? He was re-elected.
The technology literally can NEVER be there. It is completely impossible to positively identify a bulge in clothing as a handgun. But that doesn’t stop irresponsible salesmen from making the claim anyway.
Why did they waste time verifying? The police should have eliminated the threat before any harm could be done. Seconds count when you're keeping people safe.
I get that you're being sarcastic and find the police response appalling, but the sad reality of Poe's Law is that there are a lot of people who would unironically say this and would have cheered if the cops had shot this kid, either because they hate black people or because they get off on violence and police shootings are a social sanctioned way to indulge that taste.
* Even hundreds of cops in full body armor and armed with automatic guns will not dare to engage a single "lone wolf" shooter doing a killing spree in a school; the heartless cowards may even prevent the parents from going inside to rescue their kids: Uvalde school shooting incident
* Cop on a ego trip, will shoot down a clearly harmless kid calmly eating a burger in his own car (not a stolen car): Erik Cantu incident
The dispatch relayer and responding officers should at least have ready access to a screen where they can see a video/image of the raw footage that triggered the AI alert. If it is a false alarm, they will better see it and react accordingly, and if it is a real threat they will better understand the initial context and who may have been involved.
According to a news article[1], a human did review the video/image and flagged it as a false positive. It was the principal who told the school cop, who then called other cops:
> The Department of School Safety and Security quickly reviewed and canceled the initial alert after confirming there was no weapon. I contacted our school resource officer (SRO) and reported the matter to him, and he contacted the local precinct for additional support. Police officers responded to the school, searched the individual and quickly confirmed that they were not in possession of any weapons.
What's unclear to me is the information flow. If the Department of School Safety and Security recognized it as a false positive, why did the principal alert the school resource officer? And what sort of telephone game happened to cause local police to believe the student was likely armed?
Walking through TSA scanners, I always get that unnerving feeling I will get pulled aside. 50% of the time they flag my cargo pants because of the zipper pockets - There is nothing in them but the scanner doesn't like them.
Now we get the privilege of walking by AI security cameras placed in random locations, hoping they don't flag us.
There's a ton of money to be made with this kind of global frisking, so lots of pressure to roll out more and more systems.
To be fair, at least you can choose not to wear the cargo pants.
A friend of mine once got pulled aside for extra checks and questioning after he had already gone through the scanners, because he was waiting for me on the other side to walk to the gates together and the agent didn't like that he was "loitering" – guess his ethnicity...
Not sure, but I bet they were looking at their feet kinda dusting off the bottoms, making awkward eye contact with the security guard on the other side of the one way door.
Email the state congressman and tell them what you think.
Since (pretty much) nobody does this, if a few hundred people do it, they will sit up and take notice. It takes less people than you might think.
Since coordinating this with a bunch of strangers (I.e. the public) is difficult, the most effective way is to normalise speaking up in our culture. Of course normalising it will increase the incoming comm rate, which will slowly decrease the effectiveness but even post that state, it’s better than where we are, which is silent public apathy
If that's the case, why do people in Congress keep voting for things their constituents don't like? When they get booed at town halls they just dismiss it as being the work of paid activists.
Get Precheck or global entry. I only do a scanner every 5 years or so when I get pulled at random for it. Otherwise it's metal detector only. Unless your zippers have such chunky metal that they set that off you'll be fine. My belt and watch don't.
Note: Precheck is incredibly quick and easy to get; and GE is time consuming and annoying, but has its benefits if you travel internationallly. Both give the same benefits at TSA.
Second note: let's pretend someone replied "I shouldn't have to do that just to be treated...blah blah" and that I replied, "maybe not, but a few bucks could still solve this problem, if it bothers you enough that's worth it to you."
...maybe not, but a few bucks could still solve this problem
Sure, can't argue with that. But doesn't it bug you just a little that (paying a fee to avoid harassment) doesn't look all that disimilar from a protection racket? As to whether it's a few bucks or many, now you're just a mark negotiating the price.
I don't often fly, but back when I went to germany on a school trip, on the return flight I got pulled aside into a small room by whatever the german equivalent of TSA is and they swabbed the skin of my belly, and the inside of my bag. I'm guessing it was a drugs check and I must have just looked shifty because I get nervous in situations like that, but I do find it funny that they pulled me aside instead of the guys with me who almost certainly had something on them.
Also my partner has told me that apparently my armpits sometimes smell of weed or beer, despite me not coming in contact with either of those for a very long time, and now I definitely don't want to get taken into a small room by a TSA person (After some googling, apparently those smells can be associated with high stress)
I already adjust my clothing choices when flying to account for TSA's security theater make-work bullshit. Wonder how long before I'm doing that when preparing to go to other public places.
(I suppose if I attended pro sports games or large concerts, I'd be doing it for those, too)
I was getting pulled out of line in the 90’s for having long hair. I don’t dress in shitty clothes or fancy ones, I didn’t look funny, just the hair, which got regular compliments from women.
I started looking at people trying to decide who looked juicy to the security folks and getting in line behind them. They can’t harass two people in rapid succession. Or at least not back then.
The one I felt most guilty about, much later, was a filipino woman with a Philippine passport. Traveling alone. Flying to Asia (super suspect!). I don’t know why I thought they would tag her, but they did. I don’t fly well and more stress just escalates things, so anything that makes my day tiny bit less shitty and isn’t rude I’m going to do. But probably her day would have been better for not getting searched than mine was.
Getting pulled aside by TSA for secondary screening is nowhere in the ball park of being rushed at gunpoint as a teenager and told to lay down on the ground where one false move will get you shot by a trigger happy cop that probably won’t face any consequences - especially if the innocent victim is a Black male.
In fact, they will probably demonize the victim to find sn excuse why he deserved to get shot.
I wasn't implying TSA-cargo-pant-groping is comparable. My point is to show escalation in public facing systems. We have been dealing with TSA. Now we get AI Scanners. What's next?
I'm sure CLEAR is already having giddy discussion on how they can charge you to get pre-verified access to walk around in public. We can all wear CLEAR certified dog tags so the cops can hastle the non-dog-tagged people.
I got pulled aside because I absentmindedly showed them my concealed carry permit, not my driver's license. I told them I was a consultant working for their local government and was going back to Austin. No harm no foul.
If the system used any kind of logic whatsoever a CCW permit would not only allow you to bypass airport security but also carry in the airport (Speaking as both a pilot and a permit holder)
Would probably eliminate the need for the TSA security theater so that will probably never happen.
This may be mean, but we should really be careful about just handing AI over to technically illiterate people. They're far more likely to blind trust the LLM/AI output than someone who may be more experienced and take a beat. AI in an agentic-state society (what we have in America at least) is an absolute ticking time bomb. Honestly, this is what AI safety teams should be concentrated on: making sure people who think the computer is infallible understand that, no, it isn't, and you shouldn't just assume what it tells you is correct.
It's basically a failure of setting up the proper response playbook.
Instead of:
1. AI detects gun on surveillance
2. Dispatch armed police to location
It should be:
1. AI detects gun on surveillance
2. Human reviews the pictures and verifies the threat
3. Dispatch armed police to location
I think the latter version is likely what already took place in this incident, and it was actually a human that also mistook a bag of Doritos for a gun.
But that version of the story is not as interesting, I guess.
He could have been easily murdered. It's not the first time by a far margin that a bunch of overzealous cops murder a kid. I would never ever in my life set foot in a place that sends me armed cops so easily. That school is extremely dangerous.
I think the reason the school bought this silly software is because it's a dangerous school, and they're grasping at straws to try and fix the problem. The day after this false positive, a student was robbed.[1] Last month, a softball coach was charged with rape and possession of child pornography.[2] Last summer, one student was stabbed while getting off the bus.[3] Last year, there were two incidents where classmates stabbed each other.[4][5]
That certainly sounds bad, but it's all relative; keep in mind this school is in Baltimore County, which is distinct from the City of Baltimore and has a much different crime profile. This school is in the exact same town as Eastern Tech, literally the top high school in Maryland.
Based on your reporting, that's one violent crime per year, and one alleged child rapist. [0]
The crime stats seem fine to me. In a city like Baltimore, the numbers you've presented are shockingly low. When I was going through school, it was quite common for bullies to rob kids... even on campus. Teachers pretty much never did anything about it.
[0] Maybe the guy is a rapist, and maybe he isn't. If he is, that's godawful and I hope he goes to jail and gets his shit straight.
> Omnilert later admitted the incident was a “false positive” but claimed the system “functioned as intended,” saying its purpose is to “prioritize safety and awareness through rapid human verification.”
> Baltimore County Public Schools echoed the company’s statement in a letter to parents, offering counseling services to students impacted by the incident.
If false positives are known to happen, then you design a system where the image is vetted before telling the cops the perpetrator is armed. The company is basically swatting, but I'm sure they'll never be held liable.
Actually, if a system has too many false positives or false negatives, it's basically useless. There will eventually be doubts amongst the operators of it and the whole thing will implode, which is the best possible outcome.
We already went through this years ago with all those terrorism databases and we (humanity) have learned nothing--any database will have a percentage of erroneous data, it is impossible to eliminate erroneous data completely. Therefore, any database used to identify <fill in the blank> will have erroneous conclusions. It's been observed over and over again and governments can't help themselves that "this time it will be different because <fill in the blank> e.g. AI.
That was my first thought as well. A worry is police officers make mistakes which leads to anywhere from hapless people getting terrorized, harmed or killed. The bad thing about AI is it'll allow police to escape responsibility. Perhaps also where a human will realize it made a mistake they can admit it and everything is okay. But if AI says you had a gun, it won't walk that back. AI said he had a gun. But when we checked, he didn't have it anymore.
In the Menezes case the cops were playing a game of telephone that ended up with him being shot in the head.
Not sure nothing will happen. Some trial lawyers would love to sue a city, a school system, and an AI surveillance company over "trauma, anxiety, and mental pain and suffering" caused by the incident. There will probably be a settlement that nobody ever hears about.
Law enforcement officers, judicial officials, social workers, and similar generally maintain qualified immunity from liability in the course of their work. This case for example in which judges and social workers allegedly failed to properly assess a mother's fitness for child custody despite repeated indicators suggesting otherwise. The child was ultimately placed in the mother's care, and later was killed in an execution style (not due to negligence).
This case happened in the county I reside in and my sister-in-law is an attorney for the county in CP, although this was not her case directly. I can tell you what led to this: The COVID lockdowns! They stopped doing all the usual home visits and follow ups because everyone was too scared to do their jobs.
This case was a horrifying failure of the entire system that up until that point had fairly decent results for children who end up having to be taken away from their parents and later returned once the Mom/Dad clean up their act.
