Jordan: By now, you’re probably used to your data being used by all sorts of platforms to whom you gave the right to access it. It takes a lot to phase us now. Many of us no longer blink at accepting cookies. Or using our Facebook profiles to log into other applications. We click on targeted ads, which of course helps the next ad target us more accurately. We raise an eyebrow maybe when an app seems to know something about us that it shouldn’t, but even then, we usually just move on with our day. It takes a lot to make the world sit up and notice just how much of our information is out there.
News Clip: I am appalled that this check has been used. That reaction from Ontario’s former privacy commissioner after Toronto police admitted to using facial recognition technology from Clearview AI.
Jordan: When the New York times broke a story about a company creating advanced facial recognition technology and broke the story of law enforcement officials using it, that seemed to tip the scales. Since the Times broke, those stories, there have been more questions than answers and even the answers we have aren’t very clear. So what is Clearview AI’s facial recognition technology? How does it work? Who is using it and what, if anything exists to make sure that our personal information, in this case our faces, isn’t being exploited. I’m Jordan Heath Rawlings, this is The Big Story. Kate Allen is the science and technology reporter at the Toronto Star. Hi, Kate.
Kate: Hi!
Jordan: Why don’t we start, before we talk about the Toronto police or anything that’s happened here recently, just tell us what is Clearview AI?
Kate: Um, so it’s a tool made by an American technology company. Um, it’s a facial recognition app, so it allows you to take a photo of anyone or upload a photo and, uh, search that photo against its database of, it says, 3 billion images, um, scraped from the web.
Jordan: What is that supposed to do for people who use it? Like what’s the utility?
Kate: Yeah, so it’s looking for matches. So it essentially uses something called a neural network, which is a form of AI to, um, without getting into like the nitty gritty computer science, like it essentially turns the face that it’s searching into, um, a bunch of, let’s just say numbers and then it searches its enormous database for matches. Um, and so it can pull up photos of, hopefully, the same person from all across the web.
Jordan: And is this different from the kind of facial recognition software that we’ve talked about in the past on this podcast or other places have written about?
Kate: Yeah, it is. So last May, a colleague and I at the Star, uh, wrote about how the Toronto police was using a facial recognition software that it had bought with a um, grant from the province. And that facial recognition tool only searches the Toronto police’s internal database of mugshots. So that’s about 1.5 million, I think, images. And those are, they stressed to us at the time, these are lawfully obtained images, whereas this app, which is, you know, new to us, we’ve just found out about it, uh, searches, billions of images scraped off the web.
Jordan: When you say scraped off the web, where does it get them?
Kate: Well, so that’s a really great question. Uh, initial reporting around Clearview AI said that these, some of the places that these images are scraped off of include Facebook and other social media sites. Now, Clearview AI has since said that they don’t scrape images off social media if your account is private. But as anybody who’s tried to lock down their Facebook account knows, it’s not always clear what’s public and what’s private. And you know, sometimes you think something that you’ve posted is private, but it’s public or you’ve changed your settings, or they’ve made a new setting that you’re not aware of. If your photos are on the open web, Clearview AI can scrape it and add it to its database and it’s there forever.
Jordan: So, and they could come from any number of places then.
Kate: Yes. Anywhere.
Jordan: Have the major social media tech companies said anything about the fact that it’s their photos or photos on their platform that are being scraped to do this?
Kate: Yeah, absolutely. So Google, which owns YouTube and Facebook, which owns Instagram and Twitter and LinkedIn, have all demanded that Clearview AI not use it, data from their websites. Um, some of them have sent cease and desist letters. Some of them have just demanded. Um, now Facebook has already been in trouble in the past, including in Canada for sort of being extremely lax about how third parties use information from, from its website. We had the Cambridge Analytica scandal already, and, uh, the Canadian privacy commissioner and the BC privacy commissioner have slammed Facebook for being very lax about how, um, how third parties, you know, scrape data from, from Facebook.
