Benjamen Walker's Theory of Everything - The Rainbows of Inevitability

Episode Date: February 23, 2017

We take a tour of the sprawl with Metahaven to learn about the propaganda of propaganda and we travel beyond the Facebook wall to learn the real truth about targeted advertising. Plus Project Madison Valleywood!  

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 The Rainbows of Inevitability. When I began this surveillance miniseries, I put out a public call to you, dear listener, for things that I should look into. And a lot of the folks you wrote me about have made it onto the show. In fact, some of you who wrote me have made it onto the show. The craziest email I got, though, was from a guy that we'll call Jimmy. Jimmy isn't his real name. I was going to change his voice.

Yeah, but now that Trump is president, I don't think we have to worry about that. Jimmy wrote that he had to talk to me about something that was going on at Google, something surveillance-related, something that had huge implications for the national security of our nation. Jimmy didn't say much in the email, just that he had lost his job over this and would be coming to New York. So we met in a hotel room in Midtown. That's where I recorded this. I'm still not sure what to make of it, so I'm just play you the whole thing so it all started at south by and what year was this 2016 i'd been wanting to go for years i just never got around to it

but when i found out about this secret project madison valleywood meeting and that jason had cut me out i was like i'm going yeah you're just gonna have to slow down. Who's Jason? Why did he cut you out? And what's Project Madison Valleywood? Wait, I thought I wrote you about this. No. Really? Past few weeks have been kind of hectic ever since I resigned. But all right. A bunch of people who work in tech got together with some people who work in advertising and some people who make movies and decided to help the government. How? If you really want to cut off the supply of potential terrorists, then you've got to stop online radicalization. Why is it called Project Madison Valleywood though?

Madison Avenue, Silicon Valley, Hollywood. Ah. But I didn't know about this meeting because of Jason. He made the call that I wasn't need to know. So, you and this Jason guy work together at Google? Jason comes from AdWords. He's not a data scientist.

He's an ad salesman. I'm AdEye. AdEye. You know what the special forces are, right? Like SEAL Team 6? AdEye is Google SEAL Team 6. We use machine learning to create the next generation of ad products. But why is it called AdEye? Advertising plus artificial intelligence? AdEeye?

And what exactly do you do? Okay, with traditional targeted advertising, you can deliver customized ads to highly specific demographics. Say, men aged 18 to 22 with siblings who own a car and make at least 100k a year. And intuitively, this makes sense to advertisers. If you're trying to sell luxury watches, you'd prefer to advertise to an audience that can afford to purchase the watches you sell. But the limitations of this are also pretty clear. The targeting is only as valuable as your ability to describe your customers according to demographic descriptors. What I'm interested in is machine learning.

Instead of creating a set of rules for a machine to carry out, say, show this advertisement to men who make more than 100k a year, machine learning is about training a model to be goal-oriented. Like, show this advertisement to people who will actually purchase the product. This I think I'm actually following, but it sounds terrifying. Yeah, and that's what Obama said when I introduced myself at South By. Wait, so you got to go to the meeting, and Obama was there? Oh yeah, and you should have seen Jason's face when I said,

if we really want to get serious about taking out ISIS, I'm your man. Is this issue you have with this Jason guy relevant to the story? Look, Jason's a nothing, a loser. But this group, it was filled with Jasons. I'm talking most of the Silicon Valley people. Losers with no up left to fail. I mean, there were like 10 dudes from LinkedIn who all just wanted to get their James Bond on. Can you just talk about what actually

went down at this meeting? You know, this isn't a secret. Really? Even the ISIS stuff you emailed me about? Yeah. I mean, Vice even wrote about that. Well, if you don't mind, could you just explain to me how or what the ISIS targeting thing is and how it works? Okay, Google Ideas or Jigsaw, they kept changing their name, used ad targeting software to identify Muslim Americans, immigrants, and second generation Muslims who had watched ISIS YouTube videos, and then they served these users videos that had anti-ISIS messages. But I'm not here to talk about what Jason did. But where does Google get the anti-ISIS videos?

