Benjamen Walker's Theory of Everything - Targeted

Episode Date: November 8, 2016

Your host tries his hand at targeted advertising and he dooms his child to a life of exclusion with the simple gift of a stuffed frog. Plus the truth about Facial Recognition.

You are listening to Benjamin Walker's Theory of Everything. At Radiotopia, we now have a select group of amazing supporters that help us make all our shows possible. If you would like to have your company or product sponsor this podcast, then get in touch. Drop a line to sponsor at radiotopia.fm. Thanks. episode. Why is there something called influencer voice? What's the deal with the TikTok shop? What is posting disease and do you have it? Why can it be so scary and yet feel so great to block someone on social media? The Neverpost team wonders why the internet and the world because of the internet is the way it is. They talk to artists, lawyers, linguists, content creators, sociologists, historians, and more about our current tech and media moment. From PRX's Radiotopia, Never Post, a podcast for and about the Internet.

Episodes every other week at neverpo.st and wherever you find pods. This installment is called Targeted. In my quest to better understand the surveillance that makes ad tech possible, I turn to one of the smartest people I know, Gilad Lotan. So I'm Gilad, and I'm the chief data scientist at Betaworks. And most recently, I've been working on a new product that we've launched called Scale Model. And it is effectively a way to segment users and identify these cohorts of users who share interests. And not only identify the groups of users, but also understand who is important within each group.

And so I think this tool is perfect for what you're trying to do. And I suspect it'll be a great way to kind of get a sense for the kinds of people you want to target. Yeah. When Gilad told me about his new product, Scale Model, I didn't want to just have him explain how companies and marketers could use it to map out communities on social media to better target their ads. I wanted him to show me how to use it for myself.

And so with his help, I launched my very own Theory of Everything targeted Twitter ad campaign. I mean, let's face it. The election of 2016 has left a lot of folks grappling with some pretty big questions. It's like the perfect moment to swoop in and say, hey, I've got all the answers you need

right here in my podcast. Given your goal and who you want to reach, I think I have the perfect communities for you. Folks who were associated to what we now call the alt-right. But this is a mapping from probably half a year ago or even more. So before the alt-right became such a big deal, before Clinton mentioned it, before any big media outlet mentioned the Ultra.

So these are like the original Gs. The originals, yeah. That sounds like a great map because, again, what I feel I want to do with my Twitter ad is target them as like, you're interested in conspiracy theories, you're interested in different versions of reality. Hey, you might even be interested in your own. Come listen to the theory of everything.

Well said. Okay, so what we're going to do is we're going to create a campaign. All right, so we have to go to the Twitter ads dashboard. And here we are logged in through you. And when we create a campaign, we have to tell Twitter what our goals are. So we can just try for engagement, for video views, website visits, website conversions, right? If you're trying to get people to complete something. Obviously, my goal is to get the people I'm targeting to subscribe to the podcast. And even though Scale Model provides me

with a map of the exact people I want to target, I still have to craft an actual ad. But Scale Model can help me out with this, too. It lets me see all the stuff the folks I want to target are currently talking about and sharing. Podesta emails, drain the swamp. This was a week before the election, so pretty predictable stuff.

Uh, what is this? Some secret tape that Crooked Hillary wants to take in as many Syrians as possible. We cannot let this happen. You know what? I really love this. Sources have confirmed. Hashtag. Oh, we can just use the hashtag. Yeah, let's do it. Amazing. You're I really love this. Sources have confirmed. Hashtag. Oh, we can just use the hashtag.

Yeah, let's do it. Amazing. You're so good at this. It's like natural. Sources have confirmed. Okay, so want the truth? Here are the secret tapes. Listen to the Theory of Everything podcast. There's the SoundCloud link, then the hashtag as sources have confirmed. It took us a little more time to find an appropriate image. Mostly because most of the stuff these guys are sharing is anti-Semitic, horrible, and totally vile.