Not applicable - As a society we’ve countless times chosen to favour the right of the mother to keep children above the rights of other humans. Most children are killed in the home of the mother (i.e. either by the mother, or where partner choice would have avoided that, while the father was available), or even worde in the Anders Breivik situation (father available with stable job and perspectives in life, but custody refused, child grew up a mass murderer as always).
The world is doing fairly ok, thank you. The US however I’m not so sure as people here are apparently more concerned by the AI malfunction than with the idea it’s somehow sensible to live monitor high schools for gun threat.
Its not just the US. China runs the same level of surveillance, its being implemented all throughout Europe, Africa and Asia. This is becoming the norm.
Because if the "gun threat" system isn't accurate, then it's a system for false positives and false negatives and it's actually worse than having no such system. Maybe that's what you meant?
A bunch of companies and people invested unimaginable amounts of money in these technologies in the hope they will multiply that money. They will showe it down our throats no matter what, this isn't about security and making the world a better place, saving lives or preventing bad things to happen, this is strictly about those people and companies making as much money as possible, or at least for now not losing the money they invested.
The school admin has no understanding of the tech and only the dimmest comprehension of what happened. Asking them to do anything besides what the tech company told them to do is asking wayyy too much.
We blame AI here but what's up with law enforcment that comes with loaded guns in hand and send someone to the ground and cuff him before actually doing any check?
That is the real issue.
Police force anywhere else in the world that know how to behave would have approched the student, have had a small chat with him, found out all he had in hands was a bag of doritos, maybe would have asked politely to see the content of his bag, explaining the search has been triggered by an autodetection system that may lead to occasional errors and wished him a good day.
The guidance counselor does not have the training or time to "fix" the trauma you just gave this kid and his friends. Insane to put minors through this.
I wonder if the AI correctly identified it as a bag of Doritos, but was also trained on the commercial[0] where the bag appears to beat up a human (his fault for holding on too tight) and then it destroys an alien spacecraft.
An alert by one of these AI tools, which from what I understand have a terrible track record, should not be reasonable suspicion or probable cause to swarm a teenager with guns drawn. I wish more people in local communities would understand how much harm this type of surveillance and response causes. Our communities should not be using these tools.
Companies, particularly in the US, very much want to go with (2) and part of the reason they can is because there are zero consequences for incidents like this.
A couple ofexamples spring to mind:
1. the UK Royal Mail scandal where a bad system accused postmasters of theft, some of whom committed suicide over the allegations. Those allegations were later proven false and it was the system's fault. IMHO the people who signed off and deployed this should be charged with negligent homicide; and
2. The Hertz case where people who had returned cars were erroneously flagged as car thieves and report was made to police. This created hell for people who would often end up with warrants they had no idea about and would be detained on random traffic stops over a car that was never stolen.
Now these aren't AI but just like the Doritos case here, the principle is the same: companies are trying to replace people with computers. In all cases, a human should be responsible for reviewing any such complaint. In the Hertz case, a human should check to see if the car is actually stolen.
In the Royal Mail situation, the system needs to show its work. Deployment should be against the existing accounting system and discrepancies between the two need to be investigated for bugs until the system is proven correct. Particularly in the early stages, a forensic accountant (if necessary) should verify that funds were actually stolen before filing a criminal complaint.
And if "false positive" criminal complaints are filed, the people who allowed that to happen, if negligent (and we all know they are), should themslves be criminally charged.
We are way too tolerant of black box systems that can result in significant harm or even death to people. Show your work. And make a human put their name and reputation to any output of such systems.
The core of the issue is that many Americans do carry weapons which means that whatever the security system, it needs to keep in mind that the suspect might be armed and about to start shooting. This makes the police biased towards escalation because the only way against a shooter is to shoot first.
This problem doesn't exist in Europe or Japan because guns aren't that ubiquitous, which means that the police have the time to think before they act, which makes them less likely to escalate and start shooting. Obviously, for Americans, the only solution is to get rid of the gun culture, but this will never happen, so suck it up that AI gets you swatted.
I don't have kids yet, but I may someday. I went to public school myself, and would prefer to send any kid of mine to public school as well. (I'm not hard against private schools, but I'd prefer my kid gets to make friends from all walks of life, not just people who have parents who can afford private school.)
But I really wouldn't want to send my kid to a school that surveils students all the time, and uses garbage software like this that directly puts kids into dangerous situations. I feel like with a private school, I'd have more choice and ability to influence that sort of thing.
>> Omnilert later admitted the incident was a “false positive” but claimed the system “functioned as intended,” saying its purpose is to “prioritize safety and awareness through rapid human verification.”
No. If you're investigating someone and have existing reason to believe they are armed then this kind of false positive might be prioritizing safety. But in a general surveillance of a public place, IMHO you need to prioritize accuracy since false positives are very bad. This kid was one itchy trigger-pull away from death over nothing - that's not erring on the side of safety. You don't have to catch every criminal by putting everyone under a microscope, you should be catching the blatantly obvious ones at scale though.
The perceived threat of government forces assaulting and potentially killing me for reasons i have no control over, this is the kind of stuff that terminates the social contract. I'd want a new state that protects me from such stuff.
Looks like per their website it did function as intended... It surfaces potential threats for the school to look at and make a human decision. The principal decided to send the police after the school safety team dismissed it as part of the correct process. I mean fire alarms go off for lots of things that are not fire alarms... This was an alert meant to be validated by a human that messed up.
> Omnilert later admitted the incident was a “false positive” but claimed the system “functioned as intended,” saying its purpose is to “prioritize safety and awareness through rapid human verification.”
This exact scenario is discussed in [1]. The "human in the loop" failed, but we're supposed to blame the human, not the AI (or the way it was implemented). The humans serve as "moral crumple zones".
"""
The emphasis on human oversight as a protective mechanism allows governments and vendors to have it both ways: they can promote an algorithm by proclaiming how its capabilities exceed those of humans, while simultaneously defending the algorithm and those responsible for it from scrutiny by pointing to the security (supposedly) provided by human oversight.
"""
The article doesn't confirm that there was definitely a human in the loop, but it sorta suggests that police got a chance to manually verify the photo before going out to harass this poor kid.
I suspect, though, that the AI flagging that image heavily influenced the cop doing the manual review. Without the AI, I'd expect that a cop manually watching a surveillance feed would have found nothing out of the ordinary, and this wouldn't have happened.
So I agree that it's weird to just blame the human in the loop here. Certainly they share blame, but the fact of an AI model flagging this sort of thing (and doing an objectively terrible job of it) in the first place should take most of the blame here.
Q: What was the name of the Google AI Ethicist who was fired by Google for raising the concern that AI overwhelmingly negatively framed non-white humans as threats .. Timnit Gebru
We, as technologists, ARE NOT DOING BETTER. We must do better, and we are not on the "DOING BETTER" trajectory.
We talk about these "incidents" with breathless, "Wwwwellll if we just train our AI better ..." and the tragedies keep rolling.
Q2: Which of you has had a half dozen Squad Cars with Armed Police roll up on you, and treat you like you were a School Shooter? Not me, and I may reasonably assume it's because I am white, however I do eat Doritos.
> “They showed me the picture, said that looks like a gun, I said, ‘no, it’s chips.’”
So AI did the initial detection, but police looked at it and agreed. We don't see the image, but it probably did look like a gun because of a weird shadow or something.
Fundamentally, this isn't really any different from a person seeing someone with what looks like a gun and calling the cops, only it turns out the person didn't see it clearly.
The main issue is just that with increased numbers of images, there will be an increase in false positives. Can this be fixed by including multiple images, e.g. from motion of the object, so police (and the AI) can better eliminate false positives before traumatizing some poor teen?
Picture? Images? But those are just frames of footage the cameras have captured! Why would one purposefully use less information to make a decision rather than more?
Just put the full footage in front of an unbiased third party for a multi-stage verification first. The problem space isn't "is that weird shadow in the picture a gun or not?" it's "does the kid in the video have a gun?". It's not hard to figure out the difference between a bag of chips and a gun based on body language. Presumably the kid ate chips out of the bag? Using certain motions that one makes when doing that? Presumably the kids around him all saw the object in his hands and somehow did not react as if it was a gun? Jeez.
Those are a lot of "presumablies". Maybe you're right. Or maybe it was mostly obscured so you really couldn't tell. How do you know it was open and he was eating? How do you know there were other kids around and he wasn't solo? Why do you think the body language would be so different? Nobody is claiming he was using a gun or threatening anyone with it. If you're just carrying something in your hand, I don't know how you could tell what the object is or isn't from body language.
> So AI did the initial detection, but police looked at it and agreed. We don't see the image, but it probably did look like a gun because of a weird shadow or something.
Not sure I agree. The AI flagging it certainly biased the person doing the manual review toward agreeing with the AI's assessment. I can imagine a scenario where there was no AI involved, just a human watching that same surveillance feed, and (correctly) not seeing anything alarming in it.
Also I expect the AI completely failed at context. I wouldn't be surprised if the full video feed, a few minutes (or even seconds) before the flagged frame, shows the kid crumpling up the empty Doritos bag and stuffing it in his pocket. The AI probably doesn't keep all that context around to use when making a later decision, and giving just the flagged frame of video to the human may have caused them to miss out on important context.
Even better, share the frame(s) that the guess was drawn from with a human for verification before triggering ANYTHING. How much trouble could that possibly be? How many "guns" is this thing detecting in a day across all sites? I doubt more than a couple or we'd have heard about tons of incidents, false positives or not.
I don't find that especially good as a sole remedy, because lots of people are stupid. If they see a green outline box overlaid on a video image with the label 'gun', many many people will just respond to the label instead of looking at the underlying image and trying to make a decision. Probability and validation history need to be built into the product so that there are audit logs that can be pored over and challenged. Bad human decision-making, which is rampant, is always smoothed over with justifications like 'I was concerned for everyone's safety', and usually treated in isolation rather than assessed longitudinally.
It’s unsurprising, since this kind of classification is only as good as the training data.
And police do this kind of stuff all the time (or in the very least you hear about it a lot if you grew up in a major city).
So if you’re gonna automate broken systems, you’re going to see a lot more of the same.
I’m not sure what the answer is but I definitely feel that “security” system like this that are purchased and rolled out need to be highly regulated and be coupled with extreme accountability and consequences for false positives.
Everything around us: political tumult and weaponization of the justice system, ICE and other capricious projections of federal authority, the failure of drug prohibition, and on and on and on, points to a very simple solution:
Abolish SWAT teams. Do away with the idea that the state employees can be permitted to be more armed than anyone else.
Blaming the so-called 'swatter' (whether it's a human or AI) is really not getting at the root of the problem.