The State Department. So it's propaganda? That's the dumbest idea ever, and why would Google want that to go public? When your mission is to connect people with the best information possible, then there's nothing wrong. In fact, it's noble of you to serve people who are potential victims of propaganda videos that counteract said propaganda. What's dumb is going public and saying that a 2% click-through rate means success. But what's to stop the government from just saying, okay Google, hand over the list of every

single person out there who's watched one of these ISIS videos? Okay, now you need to explain something to me. Why is it that every journalist who does a story on this project Asks the same ridiculous question? Why is it ridiculous though? Just because you watch an ISIS video Does not make you a terrorist I mean the government, especially now Could really cause problems for people

Who just watch the video Okay, first of all The NSA already has this list. Really? You have it all backwards. It's not what is Google handing over to the NSA. Your question should be, what is the NSA handing over to Google? Have you heard of the Google paper building high-level features using large-scale unsupervised

learning? No. It's also been referred to as the CAP paper Building High-Level Features Using Large-Scale Unsupervised Learning? No. It's also been referred to as the CAT paper. Definitely haven't read that one. Well, it's probably one of the most important advances to come out of Google recently. A lot of people had successfully trained algorithms to identify objects in images using labeled data sets, like a collection of images tagged with whether or not they contain cats. But we created an algorithm that was able to learn from an unmarked dataset

to identify that cats were a thing. So you worked on that? Yeah, I gave them some input, but I knew my work at AdEye would have more impact. I was training algorithms to identify consumer clusters, customers, for ads. But the thing is, I needed more data. The brain team, they had it easy. They used still frames from YouTube. And how hard was that?

I needed much more sophisticated user data. Which is why I went to South By. I needed to convince the president to give me access to NSA data sets. Why? Well, obviously NSA data sets are data sets that Google could never get their hands on.

Why not? The law. And did you get access to them? Well, Google did, but Jigsaw shut me out. Jason got to use the data for his lame ISIS ad targeting experiments and I didn't get to do shit. So everyone ended up losing.

I lost. Google definitely lost. I mean, really, America lost. I'm sorry, but I'm having a hard time following. Look, I'm going to break it down for you. There's nothing amazing about targeting men who own a car and make $100K a year with an ad for a fancy watch. If we're waiting until a user has already expressed intent or targeting specific demographics, we're too late. It's a competition.

By the time a user understands they want to buy something, they should already know where they want to buy it from. And that's where Adai would come in. Yeah, I'm just so confused because when you wrote to me, you made it sound like you had this really big story. And you made it sound like you were a whistleblower. But, I mean, it sounds like you're just talking about something that never even got built. Exactly. I never got to build the product. No, you're not getting it.

If America wants to win the war on terror, we need to be leveraging machine learning. We need to identify that lone wolf terrorist before he even realizes that he is a lone wolf terrorist. Even with NSA data, Jason will never be able to do that. Only I, Adai, can accomplish this. So basically, you're mad because you never got to build your supervillain targeting

system. Yeah. And that is why I'm going public. I used to think that propaganda was about persuading people. Now it doesn't seem to be about that. It's just about deconstructing the other side, disrupting Western narratives. There's a steady stream of disinformation whose purpose seems to be to sort of undermine

the very idea that truth is provable. That's author Peter Pomerantsev speaking in a clip from The Sprawl, a film-slash-museum installation from Metahaven, the research-focused design studio founded by Amsterdam-based Winkke Kruk and Daniel van der Velden. Last fall, when I was in Amsterdam, I had dinner with Daniel and Winkekroek and Daniel van der Velden. Last fall, when I was in Amsterdam, I had dinner with Daniel and Winka. These two are deep in the maze of fake news, conspiracy theories, panoptic surveillance, and other propagandistic strategies. The sprawl is a far-reaching meditation on the propaganda

of propaganda. For them, it all started with the downing of Malaysian Airlines Flight 17. Not the Malaysian Airlines flight that went missing. That's Flight 360. Flight 17 was the one that was shot down. Flight MH17 was downed over eastern Ukraine. It was a flight from Amsterdam to Kuala Lumpur. So there was a lot of Dutch people on board. So the event had a big coverage in Dutch media.

It was kind of inescapable. But when you went on YouTube and online, there was so much material being produced, images going back and forth about, you know, fighters recording themselves, soldiers taking selfies, which then had military ammunition in the background by accident, which was then being publicly denied that it even existed by Ukraine or even the Russian state. So there was this constant kind of perpetual production of images, of statements, of stories out there surrounding what happened around this plane that became a really rich propaganda war. In the sprawl, the whole world is at propaganda war.

And Daniel and Vinca take us to numerous fronts. Russia, the Middle East, Europe, the Americas. They show us how propagandistic strategies, like alternative facts, have been used to disrupt elections, news cycles, even reality. But for Metahaven, it's propaganda itself that has been disrupted, by technology, or the planetary-scale infrastructure of information, as they call it. The idea that truth itself is not provable is not a new idea per se as well, but it's something that becomes much easier to demonstrate with the planetary scale infrastructure of information that exists today.