So I took another angle. Over the past few months, many of my friends have been sharing images of that dumpster on fire. It got a lot of traction, especially during the debates. Well, I found a gif of two healthy-looking white firemen putting a dumpster fire out. So this is how it's going to look like. There's a text above, and then there's an image of a person taking out a fire, a fireman. That feels my show. I'm putting out the fires. So perfect.

Tools like Scale Model help companies and marketers get around the restrictions that many social media platforms have on targeting specific users. In other words, when I know that my target and his followers are sharing the hashtag LockHerUp and tweeting about a particular Kentucky vape store and something called Real Truth Now, I can tailor my campaign in such a way that I'm pretty much guaranteed he will see my ad. You can also reverse engineer this. ProPublica recently took out a Facebook ad for a housing-related event that excluded African Americans, Asian Americans, and Hispanics. But as troubling as that might sound, Gilad Loten told me that most ad tech companies with any brains can do this sort of thing without ever even using racial data.

Anyone working in the ad tech space knows that you can back into demographic data. So even if I don't have racial profiles of users, zip codes carry a lot of information, demographic information. So I only need zip codes and some information about social proximity to figure out what your racial profile is or what your demographic looks like. Because of homophily, because the fact that people tend to connect to people like them. In the end, I decided to go with three ads, one aimed at a specific dude, that authority in the truth and vape community, one aimed at the alt-right, and I swapped out secret tapes for secret patterns and targeted the chemtrail community as well. After the break, find out how my ads did.

Do you lose your keys or are you married to someone who does? Well, then you need to get Tile. Tile is a tiny Bluetooth tracker and easy to use app that helps you find everyday items in seconds. Tile is the result of one man's quest to help his wife find the many, many things she lost. Tile's CEO, Mike Farley's wife, Camelia, was always losing things. Two driver's licenses, a wallet, half a dozen jackets, a cell phone, car keys, motorcycle keys, backpacks, even a rental car. Mike created Tile to help Camelia. At thetileapp.com, you can get a Tile tracker of your own. The Tile Mate, that one can be attached to almost anything, keys, luggage, backpacks. And

the Tile Slim, that's the world's thinnest Bluetooth tracker. You can put that one in your wallet or your purse. Start your tracking now at thetileapp.com and make sure to tell them the Theory of Everything podcast sent you. If we remember the three communities that we created, do you remember? There was one that was around conspiracy theory. So it was the alt-right targeted message. And that did did fairly decently, 1.5% of engagement rate. You got 17 retweets. You got a whole bunch of new followers from it and all these clicks. So people who clicked onto the SoundCloud link from this campaign.

Compared to the one that was less successful is what we called Wagner. It's that one user who seemed to be a local authority. And we said, well, let's just look at their followers. And here's what this campaign did pretty badly, 0.6%. You got like one reply, one new follower from that. So yeah, not for nothing. But the campaign that did really, really well, and it's super surprising to me. I don't know, you tell me if you're surprised. But the last one that we set up, the chemtrails, where the message was specific to the chemtrails folks, the engagement was over 2%. So people actually clicked on the ad and clicked into your SoundCloud link. Like 2% of the people who saw the ad that you posted actually like interacted with it.

2% seems low. Is it low? Yeah, no, I think it's definitely not low. I think typically if you get more than 1% of the people engaging with the content, that's an ad. That's really, that's considered good. Like you're doing a good job. It's very rarely more than 1%. Targeted advertising. 99% useless. An afternoon last week. Mathilde bursts into the apartment.

She's extremely upset. Some women were screaming at me in the playground, she says. They both called me all sorts of terrible names. One of them even shouted at our son. I look down at Artaud. He's got one arm wrapped around Mathilde's leg, and with the other, he's clutching his new stuffed Pepe the Frog doll. He can tell his mother is upset. This woman walked up to him while he was on the swing and called him a racist baby. Did he have his frog with him, I ask? Of course, Matilde replies. He takes Pepe everywhere now. It's his new favorite toy. Artaud attempts to march across the room to me, but he topples over. He can't really walk yet, even though he thinks he can.