Exactly. I wonder if this a purpose-built image-recognition system, or is it a lowest-possible effort generic image model trained on the internet? Classifying a Black high school student holding Doritos as an imminent shooting threat certainly suggests the latter.
"I am invoking my 4th and 5th amendment rights afforded to me by the Constitution of the United States of America. I have no further comment until I have consulted with and am in the presence of my legal council."
Then, just sit back and enjoy as the lawsuit unfolds.
When people wonder how can AI mistake a bag of snacks as a weapon, simply answer "42"
It is about the question, the answer will become very clear once you understand what was the question presented to the inference model, and of course what data and context was fed
Inflicting trauma on a harmless human in the name of the "safety of others" is never ok. The victim here was not unharmed, but is likely to end up with PTSD and all the mental health issues that come with it.
Imagine the head scratching that's going on with execs who are surprised things might work when a probabilistic software is being used for deterministic purposes without realizing there's a gap between it kind of by nature.
I'm sure there will be no head scratching. They already know that this can happen, and don't care, because they know that if someone gets killed because of it, they won't be held responsible. And may not even lose any customers.
the best part of the technocracy is that they're not actually all that good at anything. the second best part is that when their mistakes end in someone dead there will be some way that they're not responsible.
At least there is a check done by humans in a human way. What if this human check is removed in future, as AI decisions would be deemed no longer requiring a human inspection?
If these AI video based gun detectors are not a massive fraud I will eat one.
How on Earth does a person walk with a concealed gun? What does a woman in a skirt with one taped to her thigh walk like?
What does a man in a bulky sweatshirt with a pistol on his back walk like?
What does a teenager in wide legged cargo jeans with two pistols and a extra magazines walk like?
The brochure linked from TFA has a screenshot of a combination of segmentation and object recognition models which are fairly standard in NVRs. Quick skim of the vendor website seems to confirm this[1] and states a claim that they are not analyzing the gait.
The whole idea even accepting the core premise is OK to begin with needs to have a similar analysis applied to it that medical tests do: will there be enough false positives, with enough harm caused by them, that this is actually worse than doing nothing? Compared with likelihood of improving an outcome and how bad a failure to intervene is on average, of course.
Given that there's no relevant screening step here and it's just being applied to everyone who happens to be at a place it's truly incredible that such an analysis would shake out in favor of this tech. The false positive rate would have to be vanishingly tiny, and it's simply not plausible that's true. And that would have to be coupled with a pretty low false negative rate, or you'd need an even lower false positive rate to make up for how little good it's doing even when it's not false-positiving.
So I'm sure that analysis was either deliberately never performed, or was and then was ignored and not publicized. So, yes, it's a fraud.
(There's also the fact that as soon as these are known to be present, they'll have little or no effect on the very worst events involving firearms at schools—shooters would just avoid any scheme that involved loitering around with a firearm where the cameras can see them, and count on starting things very soon after arriving—like, once you factor in second order effects, too, there's just no hope for these standing up to real scrutiny)
The model seems pretty shitty. Does it only look on a frame-by-frame basis? Literally one second of video context and it would never make that mistake.
I can understand the outrage in this thread but literally none of what you are all calling for will be done. No one from justice or law reads HN to see what should be done. I wish folks here would keep a cooler head rather than posting lengthy rants and vents that call for punishing school staff. Really unprofessional and immature from a community that prides itself, to fall constantly into a cycle of vitriol.
Can someone outline a more pragmatic, if not likely, course of what happens next after this? Is it swept under the rug and we move on?
It wasn't sour cream and onion and didn't contain cash, so it's super sus.
But really this is typical of cop overreaction with escalation and ego rather than calm, legal, and reasonable investigation. Karens may SWAT people they don't like, but it's police officers who must use reasonableness and restraint to defend the vestiges of their impartiality and community confidence based on asking questions and gathering evidence in a legal and appropriate manner rather than rushing to conclusions. Case in point: The NYC rough false arrest of a father in front of his kid to retrieve his mis-delivered package where the egomaniacal bully cop aggressively lectures a guy for his own mistake to cover his own ego while blaming the victim: https://youtu.be/LXd-4HueHYE
With high level of hallucination, cops need to tranquilizers more. If the student had reached for his bag just before the cops arrived, BLM 2.0 would have started.
T0 be fair most commercials for Doritos, Skittles, Mentos, etc., if occurring in real life, would result in a strong police response just after they cut away.
AI is a false (political) wish, it can and never work, it is the desperation of an over extended power structure
to hold on and permanently consolodate controll of all of the worlds population, and nothing else.
the proofs are there.
philosophers mulled this over long ago and made clear statements as to why ai cant work
though not that for a second do I misdunderstand that it is "all in"
for ai, and we all get to go for the 100 trillion dollar ride to hell.
can we have truely awsome automation for manufacturing and mundane beurocratic tasks?, fuck ya we can!
but anything that requires understanding, is forever out of reach, which unfortunatly is also lacking in the people pushing, this thing, now
Before I clicked the article, I said to myself "The victim's gotta be Black", and lo and behold.
AI has inherited police's (shitty, racist, and dangerous) idea that any Black person is a dangerous monster for whom anything is a weapon.
You're free to (attempt to) amend the Second Amendment, but the Supreme Court of the United States has already affirmed and reaffirmed that individual ownership of firearms in common use is a right.
What do you propose that is "reasonable" given the frameworks established by Heller, McDonald, Caetano, and Bruen?
I can 3D print or mill basically every item you can imagine prohibiting at home: what exactly do laws do in this case?
> the Supreme Court of the United States has already affirmed and reaffirmed that individual ownership of firearms in common use is a right.
The current interpretation of 2A is actually a fairly recent invention; in the past, it's been interpreted much more narrowly. And if SCOTUS can overturn Roe v. Wade's precedent, they can do the same with their interpretation of 2A. They won't of course, at least not until some of its members age out and get -- hopefully -- replaced with people who aren't idiots.
But I'd be fine if 2A was amended away. Let the states make whatever gun laws they want, and we can see whether blue or red states end up with lower levels of gun violence as a result.
If it's taking images every 30 seconds, it's getting 86400 x 30 = 2.5 million images per day per camera. So when it causes enormous, unnecessary trauma to one student per week, the company can rightfully claim it has less than a 1 in 10 million false positive rate.
I was unduly surprised and disappointed when I saw the photo of the kid and he turned out to be black. I would love to believe that this had no impact on how the whole thing played out, but I don't.
All right they’ve gotta have a plain clothes bro go up there make sure the kid is chill. You know the difference between a murder and not can be as little as somebody being nice
Sounds like this high school is doing a great job preparing students for the real world, where they can be swarmed by jackbooted thugs at any moment for any reason.
>Omnilert later admitted the incident was a “false positive” but claimed the system “functioned as intended,” saying its purpose is to “prioritize safety and awareness through rapid human verification.”
It's ok everyone, you're safer now that police are pointing a gun at you, because of a bag of chips ... just to be safe.
/s
Absolutely ridiculous. We're living "computer said you did it, prove otherwise, at gunpoint".
AI singularity wil happen, but motherbrain as a complete moron. It will extinguish humans not as a grand plan for machines to take over, but doing horrible mistakes when trying to make things better.
If any of you had actually paid attention to the source media, you would have noticed that they were explicitly dystopias. They were always clearly and explicitly hell for normal people trying to live life.
Meanwhile, tons of you watched star trek and apparently learned(?) that the "bright future" it promised us was.... talking computers? And not, you know, post scarcity and enlightenment that allowed people to focus on things that brought them joy or they were good at, and an entire elimination of the concept of "capitalism" or personal profit or resource disparity that could allow people to not be able to afford something while some asshole in the right place at the right time gets to take a percentage cut of the entire economy for their personal use.
The primary "technology" of star trek was socialism lol.
The safest thing to do is to pull all Frito Lay products off shelves until the packaging can be redesigned to ensure that AI never confuses them for guns. It's a liability issue. THINK OF THE CHILDREN.
I think it's almost guaranteed that this model has race-related biases, so no, I don't think you're kidding at all. I think it's entirely likely that an Asian (or white) kid of the same build, wearing the same clothes, with a crumpled-up bag of Doritos in his pocket, would not get flagged as having a gun.
> Omnilert later admitted the incident was a “false positive” but claimed the system “functioned as intended,” saying its purpose is to “prioritize safety and awareness through rapid human verification.”
It just created a situation in which a bunch of people with guns were told that some teen had a gun. That's a very unsafe situation that the system created, out of nothing.
And some teen may be traumatized. Again, unsafe.
Incidentally, the article's quotes make this teen sound more adult than anyone who sold or purchased this technology product.
https://www2.ljworld.com/news/schools/2025/aug/07/lawrence-s...
Another false positive by one of these leading content filters schools use - the kid said something stupid in a group chat and an AI reported it to the school, and the school contacted the police. The kid was arrested, stripped searched, and held for 24 hours without access to their parents or counsel. They ultimately had to spend time in probation, a full mental health evaluation, and go to an alternative school for a period of time. They are suing Gaggle, who claims they never intended their system to be used that way.
These kinds of false positives are incredibly common. I interviewed at one of their competitors (Lightspeed), and they actually provide a paid service where they have humans review all the alerts before being forwarded to the school or authorities. This is a paid addon, though.
https://archive.is/DYPBL
> Gaggle’s CEO, Jeff Patterson, said in an interview that the school system did not use Gaggle the way it is intended. The purpose is to find early warning signs and intervene before problems escalate to law enforcement, he said. “I wish that was treated as a teachable moment, not a law enforcement moment,” said Patterson.
It's entirely predictable that schools will call law enforcement for many of these detections. You can't sell to schools that have "zero tolerance" policies and pretend that your product won't trigger those policies.
12 replies →
> They are suing Gaggle, who claims they never intended their system to be used that way.
Yeah, there's a shop near me that sells bongs "intended" for use with tobacco only.
The kid was arrested, stripped searched, and held for 24 hours without access to their parents or counsel. They ultimately had to spend time in probation, a full mental health evaluation, and go to an alternative school for a period of time.
All he wanted was a Pepsi. Just one Pepsi. And they wouldn't give it to him.
> They are suing Gaggle, who claims they never intended their system to be used that way.
Is there some legal way to sue a pair of actors (Gaggle and school) then let them sue each other over who has to pay what percentage?
1 reply →
> This is a paid addon, though
Holy shitballs. In my experience such pain addons have very cheap labor attached to them, certainly not what you would expect based on the sales pitch.
1 reply →
>...its purpose is to “prioritize safety and awareness through rapid human verification.
Oh look, a corporation refusing to take responsibility for literally anything. How passe.
