Amid a growing epidemic of gun violence, can AI be part of the solution? In this episode we look at some of the weapons detection technologies schools are using in an effort to try to keep students safe. 

We Meet:

Gary Hough, superintendent of Fayette County schools Mark Keierleber, investigative reporter at The 74Mike Ellenbogen, Founder, chief innovation officer at Evolv TechnologiesDonald Maye, head of operations at IPVM


Spielberg, S. (2002). Minority Report. Twentieth Century Fox.Avigilon Athena Security integration for Gun Detection, via YouTube


This episode was produced by Anthony Green and Emma Cillekens with reporting from Mark Keierleber. It was edited by Jennifer Strong, Rachel Courtland and Mat Honan, mixed by Garret Lang, with original music from Jacob Gorski.

Full Transcript:



Gary Hough: You think about what that feeling must have been like in, in the hearts and minds of, of, of those parents and grandparents and uncles and aunts and relatives and the other children in that school. 

Jennifer: He’s talking about school shootings… and the growing epidemic of gun violence in the US. 

Gary Hough: I think we have to reach out in a national movement to do our best to protect our young people. My name is Gary Hough. I am the superintendent of Fayette County schools in Fayette County, West Virginia. 

Jennifer: His district has been installing AI-enabled systems that are meant to detect guns and other weapons at school entrances.

And so far, he’s finding them useful.  

Gary Hough: We don’t have to have that long line that you deal with with the old metal detector scanners. They are able to roll straight through the process.

Jennifer: Students don’t have to slow down to pass through… stopping only if an alarm goes off. He says this is important… because long lines present a safety hazard.

Gary Hough: You know, when they’re getting off those school buses and coming in, or getting out of the cars and coming in, that period of time that they’re delayed coming through that front door where, you know, you’re enclosed in a controlled area. They’re in a vulnerable setting. This permits us to do the scanning without putting our students in that situation.

Jennifer: These machines sometimes catch things that shouldn’t be there… including brass knuckles.

But he also says, false positives are a daily occurrence… and school issued laptops are often mistaken for weapons… So students now take their Chromebooks out of their bags and hold them up as they pass through the entrance.  

Gary Hough: You know, those are the kind of things that we had to train our people to do so that it doesn’t stop them, but it’s not like what you have when you have the belts and the scanners, you have to take all the change out of your pocket. Change and keys, do not set it off. You know what I mean? It already is programmed into there that that’s not a security risk.

Jennifer: He’s talking about a system from a company called Evolv… and we’ll meet one of the founders in just a bit.

The product looks a lot like those plastic towers inside stores that detect shoplifting.

It’s there in plain sight, but you don’t really notice it unless it goes off. 

I know, because my microphone set one off recently in New York City.

But, unlike a standard metal detector… this one uses AI to keep it from going off over things like phones and keys.

Gary Hough: I think our parents were very pleased and continue to be pleased because of the level of technology that we have. Is this a kind of a fix-all for the problem? No, it’s not, but it’s a good tool in your tool chest as a school administrator to know you’ve done everything you can. 

Jennifer: I’m Jennifer Strong and this episode we explore some of the AI technologies that schools are looking to, to try to help keep students safe. 


Jennifer: The conversation on school safety has long been about preventing mass shootings. But the pandemic caused a shift in that conversation… towards stopping the spread of the virus… with products like temperature sensors and cameras that detect masks. 

For many school districts, this shift also marked the beginning of security protocols driven by AI.

Mark Keierleber:  So now some of these folks who implemented these cameras… you know I talked to the security chief in Fulton county, Georgia, which is Metro Atlanta and they installed a system that detects masks and he specifically talked about how, you know, well, in a post pandemic environment, we may not be checking for masks, but this is a precursor to other forms of technology, specifically weapons detection.

My name’s Mark Keierleber.  I’m an investigative reporter at The 74, a national K-12 education news website. 

Jennifer: He covers security and surveillance technology in schools. 