The multitude of different voices and the multitude of different possibilities and the multitude of different ways of reading certain information is actually something you can explore to create alternate realities. You know, to create realities in which a particular version of events as you want to see them is viable. In a sense, the sprawl is very much about what happens when state actors in combination with technology and with the internet kind of create a kind of overlap of these different realities or these different truths. And a very simple example of how that plays out is Google Maps. For instance, when you go to Google Maps from Ukraine, the specific area around the border is listed as being part of

Ukrainian territory. If you go to Google from Russia and you go to the exact same location and the exact same border, it is actually listed as being Russian territory. In the sprawl, multiple things can be valid at the same time in the overlays. To call these overlays fake is a mistake, though. If we hope to understand the true nature of propaganda, we must first accept that they are real. Propaganda is not a cold informational regime that's just laid over unknowing people.

In many cases, it's a lived reality. As Daniel said, thanks to planetary-scale information infrastructure, today, anyone can do the alternative fact thing. Media wars have entered new territory. But if you want to create a first-class alternative reality, well, you still need a TV channel. This television station, RT, has some kind of security threat. Since when is holding and broadcasting...

One of the central subjects of the sprawl is Russia Today, or RT. I've seen lots of your reports, and in not one report will you find you questioning the United States... RT funds a lot of independent journalists to cover stories mainstream American media often dismisses or ignores, like Occupy Wall Street and Standing Rock. Today we have ISIS. But RT also pushes the Russian agenda. On the one hand, you could say it's Russian state television, but on the other hand, it's also giving a kind of counter voice

to what we consider to be kind of standard Western media. Certainly at Russia Today, we always try to show both sides of the story. Do we show more of a Russian perspective? Of course we do, because that's the perspective that's being sidelined. But it's an absurd question. The idea is that the whole concept of legal or judicial truth, as we know it today, and as it's being expressed and given a platform in different media,

is itself part of the Western perspective that needs to be countered. But RT isn't the only TV channel broadcasting in the sprawl. Medihaven also likes to watch CNN. That is a bit of a chilling analogy, that it's the ISIS of infectious agents. In October 2014, an article appeared in the New York Times that compared Ebola to ISIS. It was picked up by CNN. If you think about Ebola as an agent...

And they had a text running in the screen saying, Ebola, the ISIS of infectious agents, question mark. Infects people and it kills people and in that way it amplified this sort of quite hysterical and absurd way of looking at things the denmark of international terrorism we have this way of speaking about things that is kind of the midwest of southern california commercial capitalist form of hyperbole. The Che Guevara of Uber. And this is what CNN did. The ISIS of gluten-free donuts.

Obviously, commercial hyperbole can be a lot of fun. Until you realize just how persuasive and pervasive it actually is. Metahaven shows us that it is now everywhere. One thing that struck us while making the sprawl is almost the inevitability of building your own truth, your own version of truth. What Metahaven wants us to see is that the sprawl is not just growing, it's also accelerating

at speeds that seem to countermand any and all attempts to put the real and the fake and the ads and the news back into their old boxes. Disinformation and confusion have become propagandistic strategies, strategies that have become extremely effective. But with the Internet, stories get accelerated and also the reaction to stories also accelerates. So you get this kind of propagandistic disinformation and confusion strategies that constantly are reacting to each other, but they're all trying to confuse each other at the same time, which results in, you know, sometimes extraordinary narratives that kind of completely lead you to something that's not even true or false, but that's just a kind of completely new reality. Yeah. For now, it's pretty much all we can do.

Marvel. After the break, what Facebook is really up to. Are you sick and tired of objective reality? Yes, we are! I know, it's exhausting, especially now. I will build a great, great wall. Nobody builds walls better than me, believe me. With the new Oculus Rift, you can curate your own reality.

With Oculus Rift, nothing is curate your own reality. With Oculus Rift, nothing is impossible. Everything is permitted. You can decide on your own what is fake and what is real. Get one now and retreat into a world all of your own making. Go to oculus.com slash rift and use the offer code theory and get 25% off your first order of next generation virtual reality. That's oculus.com slash rift. Okay, so I've had a lot of conversations about surveillance and targeted advertising for this series. And every single one of these conversations ends the same way. No matter who I'm talking to,

we hit a wall, the Facebook wall. Most of the answers to the questions I have about targeted advertising lie beyond this wall. Like how much data is being collected? How is it being used? Who gets to use it? Who is doing the targeting? And how exactly does that work? And is there any evidence that all this targeting even does work? It's frustrating, especially now that I know that what I really want to know is exactly what I don't know that I know. So I asked a professional for help to take us over the wall. I basically do like investigations, but using code instead of like traditional tools. Surya Matu is an artist and a researcher currently working as a data reporter for Gizmodo. Recently, he did some work at ProPublica, where he created a number of plugins that show how Facebook uses profile data to serve its users' targeted ads.