The invention of the corporation is virtually to eliminate responsibility/culpability from any individual.
Human car crash? Human punishment. Corporate-owned car crash? A fine which reduces salaries some negligible percent.
21 replies →
> a corporation refusing to take responsibility for literally anything. How passe
Versus all the natural people at the highest echelons of our political economic system valiantly taking responsibility for fuckall?
8 replies →
Engineer: hey I made this cool thing that can help people in public safety roles process information and make decisions more efficiently! It gives false positives but you save more time than it takes less time to weed through them.
Someone nearby: well what if they use it to replace human thinking instead of augment it?
Engineer: well they would be ridiculous. Nobody would ever think that’s a good idea.
Marketing Team: it seems like this lands best when positioning it as a decision-making tool. Let’s get some metrics on how much faster it is at making decisions than people are.
Sales Rep: ok, Captain, let’s dive into our flagship product, DecisionMaker Pro, the totally automated security monitoring agent…
::6 months later—some kid is being held at gunpoint over snacks.::
Nice fantasy, but the reality is that the "people in public safety roles" love using flimsy pretenses to harass and abuse vulnerable populations. I wish it was just overeager sales and marketing, but you're view of humanity is way too naive especially as masked thugs are disappearing people in the street as we type.
1 reply →
Refer to the post office scandal in Britain and the robodebt debacle in Australia.
The authorities are just itching to have their brains replaced with by dumb computer logic, without regard for community safety and wellbeing.
2 replies →
It’s actually “AI swarmed” since no human reasoning, only execution, was exerted - basically have an AI directing resources.
delegating decision to AI, excluding human from the "human in the loop" is kind of unexpected as a first step, as in general it was expected that exclusion will start from the other end. Sideway i wonder how is that going to happen on the battlefield.
for this civilian use case the next step - AR googles worn by police with that AI will be projecting onto the googles where that teenager has his gun (kind of Black Mirror style), and the next step after that is obviously excluding the humans even from the execution step.
Reverse Centaur. MANNA.
1 reply →
when attacked by bees am I hive swarmed?
In any system, there are false positives and false negatives. In some situations (like a high recall disease detection) false negatives are much worse than false positives, because the cost of a false positive is a more rigorous screening.
But in this case both are bad. If it was a false negative students might need therapy for a more tragic reason.
Aside from improving the quality of the detection model, we should try to reduce the “cost” of both failure modes as much as possible. Putting a human in the loop or having secondary checks are ways to do that.
In this case false positives are far, far worse than false negatives. A false negative in this system does not mean a tragedy will occur, because there are many other preventative measures in place. And never mind the fact that this country refuses to even address the primary cause of gun violence in the first place: the ubiquity of guns in our society. So systems like this is what we end up with when we ignore to address the problem of guns and choose to deal the downstream effects of that instead.
8 replies →
> But in this case both are bad. If it was a false negative students might need therapy for a more tragic reason.
Given the probability of police officers in the USA taking any action as hostile and then ending up shooting him a false positive here is the same as swatting someone.
The system here sent the police off to kill someone.
7 replies →
I was swatted once. Girlfriend's house. Someone called 911 and said they'd seen me kill a neighbor, drag their body into the house, and was now holding my gf's family hostage.
We answered the screams at the door to guns pointed at our faces, and countless cops.
It was explained to us that this was the restrained version. We got a knock.
Unfortunately, I understand why these responses can't be neutered too much. You just never know.
30 replies →
False positives can effectively lead to false negatives too. If too many alarms end in teens getting swatted (or worse) for eating chips, people might ignore the alarm if an actual school shooter triggers it. Might assume the AI is just screaming about a bag of chips again.
I think a “true positive” is an issue as well if the protocol to manage it isn’t appropriate. If the kid was armed with something other than nacho cheese, the provocative reaction could have easily set off a tragic chain of events.
Reality is there are guns in schools every day. “Solutions” like this aren’t making anyone safer. School shooters don’t fit this profile - they are planners, not impulsive people hanging out at the social event.
More disturbing is the meh attitude of both the company and the school administration. They almost engineered a tragedy through incompetence, and learned nothing.
>And some teen may be traumatized.
Um. That's not really the danger here.
The danger is that it's as clear as day that in the future someone is gonna be killed. That's not just a possibility. It's a certainty given the way the system is set up.
This tech is not supposed to be used in this fashion. It's not ready.
Did you want to emphasize or clarify the first danger I mentioned?
My read of the "Um" and the quoting, was that you thought I missed that first danger, and so were disagreeing in a dismissive way.
When actually we're largely in agreement about the first danger. But people trying to follow a flood of online dialogue might miss that.
I mentioned the second danger because it's also significant. Many people don't understand how safety works, and will think "nobody got shot, so the system must have worked, nothing to be concerned about". But it's harder for even those people to dismiss the situation entirely, when the second danger is pointed out.
1 reply →
I fully agree, but we also really need to get to a place where drawing the attention of police isn't an axiomatically life-threatening situation.
5 replies →
Americans are killed by police all the time, and by other Americans. We've already decided as a society that we don't care enough to take the problem seriously. Gun violence, both public and from the state, is accepted as unavoidable and defended as a necessary price to pay to live in a free society[0]. Having a computer call the shots wouldn't actually make much of a difference.
Hell, it wouldn't even move the needle on racial bias much because LLMs have already shown themselves to be prejudiced against minorities due to the stereotypes in their training data.
[0]Even though no other free society has to pay that price but whatever.
7 replies →
> The danger is that it's as clear as day that in the future someone is gonna be killed.
This argument can be made about almost every technology, including fire, electricity, and even the base "technology" of just letting a person watch a camera feed.
I'm not downplaying the risks. I'm saying that we should remember that almost everything has risks and benefits, and as a society we decide for or against using/doing them based mostly on the ratio between those two things.
So we need some data on the rates of false vs. true detections here. (A value judgment is also required.)
5 replies →
"“We understand how upsetting this was for the individual that was searched as well as the other students who witnessed the incident,” the principal wrote. “Our counselors will provide direct support to the students who were involved.”"
Make them pay money for false positives instead of direct support and counselling. This technology is not ready for production, it should be in a lab not in public buildings such as schools.
Charge the superintendent with swatting.
Decision-maker accountability is the only thing that halts bad decision-making.
> Charge the superintendent with swatting.
This assumes no human verification of the flagged video. Maybe the bag DID look like a gun. We'll never know, because modern journalism has no interest in such things. They obtained the required emotional quotes and moved on.
17 replies →
> Make them pay money
It already cost money paying for the time and resources to be misappropriated.
There needs to be resignations, or jail time.
The taxpayers collectively pay the money, the officers involved don't (except for that small fraction of their income they pay in taxes that increase as a result).
I wonder how much more likely it is to get a false positive from a black student.
The question is whether that Doritos carrying kid is still alive, instead of being shot at by the violent cops (who typically do nothing when an actual shooter is roaming a school on a killing spree; apropos the Uvalve school shooting, when hundreds of cops milled around the school in full body armor, refusing to engage the shooter on killing spree inside the school, and they even prevented the parents from going inside to rescue their kids) based on a false positive about a gun (and the cops must have figured that it's likely a false positive, because it is info from AI surveillance), only because he is white?
Before clicking on the article, I kinda assumed the student was black. I wouldn't be surprised if the AI model they're using has race-related biases. On the contrary, I would be surprised if it didn't.
I assume they were provide gift cards good for psychotherapy sessions.
> Make them pay money for false positives instead of direct support and counselling.
Agreed.
> This technology is not ready for production
No one wants stuff like this to happen, but nearly all technologies have risks. I don't consider that a single false positive outweighs all of its benefits; it would depend on the rates of false and true positives, and the (subjective) value of each (both high in this case, though I'd say preventing 1 shooting is unequivocally of more value than preventing 1 innocent person being unnecessarily held at gunpoint).
I think I’ve said this too many times already but the core problem here and with the “AI craze” is that nobody realy wants to solve problems, what they want is a marketable product and AI seems to be the magic wrench that fits in all the nuts and since most of the people doesn’t really knows how it works and what are its limitations the happily buy the “magic dust”.
4 replies →
In the US cops kill more people than terrorists. As long as you quantifying values take that into account.
1 reply →
Stuff like this feels like some company has managed to monetize an open source object detection model like YOLO [1], creating something that could be cobbled together relatively easily, and then sold it as advance AI capabilities. (You'd hope they've have at least fine-tuned it / have a good training dataset.)
We've got a model out there now that we've just seen has put someone's life at risk... Does anyone apart from that company actually know how accurate it is? What it's been trained on? Its false positive rate? If we are going to start rolling out stuff like this, should it not be mandatory for stats / figures to be published? For us to know more about the model, and what it was trained on?
[1] https://arxiv.org/abs/1506.02640
And it feels like they missed the "human in the loop" bit. One day this company is likely to find itself on the end of a wrongful death lawsuit.
They’ll likely still be profitable after accounting for those. This is why sociopaths are so successful at business
[flagged]
I expect that John Bryan -- who produces content as The Civil Rights Lawyer https://thecivilrightslawyer.com -- will have something to say about it.
He frequently criticizes the legality of police holding people at gunpoint based on flimsy information, because the law considers it a use of force, which needs to be justified. For instance, he's done several videos about "high risk" vehicle stops, where a car is mistakenly flagged as stolen, and the occupants are forced to the ground at gunpoint.
My take is that if both the AI automated system and a "human in the loop" police officer looked at the picture and reasonably believed that the object in the image was a gun, then the stop might have been justified.
But if the automated system just sent the officers out without having them review the image beforehand, that's much less reasonable justification.
Someday there'll be a lawyer in court telling us how strong the AI evidence was because companies are spending billions of dollars on it.
Or they'll tell us police have started shooting because an acorn falls, so they shouldn't be expected to be held to higher standards and are possibly an improvement.
And there needs to be an opposing lawyer ready to tear that argument to pieces.
You mean in the same fallacious sense of "you can tell cigarettes are good because so many people buy them"?
4 replies →
The article says the police later showed the student the photo that triggered the alert. He had a crumpled-up Doritos bag in his pocket. So there was no gun in the photo, just a pocket bulge that the AI thought was a gun... which sounds like a hallucination, not any actual reasonable pattern-matching going on.
But the fact that the police showed the photo does suggest that maybe they did manually review the photo before going out. If that's the case, do wonder how much the AI influenced their own judgment, though. That is, if there was no AI involved, and police were just looking at real-time surveillance footage, would they have made the same call on their own? Possibly not: it feels reasonable to assume that they let the fact of the AI flagging it override their own judgment to some degree.
Is use of force without justification automatically excessive force or is there a gray area?