Related work from others:  Latest from MIT : 3 Questions: What you need to know about audio deepfakes

Mark Keierleber: That can range from metal detectors to locks inside classrooms to bulletproof blankets. 

Jennifer: You heard that right… he said bulletproof blankets… which are among a whole host of bulletproof school supplies.

Mark Keierleber: There are bulletproof backpacks. There are bulletproof whiteboards. There are bulletproof shields that are designed to hang in a classroom next to the fire extinguisher. There’s basically anything that you can think about inside of a school… has been redesigned in the last few years to be bulletproof. 

Jennifer: But he says the real focus in school security is on using AI and other tech to find weapons.

Schools that use the kinds of technologies he reports on to try to prevent gun violence… could benefit from a bill introduced by Congress in June. It would pump hundreds of millions of dollars into a program started in response to previous school shootings.

Mark Keierleber: So the idea is they’ve programmed the cameras to identify what, you know, a gun looks like in terms of its shape. There’s one company that was marketing a gun detection system a few years ago where it was also equipped with speakers in the hallway. And it would say, like it would detect the gun and the speakers would, like, a robot would come across the speakers and say “warning, we have detected a gun please be advised that law enforcement is on the way.”

Jennifer: That product comes from a company called Athena Security… which makes a variety of weapons detection products.

Instructor: Okay. What you see in the bottom left hand corner is the live video feed. 

Jennifer: This is taken from a demo the company posted on YouTube.

Instructor: Upper left hand corner is our alerting platform and in the right you’re gonna see the person entering with a gun. Of course, this can also work with our fever detection platform. And if a person has the fever walking in front of the camera, it’ll send the same exact alert. 

Jennifer: Fever detection technology took off during covid, and though it’s far from foolproof… it’s probably here to stay.

Instructor: So what you see here is a person entering with a weapon. Soon as the camera sees the weapon on the lower left hand corner.. wait for it.. There it is. Boom. It shows the alert on the right hand side, it pops into Avigilon. You’ll see the image and then you can pull open the video if you like. 

Jennifer: After the Parkland shooting, Reporter Mark Keierleber says he was getting all kinds of pitches from tech companies about everything from metal detectors to those bulletproof backpacks… and he says someone from Athena reached out with a quote that really stayed with him.

Mark Keierleber: Well, I’m gonna read you the quote. // Artificial intelligence to help law enforcement stop crime before it starts or escalates, like Minority Report but in real life, is becoming a reality. // And that’s pretty interesting. Right. Um, so if you’ve ever seen the movie Minority Report, the idea here is that, in Minority Report, you’re, you’re being constantly followed by different forms of surveillance. It’s able to recognize, um, everything around you and basically predict crime before it happens. 

Announcer: [00:21] Within just one month under the pre-crime program, the murder rate in the District of Columbia was reduced 90 percent. 

Victim 1: They were gonna be waiting for me in the car.  

Victim 2: He was gonna rape me.

Victim 3: I was going to be stabbed.

Victim 4: Right here. 

Announcer: Within a year, pre-crime effectively stopped murder in

our nation’s capital. 

Jennifer: Up until recently, security cameras have been passive. Meaning, after a crime, people watch the tape to see what happened.

But we’re moving to a world where cameras are active… with AI that makes real time assessments and predictions.

Mark Keierleber: Certainly there are all kinds of other forms of weapons detection on the market. I mean, Shot Spotter is a big one and it exists in communities across the United States, especially major cities. 

Jennifer: Basically, that product is a microphone on a light pole that’s meant to detect the sound of a gunshot. 

Then, there’s products like Evolv.

Mark Keierleber: It’s these new forms of high tech, what appear to be basically metal detectors… and metal detectors play a role in the school security conversation. They certainly are discussed a lot at length in the wake of school shootings. We saw major growth in different forms of security and surveillance technology installed in schools, in the wake of the Columbine shooting and in the wake of Parkland and in the wake of Sandy Hook tragedy. And certainly nothing is going to be different with Uvalde.  