I'm not on Facebook. So when he came over, we opened up TOE producer Anders' profile for analysis. Let's do it. All right, use me, baby. All right, so we're basically opening his Facebook account and going to his ad preferences page. And now we're on that page and the... Wait, here's a quick question. So I look...

Yeah, wait, what does this mean? Hulu, Kickstarter, Bernie Sanders? So as you can see up here, you have 237 categories that Facebook is targeting ads to you for. These are some of the ones that they are. So Hulu, so it also tells you why you have this preference. So Hulu, you have this preference because you installed the app on your phone. Oh, God.

So Kickstarter says you have this preference because you clicked on an ad related to Kickstarter. Bernie Sanders because you like to page. Or you have Donald Trump on here because you clicked on an ad related to Donald Trump. That's terrible research. Right? So if you think about what that inference is, if someone likes Donald Trump, it means they clicked on something related to Donald Trump.

Whether it was by accident or actually what they wanted to do. So this is the good stuff, right? So what we've shown you so far is things you've actually clicked on. But then they make these amazing inferences of what you're doing. So one of his interests that I am reading right now is african-american culture and it says you have this preference because you clicked on an ad related to african-american

culture so jay-z is basically you only click on a beyonce ad and now you're interested in african-american culture in the last six months or so, the number of categories that they're showing me has exponentially increased. So what that leads me to believe is that they're doing this algorithmically. So I'll give you an example. There's a teacher at NYU

who has an online coding class called Coding Rainbow. And I liked his YouTube page because I wanted to support it. When I looked at my ad categories, it's on Facebook, it said one of my interests was rainbows. Right? And it says because you have clicked on a page and there's nothing else in my life that I have done that would obviously relate to rainbows. So that that to me leads me to make this inference that someone split that word into two parts and said, oh, this person must be interested in coding and rainbows.

Because, again, they don't really know, and this is not like a dig on Facebook. It's that no one really cares how accurate it is as long as they can put you in more buckets. Right? So it's like, you have to think about the person who's going to buy these ads. They aren't thinking about

how those categories look they created. They're just excited that they exist. There are so many people that love rainbows. buy these ads, they aren't thinking about how those categories look. They create it. They're just excited that they exist. There are so many people that love rainbows. Right. And just think about the meeting, right? Like you go, yeah, you go in and you're like, I have your group. I know who to sell this, your product to, and I figured it out. I've used the internet and I got it.

Right. You couldn't do that earlier. Even though platforms like Facebook and Google have an enormous amount of data, it's never enough. They want more. So Facebook doesn't need to buy data to keep you on. They need to buy data to make sure that they can sell ads to people that are more targeted. What more means is more places, right? So it's not actually just, it's not waiting for you to are more targeted. What more means is more places, right? So it's not actually just,

it's not waiting for you to do more stuff. It's trying to pick up other signals of fingerprints and digital footprints that you've left behind. Your shopping habits, what kind of places you shop at, stuff like that. Those are things that they can pick up

off credit cards and other data sources like that. They also want to know where you hang out. They use Instagram. They already own Instagram. So they pretty much know where you go if you have location services turned on on Instagram. Keep in mind, Facebook owns Instagram and Google owns Google Maps. Both use location-based data. So these platforms already know where you are and where you've been. More data helps

them fill in the blanks. Not just what you've seen or what you've been doing. What they want to know is who you really are. Right. So you have companies like Axiom and Datalogics who are like well-known big vendors of data tracking. And what they actually do is they, like along with the stuff they collect from cookies on your phones and laptops to figure out what you do online, they've also been going out and collecting your information from offline sources like the DMV and stuff like that to really build as big of a profile as they can on you.