See https://en.wikipedia.org/wiki/Graham_v._Connor
Ah, the coming age of Palantir's all seeing platform; and Peter Thiel becoming the shadow Emperor. Too bad non-deterministic ml systems are prone to errors that risk lives when applied wrongly to crucial parts of society. But in an authoritarian state, those will be hidden away anyway, so there's nothing to see here: move along folks. Yes, surveillance and authoritarianism go hand in hand, ask China. It's important to protest these methods and push lawmakers to act against them; now, before it's too late.
I might be missing something but I don' think this article isn't about palantir or any of their products
Palantir is but one head of the hydra which has hundreds of them, and all concerns about a single one apply to the whole beast hundredfold.
1 reply →
You're absolutely right, Palantir just needs a different name and then they'd have no issues.
This comment has a double negative, which makes it a false positive.
The article is about omnialert, not palantir, but don’t let the facts get in the way of your soapbox rant.
Same fallible systems, same end goal of mass surveillance.
1 reply →
American, please, wake up. The masked border police are on the streets arresting citizens, the military is being paid as a client of the president, corruption is legal, and a mass surveillance machine unfathomable to prior dictatorships is being/has been established. You're fucked. Listen to the soapbox. It is very, very relevant. Wake up.
I'm pretty sure that some people will continue to apply the term "soapbox ranting" to all opposition against the technofascism even when victims of its false positives will be in need of coroners, not psychologists.
I dont think a guy who knows so much about the anti christ could be wrong.
[flagged]
Pre-emptive compliance out of fear - then my boy, the war is already lost.
6 replies →
Omnilert later admitted the incident was a “false positive” but claimed the system “functioned as intended,” saying its purpose is to “prioritize safety and awareness through rapid human verification.”
prioritize your own safety by not attending any location fitted with such a system, or deemed to be such a dangerous environment that such a system is desired.
the AI "swatted" someone.
Calling it today. This company is going to get innocent kids killed.
How many packs of Twizzlers, or Doritos, or Snickers bars are out there in our schools?
First time it happens, there will be an explosion of protests. Especially now that the public knows that the system isn't working but the authorities kept using it anyway.
This is a really bad idea right now. The technology is just not there yet.
And then there's plenty of bullies who might put a sticker of a picture of a gun on someone's back, knowing it will set off the image recognition. It's only a matter of time until they figure that out.
2 replies →
>First time it happens, there will be an explosion of protests.
Why do you believe this? In the US, cops will cower outside of a school with an armed gunman actively murdering children, forcibly detain parents who wish to go in if the cops wont, and then re-elect everyone involved
In the US, an entire segment of the population will send you death threats claiming you are part of some grand (democrat of course) conspiracy for the crime of being a victim of a school shooting. After every school shooting, republican lawmakers wear an AR-15 pin to work the next day to ensure you know who they care about.
Over 50% of the country blamed the protesting students at Kent state for daring to be murdered by the national guard.
Cops can shoot people in broad daylight, in the back, with no justification, or reasonable cause, or can even barge into entirely the wrong house and open fire on homeowners exercising their basic right to protect themselves from strangers invading their homes, and as long as the people who die are mostly black, half the country will spout crap like "They died from drugs" or "they once sold a cigarette" or "he stole skittles" or "they looked at my wife wrong" while the cops take selfies reenacting the murder for laughs and talk about how terrified they are by BS "training" that makes them treat every stranger as a wanted murderer armed to the teeth. The leading cause of death for cops is still like heart disease of course.
Trump sent unmarked forces to Portland and abducted people off the street into unmarked vans and I think we still don't really know what happened there? He was re-elected.
> The technology is just not there yet.
The technology literally can NEVER be there. It is completely impossible to positively identify a bulge in clothing as a handgun. But that doesn’t stop irresponsible salesmen from making the claim anyway.
The corporate version of "It's a feature, not a bug."
Clearly it did not prioritize human safety.
"rapid human verification." at gunpoint. The Torment Nexus has nothing on these AI startups.
Why did they waste time verifying? The police should have eliminated the threat before any harm could be done. Seconds count when you're keeping people safe.
I get that you're being sarcastic and find the police response appalling, but the sad reality of Poe's Law is that there are a lot of people who would unironically say this and would have cheered if the cops had shot this kid, either because they hate black people or because they get off on violence and police shootings are a social sanctioned way to indulge that taste.
We all know the cops will go for the easy prey:
* Even hundreds of cops in full body armor and armed with automatic guns will not dare to engage a single "lone wolf" shooter doing a killing spree in a school; the heartless cowards may even prevent the parents from going inside to rescue their kids: Uvalde school shooting incident
* Cop on a ego trip, will shoot down a clearly harmless kid calmly eating a burger in his own car (not a stolen car): Erik Cantu incident
* Cops are not there to serve the society, they are not there to ensure safety and peace for the neighborhood, they are merely armed militia to protect the rich and powerful elites: https://www.alternet.org/2022/06/supreme-court-cops-protect-...
3 replies →
The dispatch relayer and responding officers should at least have ready access to a screen where they can see a video/image of the raw footage that triggered the AI alert. If it is a false alarm, they will better see it and react accordingly, and if it is a real threat they will better understand the initial context and who may have been involved.
According to a news article[1], a human did review the video/image and flagged it as a false positive. It was the principal who told the school cop, who then called other cops:
> The Department of School Safety and Security quickly reviewed and canceled the initial alert after confirming there was no weapon. I contacted our school resource officer (SRO) and reported the matter to him, and he contacted the local precinct for additional support. Police officers responded to the school, searched the individual and quickly confirmed that they were not in possession of any weapons.
What's unclear to me is the information flow. If the Department of School Safety and Security recognized it as a false positive, why did the principal alert the school resource officer? And what sort of telephone game happened to cause local police to believe the student was likely armed?
1. https://www.wbaltv.com/article/student-handcuffed-ai-system-...
15 replies →
Next up, a captcha that verifies you're not a robot by swatting you and checking at gunpoint.
Walking through TSA scanners, I always get that unnerving feeling I will get pulled aside. 50% of the time they flag my cargo pants because of the zipper pockets - There is nothing in them but the scanner doesn't like them.
Now we get the privilege of walking by AI security cameras placed in random locations, hoping they don't flag us.
There's a ton of money to be made with this kind of global frisking, so lots of pressure to roll out more and more systems.
How does this not spiral out of control?
To be fair, at least you can choose not to wear the cargo pants.
A friend of mine once got pulled aside for extra checks and questioning after he had already gone through the scanners, because he was waiting for me on the other side to walk to the gates together and the agent didn't like that he was "loitering" – guess his ethnicity...
How is it fair to say that? That's some "why did you make me hurt you"-level justification.
3 replies →
> guess his ethnicity...
Not sure, but I bet they were looking at their feet kinda dusting off the bottoms, making awkward eye contact with the security guard on the other side of the one way door.
Speak up citizens!
Email the state congressman and tell them what you think.
Since (pretty much) nobody does this, if a few hundred people do it, they will sit up and take notice. It takes less people than you might think.
Since coordinating this with a bunch of strangers (I.e. the public) is difficult, the most effective way is to normalise speaking up in our culture. Of course normalising it will increase the incoming comm rate, which will slowly decrease the effectiveness but even post that state, it’s better than where we are, which is silent public apathy
If that's the case, why do people in Congress keep voting for things their constituents don't like? When they get booed at town halls they just dismiss it as being the work of paid activists.
1 reply →
Get Precheck or global entry. I only do a scanner every 5 years or so when I get pulled at random for it. Otherwise it's metal detector only. Unless your zippers have such chunky metal that they set that off you'll be fine. My belt and watch don't.
Note: Precheck is incredibly quick and easy to get; and GE is time consuming and annoying, but has its benefits if you travel internationallly. Both give the same benefits at TSA.
Second note: let's pretend someone replied "I shouldn't have to do that just to be treated...blah blah" and that I replied, "maybe not, but a few bucks could still solve this problem, if it bothers you enough that's worth it to you."
"Just pay to not be harrassed or have your rights/dignity stepped on" a typical take to find on the orange site.
...maybe not, but a few bucks could still solve this problem
Sure, can't argue with that. But doesn't it bug you just a little that (paying a fee to avoid harassment) doesn't look all that disimilar from a protection racket? As to whether it's a few bucks or many, now you're just a mark negotiating the price.
1 reply →
The brownshirts will never get my money.
1 reply →
I don't often fly, but back when I went to germany on a school trip, on the return flight I got pulled aside into a small room by whatever the german equivalent of TSA is and they swabbed the skin of my belly, and the inside of my bag. I'm guessing it was a drugs check and I must have just looked shifty because I get nervous in situations like that, but I do find it funny that they pulled me aside instead of the guys with me who almost certainly had something on them.
Also my partner has told me that apparently my armpits sometimes smell of weed or beer, despite me not coming in contact with either of those for a very long time, and now I definitely don't want to get taken into a small room by a TSA person (After some googling, apparently those smells can be associated with high stress)
I already adjust my clothing choices when flying to account for TSA's security theater make-work bullshit. Wonder how long before I'm doing that when preparing to go to other public places.
(I suppose if I attended pro sports games or large concerts, I'd be doing it for those, too)
I was getting pulled out of line in the 90’s for having long hair. I don’t dress in shitty clothes or fancy ones, I didn’t look funny, just the hair, which got regular compliments from women.
I started looking at people trying to decide who looked juicy to the security folks and getting in line behind them. They can’t harass two people in rapid succession. Or at least not back then.
The one I felt most guilty about, much later, was a filipino woman with a Philippine passport. Traveling alone. Flying to Asia (super suspect!). I don’t know why I thought they would tag her, but they did. I don’t fly well and more stress just escalates things, so anything that makes my day tiny bit less shitty and isn’t rude I’m going to do. But probably her day would have been better for not getting searched than mine was.
Getting pulled aside by TSA for secondary screening is nowhere in the ball park of being rushed at gunpoint as a teenager and told to lay down on the ground where one false move will get you shot by a trigger happy cop that probably won’t face any consequences - especially if the innocent victim is a Black male.
In fact, they will probably demonize the victim to find sn excuse why he deserved to get shot.
I wasn't implying TSA-cargo-pant-groping is comparable. My point is to show escalation in public facing systems. We have been dealing with TSA. Now we get AI Scanners. What's next?
Also, no need to escalate this into a race issue.
6 replies →
Why don’t you pay the bribe and skip the security theater scanner? It’s cheap. Most travel cards reimburse for it too.
I'm sure CLEAR is already having giddy discussion on how they can charge you to get pre-verified access to walk around in public. We can all wear CLEAR certified dog tags so the cops can hastle the non-dog-tagged people.