[Chapter Change Music] 

Mike Ellenbogen: Metal detectors are 90 years old, nine zero, right? They were first deployed in Germany in the 1920s, 1930s to look for castings being stolen out of factories. And they haven’t fundamentally changed. And they were great for, you know, if, if you want, in a prison environment, if you’ve gotta make sure that nobody’s coming in with a handcuff key or half a razor blade or something like that 

Related work from others:  Latest from MIT Tech Review - AI just beat a human test for creativity. What does that even mean?

Mike Ellenbogen: Sure. Uh, Mike Ellenbogen. Founder, and chief innovation officer for Evolv Technologies.  

In an environment where you don’t expect them to have any other metal objects on them, metal detectors are great. They’re really good at finding metal. The problem is today, we’re all carrying metal, right? Everybody’s got a cell phone, everybody’s got pockets full of stuff. 

Jennifer: His goal is to automatically detect threats being carried by people.

Mike Ellenbogen: I’ve grown up if you will, in doing physical security technology. So Evolv’s actually my third company in this space. My last company was called Reveal Imaging Technologies. We made the systems that TSA uses in about 250 airports around the US to automatically look for explosives in suitcases before they get loaded into the belly of the aircraft. 

Jennifer: And after that company was acquired… he was looking for his next project.

Mike Ellenbogen: And that’s when the Sandy Hook shooting happened. So it was obvious at that point in time that what the world really needed was an answer to this active shooter problem. 

How do we keep threats like guns, bombs, you know, large threats to the crowd, tactical knives, things like that, out of venues that just want to create a safe environment. How do we enable security professionals to create a safe environment, but recognize that they still need to move a lot of people in and out quickly? So we set up Evolv to do exactly that. That’s why we exist.

Jennifer: He says the system uses a combination of sensors and machine learning to discern between everyday items and those that pose a threat.

Mike Ellenbogen: So rather than that old school analog metal detector, the Evolv system’s a digital platform at your front door. It’s got the core weapons detection capability, but there’s also cameras that we can use for different applications. The other sensors that we can add and other capabilities that we can provide, including the data that come from all that, that helps the venue optimize their operations.

Jennifer: And he says they’re working on ways to integrate it with other systems. 

Mike Ellenbogen:  We think there’s a great opportunity to combine credentialing. So ticketing is a form of credentialing, right? An ID, you know, that it takes to get into your building is credentialing. So bringing those two things together using, whether it’s face based recognition, maybe phone based. There’s a number of different ways in which we can do that. We spend a lot of time trying to understand the lay of the land and figure out which of the best, either, technologies that are either existing or that we might want to develop to address some of these needs.


Jennifer: You can find links to our reporting in the show notes… and you can support our journalism by going to tech review dot com, slash subscribe.

We’ll be back… right after this.


Donald Maye: Certainly with a product like Evolv, you know, they’ve raised hundreds of millions of dollars with aspirations to be part of everyday life in the United States and the world. And, you know, that is lofty expectations and we felt and feel that the public should really understand what this technology can and cannot do. 

I’m Donald Maye. I’m the head of operations of IPVM and IPVM is an organization that specializes in reporting on the video surveillance and security industry. 

We’re a team of researchers and reporters that really try to understand the underlying technology that’s being sold in the marketplace. And really what we try to do is be an objective resource that isn’t influenced by the companies or the sellers of this technology.

Jennifer: His team spent eight months trying to study how Evolv’s systems work.

Donald Maye: What stands out is they bill themselves as a weapons detector and, and they go so far as to say on their website what they’re not, and what they’re not is a metal detector. And like, for us, that’s wildly deceptive because when you look at the underlying technology, it is in fact a metal detector. One that’s able to better distinguish between certain metal objects and make a determination of whether something is benign or an actual object of threat.