Jordan: Do we know? Um, I guess the scale of it is, according to them 3 billion photos, but do we know who’s using this app? Uh, how widespread it is?
Kate: Yeah. So Clearview AI has said that they’ve given the app to 600 law enforcement agencies in North America, and that includes Canadian law enforcement agencies. So we are still trying to figure out which Canadian law enforcement agencies use it and which don’t, and which one tell us. Um, it’s also possible that, I mean, Clearview way, I can tell this to anyone it wants or give it away for free if it wants to. So, um, you know, it could very easily be sold to a private company who wants to use it for, you know, whatever.
Jordan: How did we find out about this technology? Because like you said, nobody’s really advertising it.
Kate: Yeah. So there was a big story in the New York Times back in January that sort of brought this company to public attention because it’s a very small company for its, you know, enormous powers. It only seems to have like a handful of employees. And so when that story came out, um, a lot of media here, obviously here in Canada, we’re interested in whether Canadian law enforcement agencies were using this tool. Um, so we asked, and so I asked the RCMP who have said that they don’t reveal investigative techniques and that they monitor, um, new technologies, but wouldn’t confirm or deny whether they use it. I asked the OPP who said the same thing, the Ontario provincial police, and I asked the Toronto Police Service. Um, so when I first asked the Toronto Police Service they didn’t answer me, um, and said they needed another day. And then I followed up and they said they needed another day. And then another day, and I was too busy to really like give it a second thought. And then finally they said yes. Um, our officers had, um, used this, um, starting last October in an informal testing, um, scenario. And the chief of police had not been aware of this. And when he found out about it in February, he put a stop to it and asked the privacy commissioner to look into the this, this tool. But it’s really interesting to compare, as I said, how that differs from the Toronto police’s other facial recognition tool. Because when they acquired that one, as I said, they got a grant from the government, the provincial government to buy it. They, you know, it went through all the sort of internal protocols for acquiring a new technology.
Jordan: It was available, anybody could see that they were using this thing.
Kate: Well, so actually not quite. Not quite. So we didn’t, so they had actually been testing it for a really long time before we were made aware of this. But they had done a, um, they had worked with the privacy commissioner of Ontario to do a privacy impact assessment, which is essentially what, the office in this province that sort of governs the use of personal information and access to information, what they want law enforcement agencies and any other public institutions to do if they’re going to use a tool that has significant impacts on privacy. They want you to come to their office and work with the office to minimize the impacts on privacy. So they had, um, done some of this when they acquired this other facial recognition tool. This is the one that only works on internal mugs, the internal mugshot database. But with this one, it’s not really clear what happened. Like the police chief apparently wasn’t aware that these officers were using it. Only then did the chief bring this to the attention of the privacy commissioner. So there was no impact assessment done on, you know, how this, how this will affect the privacy of Ontarians. So we’re still trying to find out exactly what happened here. There’s a lot we don’t know
Jordan: You may have kind of already answered this question, but are there any, beyond the privacy commissioner or even sitting with any of the governments, any regulations, uh, regarding what police can and can’t use to gather information?
Kate: Okay, so first of all, this balance where they have to go, go to the privacy commissioner and sort of talk about how to minimize the impacts on privacy. That is not by any means required by the law. That’s just something the privacy commissioner has asked police forces to do. And I should say, so Toronto police, after discovering that their officers had used this, went to the privacy commissioner. We, since, we at the Toronto Star since then, have since found out that Halton regional police and Peel regional police also have used this app. And when we asked the privacy commissioner for comments on this, he said, please, essentially, I’m paraphrasing, please stop using this and call my office immediately. So yeah, so, so you know, there’s nothing necessitating that these police forces, you know, have this kind of oversight. Um, now whether this app itself is legal is a very interesting question. So your face, this might sound funny because you walk around with it all day in public, but it is personal information. So when your, an image of your face, is turned into a bunch of data points and stored into a database with millions or billions of other faces that have also been, you know, turned into data, that is your personal information, that you have a privacy interest over.
