TED Radio Hour - What's In A Face

Episode Date: December 9, 2022

We think our faces are our own. But technology can use them to identify, influence and mimic us. This week, TED speakers explore the promise and peril of turning the human face into a digital tool. Guests include super recognizer Yenny Seo, Bloomberg columnist Parmy Olson, visual researcher Mike Seymour and investigative journalist Alison Killing. See pcm.adswizz.com for information about our collection and use of personal data for sponsorship and to manage your podcast sponsorship preferences.NPR Privacy Policy

This is the TED Radio Hour. Each week, groundbreaking TED Talks. Our job now is to dream big. Delivered at TED conferences. To bring about the future we want to see. Around the world. To understand who we are. From those talks, we bring you speakers and ideas that will surprise you.

You just don't know what you're going to find. Challenge you. We truly have to ask ourselves, like, why is it noteworthy? And even change you. I literally feel like I'm a different person. Yes. Do you feel that way? Ideas worth spreading.

From TED and NPR. I'm Manushe Zamoroti, and today on the show, what's in a face? Do you have any idea how long a face will stay in your mind? Like, once it's in there, is it there forever? I actually don't know, but, I mean, as an example, I have come across faces that I remember seeing. from when I was younger than 10. Oh, wow. Yeah, so I have come across, like, school teachers or classmates or people I remember in my

neighborhood when I lived there as a child. And it's a lot of harder to explain where I've seen them. Oh, so you're saying it's more just a sense of familiarity? Yeah, and it's almost like this intuition or this kind of ping. And I know, like, I am so confident I've seen that face before that happens instantly. This is Yenna Suh, and she is what's called a super recogniser. So super recogniser. Firstly, I find that term very cringy.

Oh, no. Basically what it is is super recognisers are people who are maybe on the top 1 to 2% who are very good at remembering faces. And I'm told that it's about 80% or so of the faces that we see we remember. So it's very high in comparison to the average person. Yenny knew she was always good at recognizing faces. But about five years ago, she realized that her ability was really unusual. I just turned on the TV and I happened to come across this show about people with different abilities. And I saw this gentleman was based in the UK and he was a police officer.

And they made him do a test where. he was standing in a really big train station with lots of people going through. You've got to try and find my four actresses who have hidden themselves away in the crown or be wandering around. And he was shown photos of, I think, a handful of different faces and he had to pick out the people that he'd seen on the photo. But the trick was they would, you know, put a wig on them or they'd be wearing a different hat or glasses.

Black leather jacket, blue jeans. Brilliant. And he was able to, I think, pick them all out. When I saw that, I just got through swamps, and I just had this really strong confidence that somehow I'd be able to do those tests. Is it the lady in the black jumper, cream top, and blue jeans? Brilliant. Yenny took some tests online, and she did really well.

So she got in touch with a researcher in Australia where she lives, who confirmed that she was indeed a super-recognising. I ended up visiting their lab in Sydney and they put some sort of sensor detector. So they saw where my eye movements, how it worked when I was exposed to a face. And it's not that I pinpoint on one feature. I would not focus on the eyes or nose or mouth or the shape of the face. It's just the whole. The face is a whole leaves kind of an imprint in my head.

So have you ever found like your ability useful then? like other than like you know fun party trick yeah i mean when i was in uni i worked at a clothing store and we i mean i ended up catching a shoplifter because we had a team meeting and there was a particular shoplifter who would repeatedly steal like the highest priced item in the store and they had this CCTV footage of her and it was just this really grainy black and white photo and they showed it to us during our team meeting in the morning and they stuck it on the wall and I looked at it and I was like, all right, I don't know if I'll be able to catch that person. I didn't really think much about it. An hour or so into my shift, that exact person walked in and I just knew straight

away it was that person, even though the photo was really grainy. I just knew. What did you do? We had security guards in our store so I just had to, we were wearing walkie-talkie type of things and I just told them, yeah, she's here. She's just walked into the store. So maybe you guys should go, I have a chat with her. They ended up catching her, and then they had to call the cops in. And so that was my one crime-fighting experience. From what I understand, a lot of super recognizers work in or work with law enforcement, is that, or in some kind of security capacity, is that not something that you sort of thought,

well, you know, I could actually make money off of this? I mean, at one point, I think I did consider it. but I think like it's still very new and the research in this area is still developing. I know like countries like the UK, like their police enforcement have started recruiting officers who have that ability. But I always thought it was a little bit creepy that I, I don't know if creepy is the right word, but I always thought that, you know, it would be perceived as being a bit like I was a stalker or something. Yeni sees and processes faces in an extraordinary way.