Pepe breaks his fall. I scoop both of them into my arms and walk over to the bookshelf. I pull out a few old comics, and then I sit Mathilde and Artaud down on the couch and attempt to explain the strange saga of Pepe the Frog. In 2006, the artist Matt Fury started self-publishing a comic book called Boy's Life. It featured four roommates, Brett, Andy, Landwolf, and Pepe. Andy's a dog, Pepe's a frog, Landwolf is a super furry thing, and to be honest, I have no idea what bread is supposed to be.

Boy's Life is really just a bunch of little scenes where these four characters mess around, eat pizza, and chill out. They are all super chill, but Pepe the frog is the chillest of them all. Flipping through the pages of issue one, Artaud comes to a picture of Pepe in the bathroom and points. Da?

This is a very typical boy's life story. Pepe's pulled his pants down around his ankles in front of the toilet. He's taking a leak. Then Andy accidentally walks in and says, Whoops, before closing the door. And then in the last panel, Landwolf says to Pepe, hey man, I hear you pull your pants down all the way around your ankles to pee. To which Pepe answers, feels good, man. Now, with hindsight, you can kind of see why Pepe was destined for internet stardom. By 2010, when the final issue of Matt's comic book, Number 4, came out,

he was already a regular on a number of message boards. Kids would use a picture of Pepe saying, Feels good, man, as shorthand to express all sorts of things, like, I got a new bong, or it's Friday. The Pepe meme I remember seeing the most, though, was a picture of him looking sad and pensive. Because it was a blank slate, you could attach any text to go with this image.

I think one of the first ones I saw said something like, that feeling when you find out your girlfriend has nudes. But there were so many more. And even though most of them were about sex, drugs, and food, the meme itself transcends all of that. Pepe's super zen and super cool. Artaud reaches for my phone as it buzzes. It's a text from my friend Tom. Hey man, you better check out the link I just sent you. Your kid is all over my New York City parents' listserv. I grab my laptop and click on the link.

Sure enough, there's a picture of Artaud clutching his Pepe toy as Mathilde is putting him in his stroller. The caption reads, racist baby terrorizes Tompkins Square Park. Reading over my shoulder, Mathilde explodes. What the fuck? Okay, I continue. Obviously, it's not just nice kids who are into internet memes. Racists and anti-Semites like memes too. And when they discovered Pepe, they saw a blank slate as well. And they used him to express all the things they wanted to share with the world. Some of these things were really awful. Like Pepe photoshopped onto a historical image from a concentration camp expressing

his catchphrase, feels good man. Racist Pepes are actually really easy to spot, because he's no longer a green frog. Pensive brown Pepe usually says things like, that feeling when you have to meet your tadpoles for the first time. But it's not like the bad Pepes cancelled out the good ones. Well, until Donald Trump came along. Last summer, I started seeing bad Pepe everywhere. Overnight, he turned into this cartoon spokesperson for rants about Mexicans and immigrants,

and of course, media Jews. Sometimes he would even have yellow hair. And while it was always disconcerting seeing Pepes in Make America Great hats, I was still totally blown away when the Hillary Clinton campaign posted an explainer online calling Trump out for using Pepe imagery. The campaign even went as far as to claim that Pepe was an official alt-right symbol who stood for anti-Semitism and white supremacy. And then the Anti-Discrimination League added Pepe the Frog to its database of hate symbols. Like the database with the swastika in it. I get another text from Tom. Dude, I sent you another link. This time, I clicked the link on my phone. It's a blog called Make America White Again. They have the same photo from the parents' listserv.