Jordan: My face numbers.
Kate: Yes, exactly. Your personal face numbers. My personal face numbers. So, um, you know, it seems like under Canadian privacy law that you shouldn’t be able to scrape all of this personal information off the internet without the consent of, you know, the people whose faces it is and have police unbeknownst to them searching this database. However, whether that is something that– like I, it doesn’t appear to be something where the privacy commissioners can say, you have to stop. This is illegal. Like, I don’t think that is–
Jordan: Right. And if it’s just taking, uh, photos from places that I’ve made them public, then I mean, I think everybody is kind of used to the fact that when you sign up for Facebook or other services, you are checking a terms of service that might give that platform some rights to do some things with your data and people just kind of, whether it’s good or not, people just let that go sometimes, but this is another level. Beyond that, this is a third party making use of this data.
Kate: It’s sort of like an extension of the contract we didn’t know that we were making when we, you know, put all of our photos online. I should say that this technology is not exclusive to this app. Like this is the exact same technology that underpins how Facebook can know that when you upload a photo of you and your friends, it knows how to label those photos by name. This is the exact same technology that does that. It’s the same technology that underlies how online translation services can instantly translate. Like neural nets and this form of AI are used for all sorts of things. And I mean, I think to your point, you know, we sort of put our photos online and we think like, well, okay, am I going to get a bunch of targeted ads? I guess I’m okay with that, or maybe I’m totally okay with that. You know, this is sort of like, I don’t think too many people other than privacy experts anticipated that their images might be scraped into an app that law enforcement agencies worldwide can search, you know, without any type of oversight.
Jordan: Well, since you mentioned them, when you took this info from the police and just about the technology in general to privacy experts, what do they say?
Kate: So, you know, it’s interesting. I would say the, the reaction that you or I might have, or, you know, other members of the public is a little different from their, their reaction. They’re kind of like, yeah, you know, they’re not, they’re really not surprised. And, you know, just like I said this, the sort of the AI that underlies this app is not special or unique. You know, anyone could have built this app and, um, you know, there’s actually like, Facebook did build something sort of like this. There was, um, reporting recently that Facebook built an app for its employees that essentially let you, you know, take a photo of someone, uh, and it would instantly use facial recognition to pull up their Facebook profile. So anyone can build this thing that’s not hard to do. So I don’t think privacy experts are at all surprised. And you know, I should say, again, this is not like this company is on Google. This company is not, Facebook appears to have a handful of employees. It’s small. And nobody had ever heard of it before January.
Jordan: It’s one of the things that makes this story a little bit unsettling is that, you know, we’re kind of used to the big boys in the tech space doing weird things with our data because we’ve given it to them. Whereas somebody else that’s just a tiny company that we’ve never heard of that can come through and scoop this stuff, uh, is a little startling to the average person, I think.
Kate: Yeah, totally. And you know, yeah, it’s interesting hearing reactions from my friends and family, like my husband, for example, is kind of like, you know what? This is inevitable, so we should have some sort of like oversight body that, you know, gives access to people who want to use it. Whereas I’m like, I don’t know if this is inevitable. Like, you know, we could, if we wanted to, you know, enact legislation provincially, municipally, federally, that would create more checks and balances for law enforcement agencies or any, any other public institutions or anyone who wants to use facial recognition software or other forms of surveillance, other forms of AI. We haven’t done that.
Jordan: When you spoke to the police, did they give you any sense of whether this was an okay, we informally tested this and now we’re stopping it, or whether or not this was something that they’d investigated and may still come back to using down the road?