But technology is quickly passing her superhuman abilities. Most of us already use facial recognition to unlock our phones and tag people in photos. Governments, law enforcement, and companies can use cameras and algorithms to collect and identify us. But where will we draw the line? Today, what's in a face? ideas about the promise and peril in turning the human face into an everyday digital tool for anyone to use. I was actually literally just today talking to a facial recognition vendor. Hmm, what about?

So they're in the middle of filing a patent where artificial intelligence or machine learning system will look at your face and determine how you feel. This is Bloomberg tech columnist Parmi Olsen. which will allow them to analyze the faces of stock market traders and bond traders to get a sense of where the market is moving based on the emotions shown on the faces of these traders. In a way, that sounds maybe a little bit innocuous, if not a very odd way, and potentially a disastrous way to determine where the market is going. I don't know that that would work. But I think the question is, well, what happens when all these things,

different vendors and stakeholders have access to our faces and can maybe get to a point where they want to start drawing inferences about us based on our faces. Now, even if making market decisions based on the minute facial expressions of day traders sounds far-fetched, Parmi says the basic technology behind it is not. So these systems are essentially trained on millions of and millions of actual photos of people. And the more data it has, the more accurate it can get. And I think the concern is that this technology is so widespread and so actually not that

difficult to build. Some of the technology is open source. There are billions of images of faces on the Internet. It's relatively cheap to do it. Yeah. And it's so cheap that you have written about how this. software is now used pretty widely in retail, even gas stations, convenience stores. Yeah, I think the main reason that retailers want to use facial recognition in their shops

is to actually look for unwanted individuals. So there was a chain of stores in the UK that hired a security system, a facial recognition security system to be installed. Let's go to Aylesbury, Southern England, and to a Budgeon's store. Parmy Olson continues from the TED stage. Now, Budgens in this particular town has been having trouble in the last few years with people coming in and stealing meat from their refrigeration aisle. So a year ago, they installed some new technology from a company called FaceWatch,

and through their usual CCTV cameras, FaceWatch's computer and software would scan every single thing, face that came into the Budgens and match it up against a watch list. Now, this watch list is processed by FaceWatch, and Budgens can also add to it if they suspect someone of stealing. And I called up the Budgens and asked how they thought it was working. And the staff member there told me that his phone gets pinged up to 10 times a day with an alert to say that someone has walked in the store who matches the watch list. So if that happens, he might call the police if it's an aggressive person,

or he might just say, hey, you're on CCTV. And actually, it works pretty well, he said. He thinks it's helped. But there's a few concerns about FaceWatch. So first of all, to get on the watch list, you don't have to be arrested, and you don't have to be charged by the police. There's no real legal due process. And the other thing is that to be uploaded onto the servers of FaceWatch

to be on a watch list. You can be on it for up to two years, and you won't be taken off. So this is a security company that relies on watch lists. And anyone with clearance, I guess that could be a store employee, they could add someone to the list and then what? That information is shared? Yes, that's right. Each store would have their own watch list,

and they would share the watch lists with each other, so you'd have an even bigger watch list. And yes, the people who are on these systems, this is a private system. This is not something where there's a court order or a warrant or anything like that. This is totally done privately by a business. It's their own private watch list that they've put together. So, Parmi, one of the underlining problems with this kind of mass surveillance is that sometimes the algorithms are wrong. Right.

When I talked to one of the people who worked at one of the, stores, they said that about 25% of the time the system was wrong. So they would get the alert, get told that person had walked in, then they'd walk around, and they'd see actually it wasn't that person. And so they had to really be careful to trust that the system was correct. In real world, when, you know, the lighting isn't that good and the image might be a little bit grainy, not surprisingly, the system was getting it wrong one out of four times. Wow, and I can imagine someone thinking like, okay, well, it's a grocery store. But if you're talking about a situation involving law enforcement, that could get quickly escalate, I would think.

Yeah. So a police officer has a body cam with facial recognition or a camera on their van with facial recognition and they detect someone. And if that person has increased melanin in their skin or they're black, essentially, then it is more likely to make a mistake, a mistake. in identifying that person. And the reason is that the databases that these facial recognition models are trained on typically have way more white people than black people. And so the system just isn't trained enough on black people, so it doesn't identify them properly. It makes more mistakes.