But here, it's being offered as proof that New York is not only filled with immigrants. We are the future, the headline screams. This one, I hide from Matilda. The morning ADL added Pepe the Frog to their hate symbol database, Matt Fury's phone started ringing off the hook. I started seeing and hearing him everywhere. The poor man's done like 30 podcasts already. The best interview I saw was in The Atlantic. A journalist asked Matt how he feels

about the whole thing. This is his reply. My feelings are pretty neutral. This isn't the first time that Pepe's been used in a negative, weird context. I think it's just a reflection of the world at large. The internet is basically encompassing some kind of mass consciousness. It's just out of my control, what people are doing with it. Obviously, that political agenda is exactly the opposite of my own personal feelings, but in terms of meme culture, it's people reappropriating things for their own agenda. That's just a product of the internet. I think Donald Trump is kind of a cartoon character himself, so I think it was just a natural thing, maybe to appeal to younger people or something. I think what's

happening now is overshadowing the importance Pepe has as a symbol for youth culture, and it's been taken out of context and turned into something other than that. I honestly just think it's a phase, and come November, it's just going honestly just think it's a phase. And come November, it's just going to go on to the next phase. Now, in order to help Pepe move to the next phase, Matt Fury's enlisted his friends

and fans to reappropriate Pepe from the bad people. And to its credit, when the ADL realized that Matt Fury was himself not personally responsible for this racist meme, they rewrote the entry in their database of hate symbols. They're even helping Matt with his campaign. I wanted to help as well. And so when I saw the stuffed Pepe the Frog doll for sale on the Giant Robot website, I bought one and I gave it to Artaud. Which brings us full circle. After Matilda and I calm down, we write an email to the administrator of the Parents List Serve.

We're very polite. We include articles about the Save Pepe campaign, as well as a link to the Anti-Defamation League's upcoming conference on anti-Semitism, a conference they've invited Matt Fury to present at. We get a response almost immediately. They take the photo and the post down and they apologize. Crisis averted. But then, this morning,

I took Artaud on an informational tour of this music and dance school in our neighborhood. You see, here in New York, if you want to get your kid into daycare, there's an interview process that begins before the kid is even two years old. I know that sounds crazy, but that's just the way it goes here. During our tour, we walk into a classroom filled with singing and dancing children. Artaud is totally into it. He jumps up and down in points.

But one of the teachers gives him this look. And on our way out, she walks up to the director of the school, the woman giving us the tour, and whispers something into her ear. I can feel the color draining from my face. And sure enough, when we walk out of the classroom, the director looks me in the eye and says, there is no place in this school for hate. We walk to the park. Before we get to the playground, I take Pepe and stuff him into my jacket. Arto starts shrieking.

He doesn't stop even when I show him the slide. He just tugs at my jacket. Tears are streaming down his face. A number of moms and dads and nannies are now staring at me. But I can't tell if they're staring because they know that I'm the worst parent in the universe, or if it's because they recognize my son from the internet. Police departments have been using biometrics to identify and convict criminals for quite some time now. But facial recognition is fast becoming the biometric go-to tech tool. This surveillance technology, though, has a number of serious flaws.

Face recognition is not terribly accurate. It's a biometric identification tool. But when compared to fingerprint identification, or maybe even iris identification, it's not accurate at all. Claire Garvey is an associate at the Center on Privacy and Technology at Georgetown Law, and the lead researcher for the Perpetual Lineup, one of the first comprehensive studies of how law enforcement is using, in many cases aggressively using, a technology that doesn't even really work. The FBI actually did look at how accurate their system performs, and they estimated that it performs about 86% accurately. What does that

mean? It means that the system is designed to return a list of possible suspects. So let's say it returns a list of 50 possible suspects. That means six out of seven times that the FBI runs a search, one of those 50 people will be the correct suspect. One out of seven searches will return a list of completely innocent individuals, even though the suspect's actually in the database. That's a pretty low accuracy rate. And when it comes to darker faces, facial recognition technology is even less accurate. Face recognition performs worse on African Americans. This was a finding of a 2012 study that was actually co-authored by one of FBI's experts in face recognition. They found that face recognition performs 5 to10% worse on African American faces.