So these data companies like Axiom, Datalogic, can I see what they have on them? They have such tight clauses on what they're sharing that they never, you never see what's happening. I spent a long time trying to figure out what experience, what Axiom thinks about me, what my profile is. But because they always claim that it's anonymized and bundled, I can never actually get down to me. And that's really hard. It makes it really problematic if I'm, as a journalist, trying to write a story about

what my online digital profile is. The best I can do right now is seeing what Facebook shows me and Google shows me because companies actually have no obligation at the moment to give me that information in a meaningful way. The information that data brokers like Experian and Datalogix collect ends up on the market. They sell it directly to Facebook or to companies who bundle it with other user data, who then sell that to Facebook. And sometimes they do the bundling themselves. So bundling basically is the accumulation of different data sets into one package to kind of build your profile. So you can think of it as if you, I don't know, if you go to a gym and you

go to the cinema and you go to the same ones a lot, someone collects those two pieces of your profile, even though they're like very different parts of your life, to figure out that, ah, this is someone who goes to this type of gym and this type of cinema hall or whatever the thing is. But basically, different points of data can be collected from different spaces. And the real value for these data companies comes in when they can make a more comprehensive profile of you. And the way they do that is they collect your credit card data.

So there's some things which are like aggregate data where they can't get individuals' information. But there are others which do get individuals' information. So sometimes your information comes under a probabilistic model that if you are a 45-year-old white man living in, I don't know, New York, you likely will shop at Whole Foods. But it might not be about you directly, whereas other times it will be your data. There are many actors who store different parts of our data, and then there are other actors who join that data. This is where things get complicated. Facebook doesn't want its users to see the data they are bundling or purchasing pre-bundled. It's up to researchers like Surya to discover their own ways to peek at what's going on beyond the wall. Well, I'm just going to show you a couple of ad categories that Facebook lets you buy ads on.

Where are you getting this from? So I pulled this information from Facebook from the ad selling thing. I'm going to say this, but you might have to edit this out. But basically what I did was I wrote a script where when you go to the create ad section that it basically made every combination of letters possible and made a request to that thing to see what ads they're selling. How many categories? So right now I think I have like 32,000 but I could get more probably. So here's one here's a good one here's one of my favorite ones actually. Here's two I want to say two. One is people in households that are heavy buyers of salad dressings.

Right? This comes from DataLogix. The source is loyalty card and transaction level household purchase data with multi-channel coverage across all product categories. Details. People who spend three times or more than the national average based on volume unit spend in the salad dressings category and have actively purchased over the last 12 months. So these are people who buy a lot of salad dressing. In other words... This is from loyalty cards, like at a Safeway or whatever.

So if you're not in the loyalty program, they're not getting your data here. So you don't get that discount because you're not giving your data. Here's another one. This is actually my personal favorite, and it's people whose activities strongly suggest they're a trendy mom. This tidbit comes from DataLogix.

The source of this is U.S. consumer data on where consumers shop, how they shop, what products and brands they purchase, the publications they read, and their demographic and psychographic attributes, details. Interests and purchases include children's boutique clothing, contemporary children's furniture, parenting, cooking, and home decor publications. So when we're talking about bundling, this is the level of granularity we're saying. We're not saying that, like, do I go to Netflix or not? It's like, do I use mayonnaise or not?

It's extremely disconcerting to imagine that platforms like Facebook have collected these dossiers on us, detailing our every want and desire. But what Surya wants us to understand is that in many cases, these dossiers are built on inferences and assumptions that are wrong. But this too is disconcerting, because all this raw data, these ad categories, these personal profiles, they are the building blocks from which

a new world, our brave new brand new world is being built. The individual sort of doesn't matter anymore, right? Like the way these systems are set up, you just need to be put into a bucket. They don't actually care what you do or what I do they just want to be able to say that they have an ad category of people who listen to podcasts and we have a hundred thousand people on that and our goal is to make it 150,000 right so in that sense it's we have to like kind of decouple our personal identity to this the real danger is that someone is selling this as though it's fact and someone is buying this thinking that they're making a factual decision and once that comes into your system

you're basically trusting rainbows You have been listening to Benjamin Walker's Theory of Everything. This installment is called The Rainbows of Inevitability. This episode was produced by myself, Benjamin Walker, and Andrew Calloway. It featured Surya Motu, Medihaven's Daniel, Vander Velden and Vinka Kruk, and Jimmy. Special thanks to Julie Bluzet, Jesse Chapins, Cara Oler, Mathilde Biot, and James Burns. Also, thanks to all the folks at PRX HQ, like publicity manager Maggie Taylor.

The Theory of Everything is a proud member of Radiotopia. Radiotopia is home to some of the world's best podcasts, like 99% Invisible, Criminal, The Illusionist. You can find a link to all of them at radiotopia.fm. you Radiotopia from PRX.