And that has happened. And it's probably going to continue to happen too. When we come back, what are we willing to stomach in a face-tracking? filled future. On the show today, what's in a face? I'm Anoush Zamoroti, and you're listening to the TED Radio Hour from NPR. Stay with us. It's the TED Radio Hour from NPR. I'm Anoush Zomerode. And on today's show, what's in a face and how companies will use them to make money? We were just talking to Bloomberg columnist Parmy Olson. Each of one of our faces has a face print. It's like a fingerprint, except it's a string of numbers that corresponds with the image of our face.

And there are lots of databases of people's face prints out there on the Internet. And even if they aren't identifying us by name, our faces can be tracked, categorized, and profited off of. For example, a few years ago, Walgreens installed cameras in some of their stores that identified shoppers by age, gender, and then displayed targeted ads. Oh, new sprite? Like, do you look male and around 20 years old? Buy a sprite. Buy a sprite. 50 and female, maybe a green tea.

Your Subaru has a driver monitoring. Carmaker Subaru's new vehicles use facial recognition tech that create user profiles. And even alert a driver if they seem distracted or tired. And then there are casinos. So there is a casino in London, which has facial. recognition cameras dotted around all the different rooms. And it uses that so that when high rollers walk in to a certain room, then the staff get an alert on their phone, which gets sent to an encrypted chat app they use called Wicker. And then they notify each other like, oh, so and so,

this hostess should go up because that's the high rollers favorite. We know that they like this particular type of food and this particular type of drink. And so they can actually provide a better service, and they call it their white glove service. And I remember asking the head of security, well, are the patrons actually a little bit put off by that? And he said, not at all, not a single one thinks that that's even the slightest bit creepy. They just see it as part of the service. It's what they expect. And I think it's a nice little allegory for just how this kind of surveillance is going to eventually come to serve the rest of us as consumers, that the convenience will ultimately be something that we just take for granted and we won't worry too much about the price that we're paying with

our privacy. And I think that is just the way it'll go. I mean, I guess I can see the appeal if you're opting into a luxury service. I mean, that just feels very different than being tracked while you're walking down the street or going into a store. I just don't think that that is something a lot of us would sign up for. Yeah. I would say that facial recognition definitely. has become a controversial subject. And I think that's made it difficult for brands, for advertisers, who might be able to benefit from using it to target their brands at people. They've had to take a step back. Yeah, there are a number of retailers now, I think this is what you're referring to, who are facing lawsuits for surveillance, for gathering data on their customers without consent. But really, like for now, it just feels like, lack of all because we have not figured out as a society what we think is okay and what isn't okay when it comes to face tracking.

Yes, that's right. Companies who develop these kinds of systems need to be very careful. There needs to be more ethical oversight of how these systems are developed. And right now, that's only going to come from regulation, which is a couple of years away, yes, but also campaign groups. And there are some really good civil liberties groups in the United States and Europe who are really keeping an eye on this and just helping keep companies on their toes. If there wasn't the amount of upset that have been created around facial recognition, I think there'd be a lot more advertisers using it right now. But because people have really rung alarm bells about it, then I think that's made companies really just take stock and just sit back and just say, okay, let's just be a little bit more cautious about how we use this. And I think that's a really good thing. Given that there still aren't major regulations out there around facial recognition, I mean, are we just at the point of no return here? I mean, can't go back. There definitely needs to be more laws and regulation, but we have sort of gone past trying to force companies to design algorithms in a way that are safe and ethical because the algorithms

already out there. But there is a law coming from the European Union called the AI Act, and it actually bans all forms of facial recognition for surveillance by police, unless it. for trying to combat terrorism. So that's a pretty blunt rule. And I mean, that's going to be the first kind of comprehensive legislation around the use of artificial intelligence algorithms. I think the issue with it is that it is so broad. It's not just about facial recognition. It's about all forms of AI. So whether that's recommendation systems on social media or facial recognition, you know, it covers a lot. And so enforcing it, I think,

think is going to be difficult. You know, one of the reasons all this tracking is possible is because we have accepted the idea that cameras are in our pockets all the time. They're on our doors. They are all over public spaces. And we're okay with it. We are okay, largely, with being surveilled. Yeah.