Well, why is that? There are a couple possible reasons. One is that it's harder to get certain contrast that the face recognition algorithm needs on darker skin. Another reason is that the training data sets, these algorithms need to be trained. And it's possible that if the training data set is predominantly white, the algorithm is going to learn to do very, very well at identifying between white faces, and it will not do well on faces of darker complexions. Pair this simple fact that these algorithms do not perform as well on African Americans with the fact that African Americans are probably overrepresented in the databases themselves. Most systems run on mugshot databases. African Americans are arrested at rates that are far disproportionate to their proportion of the population.

This doesn't necessarily mean they are found guilty of more crimes. This means they're simply arrested, even if they're never charged or the charges against them are dropped. The other thing to keep in mind is that if a face recognition algorithm does not find somebody, if a face recognition algorithm performs worse, this could mean one of two things. One, it could mean that if law enforcement is looking for an African-American suspect, they are actually more likely to not find that suspect, to not get an identification, even if that suspect is in the database. But face recognition systems are not designed to give no for an answer. They are not designed to return a, we couldn't find anybody, we couldn't find your suspect in the database response. They're designed to give a list of

potential candidates. So if they don't find the suspect, if they don't find the correct person, which is more likely to happen with African Americans, that means they're more likely to return a list of completely innocent suspects or completely innocent list of people. So what this means quite simply, at a very basic level, if you're African American and you're in a face recognition database, you are more likely to be returned as a possible suspect, even if you're completely innocent. A technology that could lead to a police officer misidentifying an African American suspect is definitely one you would imagine law enforcement passing on in 2016. But sadly, many of the police departments in the news for killing unarmed African-American civilians are the same police departments rushing to deploy these surveillance systems. But there's yet another problem with this technology, one that's even more feature than bug. Facial recognition systems not only tap into law enforcement databases of mugshots, but in many cases they now tap into

civilian databases of DMV photos as well. These are databases filled with people who have never committed any crime whatsoever. I think it's easy to understand why law enforcement would want to have access to a database that includes perhaps most adults in a given state. When they're looking for witnesses, for example, or maybe a victim. These people might not necessarily be in a mugshot database, and if they're just limited to mugshot databases, they can essentially only look for people who have committed past crimes or have been accused of committing a past crime. So I think there's a huge appeal in the law enforcement community to being able to identify as many people as they can with this mechanism.

The reason why it's troubling is that there are no laws that comprehensively regulate how face recognition should be used by law enforcement, how accurate it needs to be before it's used by law enforcement, and there's no case law that really addresses this either. So what's landed up happening is that we have to rely on the law enforcement agencies themselves, the police departments that have decided to purchase this technology to decide how to use it and whether to place any controls on how it's used, whether to determine what is an acceptable accuracy rate before it's used.

Many of the police departments in Claire Garvey's survey don't have facial recognition policies. We're talking no policy. No policy in regards to when the technology fails to meet standards of accuracy and no policies in regards to when this technology redefines our very relationship with public space. Perhaps that's the whole point. Back in 2011, a number of law enforcement agencies, including the FBI, Homeland Security, and then a number of state agencies, police departments, and departments of motor vehicles, took a very close look at the possible risks of face recognition.

And one of the things they found is that police use of face recognition may very well lead to people starting to curtail free speech, to start behaving differently in public. They said specifically, identity-based surveillance increases the government's power to control individuals' behavior. You have been listening to Benjamin Walker Theory of Everything. This installment is called Targeted. Thank you. You can get a collection of Matt Fury's Boys Club comics from Fantagraphics Books. And you can see a picture of little Pepe the Frog at toe.prx.org. I've been getting some killer surveillance-related stuff from you all out there on ListenerLand. Keep it coming. We still have a long way to go with this one.

Next stop, Russia. Special thanks to Matilde Biot, Cara Oler, Jesse Schaefens, Julie Shapiro, and all the folks at PRX. The Theory of Everything is a proud founding member of Radiotopia, home to the world's best podcasts. Check them all out at radiotopia.fm. And shout out to the Knight Foundation and MailChimp, our

original sponsors, along with listeners like you. Radiotopia from PRX