Jennifer: But his tests found that strollers, umbrellas and eyeglass cases were often mislabeled as weapons. 

Donald Maye: What each of these items have as a challenge for the Evolv system to differentiate between those in a weapon is that they have properties that are similar to weapons and so the Evolv technology struggles to differentiate between those two. If you have a high volume area, you know, where thousands of people are walking through in an hour, you’re gonna have many, many, many secondary alerts, particularly in environments where it’s not expected for people to not be carrying metal objects and an environment like that would be the subway. You know, people, especially in New York City, take all kinds of metal objects onto the subway, which is much different than, say, going to a sports venue or going to a museum. Subways have people going about their day to day lives and so you can think very clearly now if you, if you understand this underlying limitation of the system, that it’s gonna be a challenge logistically to deploy it in all the places that Evolv is seeking to deploy it.

Related work from others:  Latest from MIT : An AI challenge only humans can solve

Jennifer: It also presents a problem for schools where Evolv systems are already deployed. Because school-issued laptops are another common culprit of those false positives.

Donald Maye: There’s metal hinges in certain Chromebooks that have properties similar to a gun. There’s some cylinder shaped componentry to it and it’s curved and looks like the outline of a gun. 

Jennifer: And while he says Evolv’s systems have something of an internal workaround for that detection issue… it comes with key tradeoffs. 

Donald Maye: One of the more enlightening pieces of information that we learned was from a school board meeting where an Evolv employee was describing the trade off of settings as it relates to Chromebooks and guns. And what he stated was, he goes, if you go on setting C, which is a lower sensitivity setting, he goes, you won’t alert on Chromebooks. However, you might miss certain guns. Then he says it’s a difficult give and take. And he specifically noted subcompact handguns. And so for me, I heard it. I go, well, that’s, that is a difficult give and take. If you’re a school you’re having to decide whether or not I wanna do the work around where, where I’m asking people and students to hold up their Chromebook or go to a lower sensitivity setting and run the risk of someone, you know, not alerting on a gun.

Jennifer: And Donald Maye says it’s not clear what variables the algorithm weighs when that sensitivity is adjusted.

And this decision between efficiency and accuracy… one with potentially devastating consequences… is left up to the people using the device, which in this case, might fall to a mix of volunteers from the teaching staff. 

Donald Maye: Being presented information and not having the ability to reliably dissect it and understand it. And I think that disconnect in information creates an imbalance of power between the buyers and end users and the people who are selling it. And that can lead to a lot of problems. 

Jennifer: And what he’s describing… extends well beyond this topic… because the world is becoming ever more technical, and with it, that information imbalance just keeps on getting wider.     

It’s a bit like looking under the hood of a car. In the past drivers could understand the basic mechanics of how it all worked. Not so much anymore. And back to that laptop problem… making the system less sensitive could miss some weapons… but on the other hand, alarms going off all day are likely to be ignored. 

Mike Ellenbogen: Yeah. It’s a, it’s a real challenge within the security domain, right? Whether it’s physical or cyber, you know, we call it alert fatigue. 

Jennifer: Mike Ellenbogen… is the cofounder of Evolv

Mike Ellenbogen: If you remember, when car alarms were first introduced, right? If a car alarm went off, you know, in the street, outside your apartment, you know, you’d run to the window to see what’s going on. And then they just became annoying and you would ignore them because they were happening all the time. And then people kind of just turned them off. And that’s what happens with a lot of these technologies that aren’t quite ready for primetime. That those false alarms create. That they undermine people’s faith in the technology. And then it just becomes a nuisance.


Jennifer: This episode was produced by Anthony Green and Emma Cillekens with reporting from Mark Keierleber. 

It was edited by me, Rachel Courtland and Mat Honan… and mixed by Garret Lang… with original music from Jacob Gorski.  

If you have an idea for a story or something you’d like to hear, please drop a note to podcasts at technology review dot com.

Thanks for listening… I’m Jennifer Strong.


Similar Posts