There's something like 20 million homes in the U.S. have a video doorbell. The thing about ring doorbells that I think is really important. interesting is that actually the studies that have been done about just how affected these cameras are in reducing neighborhood crime show that the evidence is really flimsy. There's actually not much evidence that they do reduce crime, but the big impact is on human sentiment. So the owners of these cameras feel a greater sense of security and a greater sense of control. But then on the other hand, we also collectively come to accept that our behavior is being watched. So, yeah, take that how you will.

I think we are just an increasingly surveilled society. And I think people are just like slow-boiled frogs. We're increasingly accepting of it because it's just what's happening for better or worse. That's Parmy Olson. She's a tech columnist at Bloomberg, and you can see her full talk at ted.npr.org. And earlier we heard from super recogniser Yeni Sa, who works as a translator in Australia. On the show today, what's in a face?

Often, to understand how technology will change our lives, we just need to watch a movie. Like this one, released in 2021. So the champion was a film shot, and made in Poland. So everyone's speaking basically Polish or German. This is Mike Seymour. He's a researcher at the University of Sydney and works in the film industry, in special effects. And it's a great film. It's a true story about one of the first members of Auschwitz, who was a boxer. Terribly moving story, but of course, only in Polish or German.

And usually when there's a foreign film that wants to break into the English-speaking market, there are three options. Dubbing it, so we get somebody else to voice up. over a different piece of dialogue, but of course the lips aren't right, so it looks kind of odd. Or there's subtitles. Or we have the new version of what we call facial reenactment. Facial reenactment. It's a new technique that Mike and his team used on The Champion. We got involved as part of a team to convert the entire film to English.

So now if you were to watch the film in English, every actor speaks as if they'd been shot in English. The Dodge King of Warsaw, totaling almost a hundred fights in bantam and featherweight. So we've replaced effectively the actor's faces with their own faces, saying the lines in English. He looks more like a small rooster plucked of its feathers than a champion. Okay, so Mike, when I watch this English version, it's seamless. It's like their mouths, their faces, everything looks like it was originally shot this way. In English, is this common in the industry? Is it normal?

Well, it's the first time anyone's done it in the world, but hopefully it's going to become normal. As Mike went on to explain, months after filming, the actors re-recorded all their lines in English, as cameras taped their voices and facial expressions. And then through a process called neural rendering technology, their faces were replaced. So it looks like the film was just shot twice in two different languages. Come on, Jenny. Come on. The film industry is always pioneering new tech to trick our eye, to make someone or something look real.

But over the last few years, Mike has been developing ways to use these techniques in our real lives. Yeah, could we take this tech and just sort of use it outside the film industry that fascinated me, where I say, well, hey, I don't want to get shaven and put on a suit for my important meeting today. So I'll just flip a switch and get digital makeup. and I'll look a whole lot better and a whole lot smarter. And I would be able to say speak in Korean when I absolutely can't speak in Korean. And that we hope would facilitate much more genuine communications across cultural divides. Wow.

You're saying that maybe one day, if I have relatives all over the world who speak all different languages, but maybe one day we could do FaceTime and it would sound as though I was speaking fluent, Swedish, and they were speaking back to me, well, they can't speak English, but that we would hear each other's native tongue and wouldn't know the difference. It would look as though I could speak fluent Swedish, but it wouldn't look like I was not myself.

Yes, there is a lot of modern technology that's very sophisticated. That would, we think, benefit from being able to have an extra layer of communication that you get from face-to-face interaction. We're kind of this nexus point where that's possible. Mike Seymour picks up from the TED stage.

We're interested in being able to see if we can put a face on technology because how would you react when a computer reacts to you with a smile? Would a six-year-old learn maths better if there was a six-year-old teacher on the screen? What about if it was a slightly older version of herself? Would a grandparent having a cup of tea be more likely to check in with a computer system if they didn't have to log in and type, they could just talk to a virtual agent that actually was somebody from their past.

This is what we're excited to explore with digital humans. Our ability to produce digital humans up until recently has been quite limited, but we're now seeing interactive digital humans starting to appear. The doors are opening. We are at an inflection point. We have this perfect storm of faster GPU graphics cards, new artificial intelligence, deep learning algorithms,

and great advances in game engines. It's an incredible combination of things coming together. This tremendous nexus of points is just providing us with an extraordinary opportunity of things that we can do. The important thing about this technology is that we can now use this to get these faces to work with us in real time.