I got pulled aside because I absentmindedly showed them my concealed carry permit, not my driver's license. I told them I was a consultant working for their local government and was going back to Austin. No harm no foul.
If the system used any kind of logic whatsoever a CCW permit would not only allow you to bypass airport security but also carry in the airport (Speaking as both a pilot and a permit holder)
Would probably eliminate the need for the TSA security theater so that will probably never happen.
7 replies →
The TSA scanners also trigger easily on crotch sweat.
I enjoy a good grope, so I’ll keep that in mind the next time I’m heading into the us.
This may be mean, but we should really be careful about just handing AI over to technically illiterate people. They're far more likely to blind trust the LLM/AI output than someone who may be more experienced and take a beat. AI in an agentic-state society (what we have in America at least) is an absolute ticking time bomb. Honestly, this is what AI safety teams should be concentrated on: making sure people who think the computer is infallible understand that, no, it isn't, and you shouldn't just assume what it tells you is correct.
We already handed over the Internet to technically illetrate people long time ago.
It's basically a failure of setting up the proper response playbook.
Instead of:
1. AI detects gun on surveillance
2. Dispatch armed police to location
It should be:
1. AI detects gun on surveillance
2. Human reviews the pictures and verifies the threat
3. Dispatch armed police to location
I think the latter version is likely what already took place in this incident, and it was actually a human that also mistook a bag of Doritos for a gun. But that version of the story is not as interesting, I guess.
He could have been easily murdered. It's not the first time by a far margin that a bunch of overzealous cops murder a kid. I would never ever in my life set foot in a place that sends me armed cops so easily. That school is extremely dangerous.
I think the reason the school bought this silly software is because it's a dangerous school, and they're grasping at straws to try and fix the problem. The day after this false positive, a student was robbed.[1] Last month, a softball coach was charged with rape and possession of child pornography.[2] Last summer, one student was stabbed while getting off the bus.[3] Last year, there were two incidents where classmates stabbed each other.[4][5]
1. https://www.nottinghammd.com/2025/10/22/student-robbed-outsi...
2. https://www.si.com/high-school/maryland/baltimore-county-hig...
3. https://www.wbaltv.com/article/knife-assault-rossville-juven...
4. https://www.wbal.com/stabbing-incident-near-kenwood-high-sch...
5. https://www.cbsnews.com/baltimore/news/teen-injured-after-re...
That certainly sounds bad, but it's all relative; keep in mind this school is in Baltimore County, which is distinct from the City of Baltimore and has a much different crime profile. This school is in the exact same town as Eastern Tech, literally the top high school in Maryland.
2 replies →
That sounds to me like it's pretty close to the middle of the curve a large High School in the US.
3 replies →
Based on your reporting, that's one violent crime per year, and one alleged child rapist. [0]
The crime stats seem fine to me. In a city like Baltimore, the numbers you've presented are shockingly low. When I was going through school, it was quite common for bullies to rob kids... even on campus. Teachers pretty much never did anything about it.
[0] Maybe the guy is a rapist, and maybe he isn't. If he is, that's godawful and I hope he goes to jail and gets his shit straight.
>the system “functioned as intended,”
Behold - a real life example of a "Not a hotdog" system, except this one is gun / not-a-gun.
Except the fictional one from the series was more accurate...
I think the most amazing part is that the school doubled down on the mistake by parroting the corporate line.
I expect a school to be smart enough to say “Yes, this is a terrible situation, and we’re taking a closer look at the risks involved here.”
Lawyer's advice?
I would think "no comment" would be safer/smarter than "yeah, your kids are at risk of being shot by police by attending our school, deal with it".
1 reply →
Except they never say that.
> Omnilert later admitted the incident was a “false positive” but claimed the system “functioned as intended,” saying its purpose is to “prioritize safety and awareness through rapid human verification.”
> Baltimore County Public Schools echoed the company’s statement in a letter to parents, offering counseling services to students impacted by the incident.
(Emphasis mine)
If false positives are known to happen, then you design a system where the image is vetted before telling the cops the perpetrator is armed. The company is basically swatting, but I'm sure they'll never be held liable.
Actually, if a system has too many false positives or false negatives, it's basically useless. There will eventually be doubts amongst the operators of it and the whole thing will implode, which is the best possible outcome.
We already went through this years ago with all those terrorism databases and we (humanity) have learned nothing--any database will have a percentage of erroneous data, it is impossible to eliminate erroneous data completely. Therefore, any database used to identify <fill in the blank> will have erroneous conclusions. It's been observed over and over again and governments can't help themselves that "this time it will be different because <fill in the blank> e.g. AI.
Memories of Jean Charles de Menezes come to mind: https://en.wikipedia.org/wiki/Killing_of_Jean_Charles_de_Men...
That was my first thought as well. A worry is police officers make mistakes which leads to anywhere from hapless people getting terrorized, harmed or killed. The bad thing about AI is it'll allow police to escape responsibility. Perhaps also where a human will realize it made a mistake they can admit it and everything is okay. But if AI says you had a gun, it won't walk that back. AI said he had a gun. But when we checked, he didn't have it anymore.
In the Menezes case the cops were playing a game of telephone that ended up with him being shot in the head.
And Amadou Diallo:
<https://en.wikipedia.org/wiki/Killing_of_Amadou_Diallo>
What is happening in the world. There should be some liability for this but nothing will happen.
Not sure nothing will happen. Some trial lawyers would love to sue a city, a school system, and an AI surveillance company over "trauma, anxiety, and mental pain and suffering" caused by the incident. There will probably be a settlement that nobody ever hears about.
A settlement paid by the taxpayers with no impact at all on anyone actually responsible.
Law enforcement officers, judicial officials, social workers, and similar generally maintain qualified immunity from liability in the course of their work. This case for example in which judges and social workers allegedly failed to properly assess a mother's fitness for child custody despite repeated indicators suggesting otherwise. The child was ultimately placed in the mother's care, and later was killed in an execution style (not due to negligence).
https://www.youtube.com/watch?v=wzybp0G1hFE
This case happened in the county I reside in and my sister-in-law is an attorney for the county in CP, although this was not her case directly. I can tell you what led to this: The COVID lockdowns! They stopped doing all the usual home visits and follow ups because everyone was too scared to do their jobs.
This case was a horrifying failure of the entire system that up until that point had fairly decent results for children who end up having to be taken away from their parents and later returned once the Mom/Dad clean up their act.
Not applicable - As a society we’ve countless times chosen to favour the right of the mother to keep children above the rights of other humans. Most children are killed in the home of the mother (i.e. either by the mother, or where partner choice would have avoided that, while the father was available), or even worde in the Anders Breivik situation (father available with stable job and perspectives in life, but custody refused, child grew up a mass murderer as always).
1 reply →
The world is doing fairly ok, thank you. The US however I’m not so sure as people here are apparently more concerned by the AI malfunction than with the idea it’s somehow sensible to live monitor high schools for gun threat.
Its not just the US. China runs the same level of surveillance, its being implemented all throughout Europe, Africa and Asia. This is becoming the norm.
Because if the "gun threat" system isn't accurate, then it's a system for false positives and false negatives and it's actually worse than having no such system. Maybe that's what you meant?
So you’re okay with trigger happy cops forcing a teenager to the ground because he had a bag of Doritos?
2 replies →
A bunch of companies and people invested unimaginable amounts of money in these technologies in the hope they will multiply that money. They will showe it down our throats no matter what, this isn't about security and making the world a better place, saving lives or preventing bad things to happen, this is strictly about those people and companies making as much money as possible, or at least for now not losing the money they invested.
[flagged]
[flagged]
The people building these things are good friends with the bullies and scammers now.
> “They didn’t apologize. They just told me it was protocol. I was expecting at least somebody to talk to me about it.”
I wonder how effective an apology and explanation would have been? Just some respect.
The school admin has no understanding of the tech and only the dimmest comprehension of what happened. Asking them to do anything besides what the tech company told them to do is asking wayyy too much.
Effective at what? No one is facing any consequences anyway.
More's the pity. The school district could use some consequences.
Except for the kids who experienced the “rapid human verification” firsthand.
2 replies →
We blame AI here but what's up with law enforcment that comes with loaded guns in hand and send someone to the ground and cuff him before actually doing any check?
That is the real issue.
Police force anywhere else in the world that know how to behave would have approched the student, have had a small chat with him, found out all he had in hands was a bag of doritos, maybe would have asked politely to see the content of his bag, explaining the search has been triggered by an autodetection system that may lead to occasional errors and wished him a good day.
Or they'd have looked at whatever images the system had used for its decision and see it was a false positive without having to send anyone over.
>That is the real issue.
No. Trusting AI is clearly the issue.
If there was a 9-1-1 call to the police that there was an active shooter at your kids school, how would you want the police to show up?
The guidance counselor does not have the training or time to "fix" the trauma you just gave this kid and his friends. Insane to put minors through this.
I wonder if the AI correctly identified it as a bag of Doritos, but was also trained on the commercial[0] where the bag appears to beat up a human (his fault for holding on too tight) and then it destroys an alien spacecraft.
[0] https://www.youtube.com/watch?v=sIAnQwiCpRc
An alert by one of these AI tools, which from what I understand have a terrible track record, should not be reasonable suspicion or probable cause to swarm a teenager with guns drawn. I wish more people in local communities would understand how much harm this type of surveillance and response causes. Our communities should not be using these tools.
There are two basic ways AI can be used:
1. To enhance human productivity; or
2. To replace humans.
Companies, particularly in the US, very much want to go with (2) and part of the reason they can is because there are zero consequences for incidents like this.
A couple ofexamples spring to mind:
1. the UK Royal Mail scandal where a bad system accused postmasters of theft, some of whom committed suicide over the allegations. Those allegations were later proven false and it was the system's fault. IMHO the people who signed off and deployed this should be charged with negligent homicide; and
2. The Hertz case where people who had returned cars were erroneously flagged as car thieves and report was made to police. This created hell for people who would often end up with warrants they had no idea about and would be detained on random traffic stops over a car that was never stolen.
Now these aren't AI but just like the Doritos case here, the principle is the same: companies are trying to replace people with computers. In all cases, a human should be responsible for reviewing any such complaint. In the Hertz case, a human should check to see if the car is actually stolen.
In the Royal Mail situation, the system needs to show its work. Deployment should be against the existing accounting system and discrepancies between the two need to be investigated for bugs until the system is proven correct. Particularly in the early stages, a forensic accountant (if necessary) should verify that funds were actually stolen before filing a criminal complaint.
And if "false positive" criminal complaints are filed, the people who allowed that to happen, if negligent (and we all know they are), should themslves be criminally charged.