Kate: Um, they haven’t ruled out. What they said was that they will work with the privacy commissioner’s office and the crown attorney’s office to determine if and how they will use this thing. You know, something mentioning the Crown attorney’s office is interesting because it’s unclear whether, if this was used in an investigation, whether anyone would find out, because if it wasn’t sort of central to the narrative, if it was just one tool used to make sort of an initial match, um, at the beginning of an investigation, it might never be disclosed to the defense. Um, so that’s interesting. On the other hand, you know, if it does infringe on people’s charter rights, I don’t think you could build a case on it. So I’m actually curious if they have used it in investigations and it is determined to be unlawful or infringing on people’s rights. Well, those investigations have to be thrown out. I have no, I have no idea. We’re very early in trying to sort out all the implications of this.
Jordan: That’s really interesting though, that it could end up just being tested in court and all of a sudden you find out one way or the other, whether this is a tool that’ll work, right?
Kate: Yeah. Yeah. And that would be a nightmare, obviously for the police, if they, and you know, I should say, um, we’ve talked a lot about what the privacy experts think about this. You can also understand why a police officer would want to use this. Let’s say they have a cold case, you know, a murder or rape, and they have photos of a suspect and it hasn’t turned up any hits on their mugshot database. Maybe that person was just never convicted of a crime. And they have this tool that could potentially pull photos from, you know, anywhere online and make a match. Like you can understand why that is a very attractive tool for law enforcement agencies.
Jordan: Yeah I think we probably aren’t having this conversation unless there’s some serious utility to this thing.
Kate: Of course. Yeah. And actually, Clearview AI had a blurb on their website and we don’t know whether it’s– anything more about this blurb, but it’s, it was from an apparently from a Canadian law enforcement, uh, um, um, officer. And it said that in a week and a half, they were able, using Clearview AI, they were able to make, um, eight matches on either victims or suspects of, um, sex crimes. And so, you know, this, if this thing works as well as they say it does, you can see why police would want to use it.
Jordan: When you talk to, uh, the privacy experts, is there any way that this goes back in the bottle? Like it always feels to me, cause we have conversations about technology a lot on this podcast and like, whether it’s good or bad, once it’s out there, it’s out there.
Kate: Yeah, it’s, it’s a great question. I have, I have no idea. You know, it’s hard to see how you dial this back because even if Clearview AI, you know, goes away or it becomes too controversial, controversial for law enforcement agencies to use, there’s nothing stopping another company from coming up and doing the exact same thing. And like I said. This is not like rocket science. This isn’t like groundbreaking, you know, anything like it’s, it’s, you know, it’s something that, it’s technology that has been waiting out there for someone to put together.
Jordan: What do you think happens next? What are you waiting to hear from or find out?
Kate: You know, I am still really interested on the legal implications of this. I’m, I’m, it’s unclear to me exactly what the privacy laws that we have now can and can’t do with respect to this tool. Um, I’m also really, I want to know who else is using it. Like we’ve, we’ve found out that, you know, three, um, law enforcement agencies in the GTA are, and by the way, these agencies service almost 5 million people. We haven’t heard back from one. You know, as I said, the OPP and the RCMP haven’t said either, won’t confirm either way. Um, so I’m, I just wanna know, you know, who’s using this thing and, and how? Are they using it for investigations? Where they just like, was informal testing, like testing it on their buddies? Like I have no idea. So I have, I have a lot of questions.
Jordan: Well. Thank you for answering some of mine.
Kate: You’re very welcome.
Jordan: Kate Allen is the science and technology reporter at the Toronto Star. That was The Big Story. If you want more, they’re at our website, thebigstorypodcast.ca. You can also write to us there. There is an email form you can fill out and I promise you, I and Claire, Claire and I, read every single one. You can also find us on Twitter, the more primitive mode of communication at @thebig storyFPN or at @frequencypods. And we are everywhere you get your podcasts and Apple and Google and Stitcher and Spotify on the free version of Luminary and I guess on the paid version too. Thanks for listening. I’m Jordan Heath Rawlings. We’ll talk tomorrow.
Back to top of page