In other words, this is a really key point. The faces that we're talking about can talk, interact, and see us. Okay, putting faces on our technology. Tell me more about how you see this working and the reasons why we would want it. Yeah, I mean, there are a lot. Already in New Zealand, there is automatic sign languaging. So if somebody is speaking, digital human signs for the deaf community, you might have an assistant sitting in on a Zoom call that you can ask to help book future things, take notes, do stuff. In age care, you could have an assistant. that logs in with somebody each day and make sure that they're okay and are lucid and they've taken their pills, not to replace a healthcare worker, but just simply that to make sure that they're okay and facilitate them staying in the home longer. And so in a world where we're saying,

hey, you know, even to use the phone, there are no buttons now. You have to, you know, swipe up, swipe left, do all this stuff. People are like, I have trouble with that. And so we could bring a face from their past that would be the one that they interact with that. technology. You don't think that would be odd to someone that if you said, well, this is your sister. She's not actually your sister. She just kind of looks like your sister and she's going to help you use your phone. I don't know. That might freak me out. Yeah. You know, you've just touched on a really interesting point. People when asked traditionally say, I wouldn't like that. So if you project ahead, you say, hey, would you have a digital human tell you what to do? No, no, no, absolutely not. That would

be freaky. And yet, every time we do a lab test, they completely don't do that. I looked up one of those services that might be available in the near future. Hello, this is Sol, Dr. Beanie's assistant. Hello, Sol. This is Tyler. I wanted to ask you about my recent surgery. In the demo video, a man is home after knee surgery and consults on his laptop with his AI nurse. Soul seemed to know her stuff. The discharge summary states that you should take the medicine about 20 minutes before you put on your headset for your virtual reality meditation therapy. Mike, I don't know. I have to be honest, I was a little unnerved by soul.

Sure. What I would say is it was her lack of authenticity that probably bothered you, not the digital representation of the face. Once you get to a certain level of quality, you kind of pass what we've referred to as the Uncanny Valley. So you now got something that looks pretty darn good, it doesn't matter whether you can tell it's real or not. That's not the deciding factor. It's the authenticity of the emotional kind of response that matters. And that's the driving factor. And so for us to succeed in those cases, we really need to make sure that it's the sort of the back end behind the face that's delivering what's wanted, not so much the face itself. Speaking of the Uncanny Valley, you did a demonstration. You did a demonstration

on stage where you showed off a very realistic digital version of your head, your face on a screen that you could control. Hi, I'm Mike. Well, kind of virtual Mike, really. This is our digital human project, which is a collaboration of a whole bunch of people coming together to produce, well, a virtual human. I mean, people, like right now, we can make digital avatars of ourselves, but not like this, not to this realistic extent.

How hard was that to build? Yeah. I mean, we sort of are close. I mean, that one took a lot of people. So we scanned my face, and I got my face done in one of the most high-resolution facial scanning systems in the world. It produced this super realistic version of my head. Then I could puppeteer that in real time or have it driven. So how do we do it? So first, we scanned my face. This allowed us to produce a very complex digital avatar of my head, or a digital puppet. Then with a camera mounted on a head rig, the computer can actually read my face.

An advanced AI engine then basically interprets that into expressions. Now, the computer can tell the digital puppet what to do. In effect, what's happening is it's the computer telling the muscles in the digital mic how to smile, talk, or do things. It makes me wonder if we might get to a point where kids think, well, I would much rather deal with my extremely realistic looking too, on my laptop who responds to me, but who doesn't actually give me a hard time and won't be offended if I tell it or shut up. How do we make sure that people don't choose these artificially

intelligent agents over humans? Brilliant question, and I wish I had a definitive answer. I can only give you my hope. So imagine I'm a vet, I come back, but I'm an 18-year-old guy. I've experienced some horrendous experiences in a conflict. And I'm now suffering from all sorts of sexual dysfunction. I cry at night. I have like things I'm really embarrassed about and ashamed of. I kind of want help, but I don't want to have to sit there and tell a doctor that. But I'd actually like my doctor to know all that so that they can help me.