We are way too tolerant of black box systems that can result in significant harm or even death to people. Show your work. And make a human put their name and reputation to any output of such systems.
The core of the issue is that many Americans do carry weapons which means that whatever the security system, it needs to keep in mind that the suspect might be armed and about to start shooting. This makes the police biased towards escalation because the only way against a shooter is to shoot first.
This problem doesn't exist in Europe or Japan because guns aren't that ubiquitous, which means that the police have the time to think before they act, which makes them less likely to escalate and start shooting. Obviously, for Americans, the only solution is to get rid of the gun culture, but this will never happen, so suck it up that AI gets you swatted.
You had me up until the last sentence...
>Obviously, for Americans, the only solution is to get rid of the gun culture, but this will never happen, so suck it up that AI gets you swatted.
...and you are correct.
I don't have kids yet, but I may someday. I went to public school myself, and would prefer to send any kid of mine to public school as well. (I'm not hard against private schools, but I'd prefer my kid gets to make friends from all walks of life, not just people who have parents who can afford private school.)
But I really wouldn't want to send my kid to a school that surveils students all the time, and uses garbage software like this that directly puts kids into dangerous situations. I feel like with a private school, I'd have more choice and ability to influence that sort of thing.
The regular types of school shootings weren't enough, so they invented AI-powered police school shootings to the mix.
Up next: gun kitted drones patrolling the school playground.
>> Omnilert later admitted the incident was a “false positive” but claimed the system “functioned as intended,” saying its purpose is to “prioritize safety and awareness through rapid human verification.”
No. If you're investigating someone and have existing reason to believe they are armed then this kind of false positive might be prioritizing safety. But in a general surveillance of a public place, IMHO you need to prioritize accuracy since false positives are very bad. This kid was one itchy trigger-pull away from death over nothing - that's not erring on the side of safety. You don't have to catch every criminal by putting everyone under a microscope, you should be catching the blatantly obvious ones at scale though.
The perceived threat of government forces assaulting and potentially killing me for reasons i have no control over, this is the kind of stuff that terminates the social contract. I'd want a new state that protects me from such stuff.
Looks like per their website it did function as intended... It surfaces potential threats for the school to look at and make a human decision. The principal decided to send the police after the school safety team dismissed it as part of the correct process. I mean fire alarms go off for lots of things that are not fire alarms... This was an alert meant to be validated by a human that messed up.
Its pretty clearly documented how it works here:
https://www.omnilert.com/solutions/gun-detection-system https://www.omnilert.com/solutions/ai-gun-detection https://www.omnilert.com/solutions/professional-monitoring
> Omnilert later admitted the incident was a “false positive” but claimed the system “functioned as intended,” saying its purpose is to “prioritize safety and awareness through rapid human verification.”
This exact scenario is discussed in [1]. The "human in the loop" failed, but we're supposed to blame the human, not the AI (or the way it was implemented). The humans serve as "moral crumple zones".
""" The emphasis on human oversight as a protective mechanism allows governments and vendors to have it both ways: they can promote an algorithm by proclaiming how its capabilities exceed those of humans, while simultaneously defending the algorithm and those responsible for it from scrutiny by pointing to the security (supposedly) provided by human oversight. """
[1]: https://pluralistic.net/2024/10/30/a-neck-in-a-noose/
The article doesn't confirm that there was definitely a human in the loop, but it sorta suggests that police got a chance to manually verify the photo before going out to harass this poor kid.
I suspect, though, that the AI flagging that image heavily influenced the cop doing the manual review. Without the AI, I'd expect that a cop manually watching a surveillance feed would have found nothing out of the ordinary, and this wouldn't have happened.
So I agree that it's weird to just blame the human in the loop here. Certainly they share blame, but the fact of an AI model flagging this sort of thing (and doing an objectively terrible job of it) in the first place should take most of the blame here.
Sincere, and snarky summary:
"Omnilert" .. "You Have 10 Seconds To Comply"
-now targeting Black children!
Q: What was the name of the Google AI Ethicist who was fired by Google for raising the concern that AI overwhelmingly negatively framed non-white humans as threats .. Timnit Gebru
https://en.wikipedia.org/wiki/Timnit_Gebru#Exit_from_Google
We, as technologists, ARE NOT DOING BETTER. We must do better, and we are not on the "DOING BETTER" trajectory.
We talk about these "incidents" with breathless, "Wwwwellll if we just train our AI better ..." and the tragedies keep rolling.
Q2: Which of you has had a half dozen Squad Cars with Armed Police roll up on you, and treat you like you were a School Shooter? Not me, and I may reasonably assume it's because I am white, however I do eat Doritos.
It sounds like the police mistook it as well:
> “They showed me the picture, said that looks like a gun, I said, ‘no, it’s chips.’”
So AI did the initial detection, but police looked at it and agreed. We don't see the image, but it probably did look like a gun because of a weird shadow or something.
Fundamentally, this isn't really any different from a person seeing someone with what looks like a gun and calling the cops, only it turns out the person didn't see it clearly.
The main issue is just that with increased numbers of images, there will be an increase in false positives. Can this be fixed by including multiple images, e.g. from motion of the object, so police (and the AI) can better eliminate false positives before traumatizing some poor teen?
Picture? Images? But those are just frames of footage the cameras have captured! Why would one purposefully use less information to make a decision rather than more?
Just put the full footage in front of an unbiased third party for a multi-stage verification first. The problem space isn't "is that weird shadow in the picture a gun or not?" it's "does the kid in the video have a gun?". It's not hard to figure out the difference between a bag of chips and a gun based on body language. Presumably the kid ate chips out of the bag? Using certain motions that one makes when doing that? Presumably the kids around him all saw the object in his hands and somehow did not react as if it was a gun? Jeez.
Those are a lot of "presumablies". Maybe you're right. Or maybe it was mostly obscured so you really couldn't tell. How do you know it was open and he was eating? How do you know there were other kids around and he wasn't solo? Why do you think the body language would be so different? Nobody is claiming he was using a gun or threatening anyone with it. If you're just carrying something in your hand, I don't know how you could tell what the object is or isn't from body language.
2 replies →
> So AI did the initial detection, but police looked at it and agreed. We don't see the image, but it probably did look like a gun because of a weird shadow or something.
Not sure I agree. The AI flagging it certainly biased the person doing the manual review toward agreeing with the AI's assessment. I can imagine a scenario where there was no AI involved, just a human watching that same surveillance feed, and (correctly) not seeing anything alarming in it.
Also I expect the AI completely failed at context. I wouldn't be surprised if the full video feed, a few minutes (or even seconds) before the flagged frame, shows the kid crumpling up the empty Doritos bag and stuffing it in his pocket. The AI probably doesn't keep all that context around to use when making a later decision, and giving just the flagged frame of video to the human may have caused them to miss out on important context.
Computer says it looks like a gun.
https://en.wikipedia.org/wiki/Computer_says_no
Systems like this need to report confidence in their assertions.
e.g. Not "this student has a gun" but "this model says the student has a gun with a probability of 60%".
If an AI can't quantify it's degree of confidence, it shouldn't be used for this sort of thing.
Even better, share the frame(s) that the guess was drawn from with a human for verification before triggering ANYTHING. How much trouble could that possibly be? How many "guns" is this thing detecting in a day across all sites? I doubt more than a couple or we'd have heard about tons of incidents, false positives or not.
I wanna see the frames too.
I don't find that especially good as a sole remedy, because lots of people are stupid. If they see a green outline box overlaid on a video image with the label 'gun', many many people will just respond to the label instead of looking at the underlying image and trying to make a decision. Probability and validation history need to be built into the product so that there are audit logs that can be pored over and challenged. Bad human decision-making, which is rampant, is always smoothed over with justifications like 'I was concerned for everyone's safety', and usually treated in isolation rather than assessed longitudinally.
Doesn’t work. Competition will make them report higher accuracy to make the product look better.
It’s unsurprising, since this kind of classification is only as good as the training data.
And police do this kind of stuff all the time (or in the very least you hear about it a lot if you grew up in a major city).
So if you’re gonna automate broken systems, you’re going to see a lot more of the same.
I’m not sure what the answer is but I definitely feel that “security” system like this that are purchased and rolled out need to be highly regulated and be coupled with extreme accountability and consequences for false positives.
Everything around us: political tumult and weaponization of the justice system, ICE and other capricious projections of federal authority, the failure of drug prohibition, and on and on and on, points to a very simple solution:
Abolish SWAT teams. Do away with the idea that the state employees can be permitted to be more armed than anyone else.
Blaming the so-called 'swatter' (whether it's a human or AI) is really not getting at the root of the problem.
How likely is it that the AI system would have classified the bag of Doritos as a weapon had the person carrying it been white instead of black?
I would be certainly curious to test ethnicity with this system. Will white students with a bag of Doritos be flagged, or only if they’re black?
Exactly. I wonder if this a purpose-built image-recognition system, or is it a lowest-possible effort generic image model trained on the internet? Classifying a Black high school student holding Doritos as an imminent shooting threat certainly suggests the latter.
Wait… AI hallucinated and the police overreacted to a black kid who actually posed no threat?
I thought those two things were impossible?
> “They showed me the picture, said that looks like a gun, I said, ‘no, it’s chips.’”
"Sorry, that's Nacho gun"
Armed and dangerous until proven chips.
"I am invoking my 4th and 5th amendment rights afforded to me by the Constitution of the United States of America. I have no further comment until I have consulted with and am in the presence of my legal council."
Then, just sit back and enjoy as the lawsuit unfolds.
When people wonder how can AI mistake a bag of snacks as a weapon, simply answer "42"
It is about the question, the answer will become very clear once you understand what was the question presented to the inference model, and of course what data and context was fed
Inflicting trauma on a harmless human in the name of the "safety of others" is never ok. The victim here was not unharmed, but is likely to end up with PTSD and all the mental health issues that come with it.
I hope they sue the police department over this.
Sad for the student.
Imagine the head scratching that's going on with execs who are surprised things might work when a probabilistic software is being used for deterministic purposes without realizing there's a gap between it kind of by nature.
> Imagine the head scratching that's going on with execs
I can't. The execs won't care and probably in their sadist ways, cheer.
Fair. Only a matter of time until it's big enough that it can't be avoided.
I'm sure there will be no head scratching. They already know that this can happen, and don't care, because they know that if someone gets killed because of it, they won't be held responsible. And may not even lose any customers.
the best part of the technocracy is that they're not actually all that good at anything. the second best part is that when their mistakes end in someone dead there will be some way that they're not responsible.
Wouldn’t have thought AI assessment of security image is enough for probable cause
Very ripe for a lawsuit. I would expect lawyers to be calling daily.