If there are ways where you can communicate that to effectively like a digital nurse, a digital doctor substitute so that the system can know it, but you don't have to face them and look them in the eye and say, you know, I have sexual dysfunction. But you can then get treatment and help and the system knows and can look after you. That's a tremendous benefit. So hopefully for that generation, there'll be tools that appear in their everyday life that just make it a bit easier and reconnects them with people, not takes them away. And I'd like to think that if I had teenagers who were in distress and teenagers,

that were struggling, if there were tools that helped them, that that would do just that, it would help them. It wouldn't replace human contact. In a minute, the ethical dilemmas with giving our technology a face. I'm Manus Zamoroti, and you're listening to the TED Radio Hour from NPR. Stick with us. Hey, if you're looking for a way to support our show and public media, I hope you will consider signing up for the NPR Plus podcast bundle. You can listen to a bunch of NPR podcasts, including this one, without any sponsor breaks, and you can even access behind the scenes episodes from some of your favorite shows.

Go find out more at plus.npr.org. And thanks. It's the TED Radio Hour from NPR. I'm Anoush Zamorodi. And on the show today, what's in a face? We were just hearing from Mike Seymour, a film industry veteran who now wants, are virtual helpers to look and act more human. Hi, I'm Mike. Well, kind of virtual, Mike, really.

This is our digital human project, which is a virtual human, but one rendered in real time. Puppeteered or driven in real time, rendered in real time, and not only that. Mike says that technology with a face can better interact with us. Talk to patients about how they're feeling. Ask students where they're struggling in algebra. Coach brain injury survivors to be more. self-sufficient. So many different types of technology, so many different use case scenarios, as you've mentioned,

I expect we're going to see this area explode in the next few years. But along the way, wow, do we have a lot of ethical dilemmas to sort out? I mean, you're just reminding me, we've talked about this on the show a lot, the deep fakes. You know, there's the famous case of seeing President Obama. Obama giving a speech that he never gave, where do you see some of the pitfalls that we need to watch out for? Mnisha, I completely agree with you. Some of the applications of this technology by what I would describe as bad actors is appalling

and just absolutely indefensible. It's a really interesting sort of fundamental ethical question, is the technology good or evil or is it the use of it and the application of it? And I only can say this. For me personally, you can use steel to make ambulances or tax. tanks. I'm in the business of trying to see if we can't use it to make lots of good ambulances. I know some people are going to make tanks, but that's something I can't have any control over, but I do feel that it's going to happen. And the best line of defense we have to the

deception that can be done by this technology is an informed public. So if you see something that's highly improbable, you're going to say, hang on a second, that's probably been faked or not real, or you'll dig in to try and discover its authenticity. So there's a lot of ways we can produce inaccurate material, but an informed public that is aware of what's going on, that understands what the sort of limits of technology are, and, you know, where it's going is vital to being able to do this sort of stuff. But are we just barreling towards a future where, you know, your identity gets stolen, but it's not just your social security number.

It's your face that can make it look like you're handing over your bank account number to someone on a Zoom call. Or is my imagination way ahead of the technology? Gosh, I mean, people will deceive people with this technology. But yeah, when I'm talking about these AIs, they work very well when you've got a limited amount of stuff that you're asking them to do. So if I was having an agent that was helping you as a maths tutor, and it was discussing maths and explaining mathematical concepts, that could be completely plausible and look photorealistic and wonderful. But if I asked my maths assistant, what does it mean to understand existential philosophy in France?

It would completely blank out. So we're not talking about a general intelligence. People quickly extrapolate to that, but we are so far away from that. General AI intelligence is a long way off. But as I say, these plausible, realistic, domain-specific applications in health, in age care, in all of these sort of parts of everyday life, completely plausible and extremely likely to happen because we just love faces. We love face-to-face communication.

We love seeing people face-to-face. Humanity just likes faces. We're talking about just putting a face on technology so that it's a bit more friendly, a bit more empathetic, a bit more engaging that has an emotional response, and therefore we find it to be a better, more pleasurable experience. That's Mike Seymour. He's a researcher and academic at the University of Sydney. You can see his talk at ted.com. On the show today, what's in a face? how our faces are captured, where that data ends up, and who has access to it.

Do you use Instagram? Do you let Google Maps track you? Do you, I don't know, let open your iPhone with your face? Oh, God. So I'm pretty privacy conscious, as you might imagine. I don't allow Google to store my location. I don't have the face unlock turned on on my phone. And that is in part because,

I'm just aware of how sensitive the data is and I feel so self-conscious about it. This is Alison Killing. She's a journalist who, ironically, uses all sorts of data that's available online to track the actions of authoritarian governments. So all of the digital traces that we leave behind on the internet, like how can we use those to investigate? And I mostly focus on human rights. In 2021, Alison won the Pulitzer Prize for her investigations into China, a place where

people's faces and movements are constantly being watched. They've really worked to sort of cover cities in a way that they are able to obtain as much data as possible. So placing cameras in high traffic areas, so for example, at the entrance to a neighborhood, where they can then say, okay, we know everybody who is in this neighborhood now. We know whether they're in or when they've left. China has the world's largest surveillance network. And cameras watch over residential. commercial complexes, office buildings, train stations, shopping malls. So these very high traffic places where they can then say, like, okay, these are the people

who are in this area so that they can then control that area. They're collecting a lot of data and there's huge ambition about the things that they would like to do with it. A lot of work has gone into the processing tools at the back end of this software to identify people by gender and age and then controversially also by ethnicity. And as you may know, the Chinese government has been tracking one large group of people in particular. The Uyghurs, a Muslim ethnic minority in a Western region called Xinjiang. Yeah, there's been a lot of discrimination.

There's been intimate and crackdowns on the practice of Islam. But then in 2009, there were two Uyghur workers killed. And that led to protests which turned violent and about 200 people were killed. And this was kind of the start as well of the Chinese authorities starting to crack down on the region and seeing it as a very violent place, seeing it as a site of terrorism. The incident ushered in an era of Chinese control of the Uyghurs using all kinds of tactics. So I think from sort of 2013, 2014, we saw the start of this real campaign of oppression in Xinjiang with the installation of this incredibly invasive surveillance state.

And the New York Times has done a lot of investigation on this topic where they actually found documents from tech companies, which were boasting that they could identify Uyghurs using facial recognition software. So one of the first things that we saw was the creation of this network of detention camps. You know, Alison, we were just talking to Parmy Olson. tech reporter about how people view facial recognition in the Western world. And it often feels like what if scenarios. But here in China, we are talking about the worst case scenario come true with proof that a

minority are being tracked and rounded up because you could see the camps on satellite imagery. Yeah. In the satellite imagery, we saw them starting to appear in late 2016. And these stories started to emerge that hundreds of thousands of people. had been disappeared into these camps. And nobody knew where they were. In the far west of China, evidence is building that a monstrous crime is taking place.

An estimated one million Chinese Muslims had vanished. Uyghurs are now being rounded up by the hundreds of thousands. There are many accounts of people who have had their relatives disappear into the camps. And we don't really know what's happening to them. Alison Killing picks up the story from the TED stage. I got involved in investigating she. in the summer of 2018, when I met Megha Rajagopalan, an American journalist who had been working in China for several years. Over the past few years, China has been carrying out a campaign of

forcible assimilation, and several nations have described it as a genocide. It's estimated that over a million people have been disappeared into detention camps. And while the Chinese government claims that these are part of a benign program of re-education, dozens of former detainees described being tortured and abused, and women being forcibly sterilized. And yet, for a long time, we lacked information about what was happening in Xinjiang because the Chinese government controls the internet tightly

and restricts journalists' work in the region. Journalists would be followed or detained, and the authorities occasionally even went so far as to set up fake roadworks or stage car crashes to prevent access to certain roads. Local people who did speak to journalists, faced the risk of being sent to a detention camp for doing so. Megga had been the first journalist to visit one of the camps,

but shortly after publishing her article, the Chinese authorities declined to renew her visa, and she had to leave. Other journalists had managed to visit a handful of the camps, but this still represented a fraction of what we believed was out there, and no one knew where the others were. But Mega was keen to find the rest. She just needed to find a way to work effectively from outside China.

China. And so this is where you come into the story, Alison, because you and Megha decided to team up. Yeah. So I met Maga at this workshop in the summer of 2018. I've been doing a lot of then cartography work and satellite imagery. And we got talking and we realized that we maybe had a complementary skill set to be able to find these camps. You know, the way that Megga had found this first camp was through satellite imagery. And so she had the idea that that could be a good way to find the rest. But it's still, like, Xinjiang is absolutely massive. So you can't just like scour all of the satellite imagery of the region.

We needed to work out where to look. There was no street-level imagery, but as I zoomed in on the satellite images, this weird thing happened. A light grey square suddenly appeared above the location of the camp, and then disappeared just as quickly as I zoomed in further. It was a bit like the map wasn't loading properly, but then I zoomed out and in again, only for the same thing to happen.