At least there is a check done by humans in a human way. What if this human check is removed in future, as AI decisions would be deemed no longer requiring a human inspection?
Isn't that what happened here?
If these AI video based gun detectors are not a massive fraud I will eat one.
How on Earth does a person walk with a concealed gun? What does a woman in a skirt with one taped to her thigh walk like? What does a man in a bulky sweatshirt with a pistol on his back walk like? What does a teenager in wide legged cargo jeans with two pistols and a extra magazines walk like?
The brochure linked from TFA has a screenshot of a combination of segmentation and object recognition models which are fairly standard in NVRs. Quick skim of the vendor website seems to confirm this[1] and states a claim that they are not analyzing the gait.
[1]: https://www.omnilert.com/blog/what-is-visual-gun-detection-t...
The whole idea even accepting the core premise is OK to begin with needs to have a similar analysis applied to it that medical tests do: will there be enough false positives, with enough harm caused by them, that this is actually worse than doing nothing? Compared with likelihood of improving an outcome and how bad a failure to intervene is on average, of course.
Given that there's no relevant screening step here and it's just being applied to everyone who happens to be at a place it's truly incredible that such an analysis would shake out in favor of this tech. The false positive rate would have to be vanishingly tiny, and it's simply not plausible that's true. And that would have to be coupled with a pretty low false negative rate, or you'd need an even lower false positive rate to make up for how little good it's doing even when it's not false-positiving.
So I'm sure that analysis was either deliberately never performed, or was and then was ignored and not publicized. So, yes, it's a fraud.
(There's also the fact that as soon as these are known to be present, they'll have little or no effect on the very worst events involving firearms at schools—shooters would just avoid any scheme that involved loitering around with a firearm where the cameras can see them, and count on starting things very soon after arriving—like, once you factor in second order effects, too, there's just no hope for these standing up to real scrutiny)
The real issue is that they obviously can't detect what's in a backpack or similar large vessel.
Gait analysis is really good these days, but normal, small objects in a bag don't impact your gait.
The only way we could have foreseen this was immediately.
The model seems pretty shitty. Does it only look on a frame-by-frame basis? Literally one second of video context and it would never make that mistake.
If I was that kid I'd be suing the school, the AI company, the police, anyone and everyone who had to be subjected to the mistake.
I can understand the outrage in this thread but literally none of what you are all calling for will be done. No one from justice or law reads HN to see what should be done. I wish folks here would keep a cooler head rather than posting lengthy rants and vents that call for punishing school staff. Really unprofessional and immature from a community that prides itself, to fall constantly into a cycle of vitriol.
Can someone outline a more pragmatic, if not likely, course of what happens next after this? Is it swept under the rug and we move on?
This is only the beginning of AI-hallucinated policing. Not a good start, and I don't think it's going to end well for citizens.
"end well for citizens."
That ship has long sailed buddy.
Yeah ask all those citizens getting “detained” by ICE how it worked out for them.
This is Brazil-level insanity, but half the people on this forum hope to make money off it so it's fine
* AI mistakenly reports the student as a threat
* the student was black
Is that really a coincidence?
It's just a matter of time before this or something worse happens.
The solution is easy. Gun control. We dont feel the need to have AI surveillance on people to detect guns in ROTW.
Was this really an AI gun detection system, or just a machine that goes off randomly?
In 1987, Paul Verhoeven predicted exactly this in the original Robocop.
ED-209 mistakenly viewed a young man as armed, blows him away in the corporate boardroom.
The article even included an homage to:
“Dick, I’m very disappointed in you.”
“It’s just a small glitch.”
Robocop.
Edit: And racism. Just watched the video.
The "AI mistake" part is a red herring.
The real question is: Would this have happened in an upper/middle class school.
The student has dark skin. And is attending a school in a crime ridden neighborhood.
Were it a white student in a low crime neighborhood, would they have approached him with guns drawn?
The AI failure is masking the real problem - bad police behavior.
Durely a human being should review the evidence before going off half cocked.
America does American things.
And so begins the ending of the "unfinished fable of the sparrows"
I thought on first glance the source was from doritos.com
That would have been bold
It wasn't sour cream and onion and didn't contain cash, so it's super sus.
But really this is typical of cop overreaction with escalation and ego rather than calm, legal, and reasonable investigation. Karens may SWAT people they don't like, but it's police officers who must use reasonableness and restraint to defend the vestiges of their impartiality and community confidence based on asking questions and gathering evidence in a legal and appropriate manner rather than rushing to conclusions. Case in point: The NYC rough false arrest of a father in front of his kid to retrieve his mis-delivered package where the egomaniacal bully cop aggressively lectures a guy for his own mistake to cover his own ego while blaming the victim: https://youtu.be/LXd-4HueHYE
I would get my GED at that point. Screw that school.
With high level of hallucination, cops need to tranquilizers more. If the student had reached for his bag just before the cops arrived, BLM 2.0 would have started.
T0 be fair most commercials for Doritos, Skittles, Mentos, etc., if occurring in real life, would result in a strong police response just after they cut away.
AI is a false (political) wish, it can and never work, it is the desperation of an over extended power structure to hold on and permanently consolodate controll of all of the worlds population, and nothing else.
the proofs are there.
philosophers mulled this over long ago and made clear statements as to why ai cant work
though not that for a second do I misdunderstand that it is "all in" for ai, and we all get to go for the 100 trillion dollar ride to hell.
can we have truely awsome automation for manufacturing and mundane beurocratic tasks?, fuck ya we can!
but anything that requires understanding, is forever out of reach, which unfortunatly is also lacking in the people pushing, this thing, now
we need personal liability for the owners of companies that make things like this
Can someone write the novel
“Computer says die”
Who knew eating Doritos could make you a millionaire?
I hope this kid gets what he deserves.
What a tragedy. I'm sure racial profiling on behalf of the AI and the police had absolutely nothing to do with it.
How is this not slander? I would absolutely sue the fuck out of this system where it puts people's lives in danger.
> How is this not slander?
Because that's not what slander is.
Hallucinate much?
Before I clicked the article, I said to myself "The victim's gotta be Black", and lo and behold. AI has inherited police's (shitty, racist, and dangerous) idea that any Black person is a dangerous monster for whom anything is a weapon.
This is what we get instead of reasonable gun control laws.
You're free to (attempt to) amend the Second Amendment, but the Supreme Court of the United States has already affirmed and reaffirmed that individual ownership of firearms in common use is a right.
What do you propose that is "reasonable" given the frameworks established by Heller, McDonald, Caetano, and Bruen?
I can 3D print or mill basically every item you can imagine prohibiting at home: what exactly do laws do in this case?
> the Supreme Court of the United States has already affirmed and reaffirmed that individual ownership of firearms in common use is a right.
The current interpretation of 2A is actually a fairly recent invention; in the past, it's been interpreted much more narrowly. And if SCOTUS can overturn Roe v. Wade's precedent, they can do the same with their interpretation of 2A. They won't of course, at least not until some of its members age out and get -- hopefully -- replaced with people who aren't idiots.
But I'd be fine if 2A was amended away. Let the states make whatever gun laws they want, and we can see whether blue or red states end up with lower levels of gun violence as a result.
1 reply →
"Omnilert Gun Detect delivers instant gun detection, near-zero false positives".
If it's taking images every 30 seconds, it's getting 86400 x 30 = 2.5 million images per day per camera. So when it causes enormous, unnecessary trauma to one student per week, the company can rightfully claim it has less than a 1 in 10 million false positive rate.
(* see also "how to lie with statistics").
I was unduly surprised and disappointed when I saw the photo of the kid and he turned out to be black. I would love to believe that this had no impact on how the whole thing played out, but I don't.
All right they’ve gotta have a plain clothes bro go up there make sure the kid is chill. You know the difference between a murder and not can be as little as somebody being nice
Sounds like this high school is doing a great job preparing students for the real world, where they can be swarmed by jackbooted thugs at any moment for any reason.
its not gun detection that ai is racist just like its white creators
> “false positive” but claimed the system “functioned as intended,”
Fuck you.
[dead]
>Omnilert later admitted the incident was a “false positive” but claimed the system “functioned as intended,” saying its purpose is to “prioritize safety and awareness through rapid human verification.”
It's ok everyone, you're safer now that police are pointing a gun at you, because of a bag of chips ... just to be safe.
/s
Absolutely ridiculous. We're living "computer said you did it, prove otherwise, at gunpoint".
We got our cyberpunk future, except none of it's cool and everything's extremely stupid.
They could at least thrown in some good music and cute girls with colored hair to make us feel better :(
6 replies →
AI singularity wil happen, but motherbrain as a complete moron. It will extinguish humans not as a grand plan for machines to take over, but doing horrible mistakes when trying to make things better.
I recommend rewatching the trilogy of Brazil, 12 Monkeys and Zero Theorem.
It's sadly the exact future that we are already starting to live in.
1 reply →
Pretty sure cyberpunk was always this dark.
1 reply →
If any of you had actually paid attention to the source media, you would have noticed that they were explicitly dystopias. They were always clearly and explicitly hell for normal people trying to live life.
Meanwhile, tons of you watched star trek and apparently learned(?) that the "bright future" it promised us was.... talking computers? And not, you know, post scarcity and enlightenment that allowed people to focus on things that brought them joy or they were good at, and an entire elimination of the concept of "capitalism" or personal profit or resource disparity that could allow people to not be able to afford something while some asshole in the right place at the right time gets to take a percentage cut of the entire economy for their personal use.
The primary "technology" of star trek was socialism lol.
1 reply →
gustapo
The safest thing to do is to pull all Frito Lay products off shelves until the packaging can be redesigned to ensure that AI never confuses them for guns. It's a liability issue. THINK OF THE CHILDREN.
The only thing that can stop a bag guy with Doritos is ...
[dead]
Feed the same system an image of an Asian kid and it will think the bag of chips is a calculator /s
Or am I kidding? AI is only as good as its training and humans are...not bastions of integrity...
I think it's almost guaranteed that this model has race-related biases, so no, I don't think you're kidding at all. I think it's entirely likely that an Asian (or white) kid of the same build, wearing the same clothes, with a crumpled-up bag of Doritos in his pocket, would not get flagged as having a gun.
Using humans for training gurantees bad outcomes because humans cannot demonstrate sociality at the same scale as antisociality.
Let's hope that, thanks to AI, the young man will now have a healthier diet! /s
Poor kid, and what an incompetent police department not to use their own judgement ……
But ……
Doritos should definitely use this as an advertisement, Doritos - The only weapon of mass deliciousness, or something like that
And of course pay the kid, so something positive came come out of the experience for him
The snack you'll only SWAT from my cold dead hands
[flagged]
[flagged]
3 replies →