I realized it couldn't be a problem with the map loading because the tiles would have been in the browser's cache. And when I found the same thing happening at the other locations we knew to be camps, I realized that we had a technique we could use to find the rest of the network. It's quite rare for maps and satellite images, to have these blank spots

because blank areas tend to draw attention to themselves. But here we got lucky. Obscuring the camps had inadvertently revealed all of their locations. We worked with developer Christo Bouchek who specialises in documenting human rights issues and building tools for open source researchers to map the mask tile locations. We had to work quickly and secretively

to map the mask tiles before anyone found out what we were doing and removed them because our investigation relied on access to that information. The idea was that we could go and look at the masked tile locations and then look at that same location in other unaltered satellite imagery and see what was there. Zooming in on the satellite imagery, we can see the barbed wire in the courtyards

that creates exercise pens for the detainees adjacent to the buildings. In other images, we can even see people, all wearing red uniforms, lined up in the courtyard. These features could help us decide whether a location was a camp or not. As we investigated further, we realized that the camps program had evolved, away from the early days of makeshift camps in former schools and hospitals, and had become more permanent, that the camps were now larger, higher security, and purpose-built. This is the largest camp that we know of. It's in DeBan Cheng.

The complex is two miles long, and it would cover a quarter of New York Central Park. In the satellite images, we can see the thick perimeter walls, the guard towers, and these blue roof buildings, which we believe to be factories. We estimate that this complex can hold over 40,000 people without overcrowding. In total, we found 348 locations bearing the hallmarks of camps and prisons, and we believe that this is close to being the full network. We estimate that these facilities have been built to hold more than a million people. That's enough space.

to detain one in every 25 of Xinjiang's residents. Wow, your one little lucky revelation, finding that quirk on the digital map, led to a horrifying and huge discovery. And how did China respond to the allegations? So at the beginning, when the rumors were first emerging of all of these people disappearing into camps, there was denial on the part of the Chinese government that this was happening.

By mid-2018, the UN had made a statement about what was happening in Xinjiang and raising concerns and sort of saying it was one of the most urgent human rights crises in the world at that time. And the Chinese government was then under pressure to respond to that. And what they started to say was like, well, you know, these places do exist, but their education and vocational schools. People are there voluntarily, their learning skills which will allow them to get higher paid factory jobs. That wasn't true. People were taken there forcibly. In fact, the people who were initially targeted to be sent to the camps were the most highly educated people in those

communities. So, you know, the Chinese government's claims about these being vocational schools just weren't credible. So where do things stand now in terms of what you can do with this knowledge that you have accumulated other than share it with us? Yeah. One of the big things that has been done. I mean, we've seen sanctions on key individuals within the Chinese Communist Party. We've also seen sanctions on goods coming out of Xinjiang. The Uighur forced labor prevention act came into force earlier this year, and that bans any products coming out of Xinjiang because it's very, very likely that goods coming out of Xinjiang have involved forced labor, and it's very difficult to prove that they haven't. And so that has also been a big impact.

that we've seen. With social media data and satellite imagery, we can provide evidence of human rights abuses in a way that wasn't possible before. We can move beyond looking at individual instances of human rights violations to show the scale of what's happened. We can corroborate the testimony of eyewitnesses

and provide further proof of their stories. We can build a more detailed picture of what's happening to inform policy makers or to provide evidence that can be presented in court. With open-source data, we can provide the evidence needed for accountability. And then, hopefully, action. Thank you. That's Alison Killing. She's an investigative journalist and an architect. In 2021, she won the Pulitzer Prize for her reporting. You can see her full talk at ted.com. Thank you so much for

listening to our show this week. What's in a Face? This episode was produced by Andrea Gutierrez, James Delahousie, and Katie Montalione. It was edited by Sanaz Meshkampur, James Delahousie, Rachel Faulkner White, and me. Our production staff at NPR also includes Matthew Cloutier, Fiona Gehrin, and Catherine Seifer. Our theme music was written by Ramteen Arablewe. Our audio engineer was Quasi Lee. Research support came from Cecil Davis Vasquez. Our partners at TED are Chris Anderson, Colin Helms, Anna Feele.

Michelle Quint, Jimmy Gutierrez, and Daniela Ballerazzo. I'm Manus Shameroi, and you've been listening to The TED Radio Hour from NPR.