[MUSIC UP & UNDER]
BROOKE GLADSTONE: From WNYC in New York, this is On the Media. I’m Brooke Gladstone.
BOB GARFIELD: And I’m Bob Garfield. Late last week, the New York Times and The Observer of London released the results of a months-long investigation into the political data firm, Cambridge Analytica. The story was delicious and the cast of characters straight out of a Bond movie. There was the wealthy arch-conservative puppet master.
MALE NARRATOR: They had the money from billionaire Republican backer Robert Mercer and his daughter Rebecca.
BOB GARFIELD: His man on the inside.
MALE CORRESPONDENT: Steve Bannon oversaw Cambridge Analytica’s collection of Facebook data.
BOB GARFIELD: The arrogant operative caught red handed in a sting.
MALE CORRESPONDENT: The videos show a reporter posing as a prospective client of Cambridge Analytica and executives bragging about using bribes and prostitutes to entrap politicians.
BOB GARFIELD: The guilt-stricken whistleblower.
MAN: I was naïve. I made a, a big mistake, and that’s why I’m talking to you.
BROOKE GLADSTONE: And we were naïve too, or at least the 50 million of us Facebook users who were touched by the Cambridge Analytica scandal when our personal information was divulged by Facebook without our consent and maybe weaponized against us during the campaign. Many talking heads saw these revelations as the answer to the question that had long plagued them, namely, how did Donald Trump defy all expectations and win the presidency? It was the data, stupid!
MALE CORRESPONDENT: Did Cambridge Analytica win the election for the president and were laws broken along the way?
MALE CORRESPONDENT: A psychological warfare weapon --
MALE CORRESPONDENT: Yeah.
MALE CORRESPONDENT: -- something you say Steve Bannon wanted, the ability to play with the psychology of an entire country.
FEMALE CORRESPONDENT: It’ what Cambridge Analytica does, it kind of identifies the kind of voters who might be vulnerable to persuasion, some might say manipulation.
BOB GARFIELD: Antonio García Martínez used to work at Facebook as a product manager heading its ad-targeting business. He’s now a contributing writer for Wired, where he recently reported on the age-old marketing tool that Cambridge Analytica claims to have used to direct Trump campaign ads to susceptible voters: psychographics.
ANTONIO GARCÍA MARTINEZ: Yeah. One of the funniest things about the coverage of the story is that those who know the least about ads are the most convinced of the supernatural powers of advertising and what Cambridge Analytica did and those like me or some of my colleagues who most know ads are the ones who are most skeptical of their claims. [LAUGHS]
BOB GARFIELD: On Facebook, drawing psychographic profiles has involved quiz apps that supposedly tell you what kind of personality you have or which Beatle or Harry Potter character you most resemble. But it all boiled down to the same thing.
ANTONIO GARCÍA MARTÍNEZ: You know, this whole psychographic thing is kind of a riff on demographics but instead of being age and gender and geography it refers to a certain psychological state of being that presumably the target of your ad is somehow in, which gives that person an affinity for your product or, in this case, you know, political philosophy.
BOB GARFIELD: Antonio Garcia Martinez is a contributor to Wired, a former Facebook product manager and author of Chaos Monkeys: Obscene Fortune and Random Failure in Silicon Valley.
ANTONIO GARCÍA MARTÍNEZ: The model that they used is what’s called the Big Five personality model. You can actually look it up on Wikipedia. OCEAN is, is a handy mnemonic for five dimensions that you project a human personality on, and, and those are openness, conscientiousness, extraversion, agreeableness and neuroticism, and the idea here being that if you’re neurotic then you’re either going to feel strongly about guns or immigration or as the case may be. And so, by breaking somebody down into one of these five categories you can figure out whether they’d react positively or negatively to a certain political message. That's the thought, at least.
BOB GARFIELD: The four marketers had actual real-time consumer behavior data in hand. They segmented consumers based on general demographic similarities, neighborhood and media habits and so on. It’s not complete baloney, is it?
ANTONIO GARCÍA MARTÍNEZ: Yeah, I, I don’t think it’s completely wrong. Again, if you refer to one of the classic psychographic models, it’s called the PRIZM, PRIZM with a Z, segments, now owned by Nielsen. And you’ll find there a mythology of 66 types of American consumer. And yeah, if you read it you’ll recognize yourself in some of them, in the same way that you kind of read your horoscope and think it's true. But it has such gems like shotguns and pickups, for example, as the sort of white working-class rural voter. There's kids in cul-de-sacs as a sort of middle-class suburban family.
The idea behind this model is, and, in fact, what happened with Cambridge Analytica is, you know, one of their researchers in a slightly freelance capacity created a Facebook app. Something like 300,000 Facebook users had to opt in via a Facebook platform and then, based on the questions in that quiz, that person tried to figure out what sort of Facebook data is associated with, for example, neuroticism or openness or conscientiousness or the various dimensions of this model. And then armed with that you sort of guesstimate, well, what sort of personality profile, for example, feels strongly about immigration or gun control or Trump or, as the case may be, and then turn around and try to target that person on Facebook via the ad system.
BOB GARFIELD: You would think that joining even these kind of rough-hewn psychographic profiles with the other data sets that Facebook has would really, as Cambridge Analytica claims, really sharpen that lo-res image into something deeply personal and illuminating.
ALEXANDER NIX: Well, at Cambridge we’ve rolled out a long-form quantitative instrument to probe the underlying traits that inform personality.
ANTONIO GARCÍA MARTÍNEZ: Yeah, I mean, it’s funny. I think people tend to overestimate how much actual actionable targetable data they have. The only thing you can actually target on Facebook are things like age and gender, which are admittedly important but are still kind of, you know, basic geography, obviously, and then what's known as interests. It’s mostly basically what pages you’ve liked. And while that might be of interest to some marketers, the fact that you've liked, say, BMW’s page on Facebook doesn't mean you're actually in the market for a BMW, right, which is the discovery we made in 2011 when we built a lot of these technologies when we realized that they actually weren’t working for most marketers.
Even assuming you do a poll on someone, how do you actually find a neurotic person on Facebook, right? I want to find someone who wants to buy a BMW or I, I want to find someone who’s neurotic because a neurotic person is anti-immigration, how do you then go around and find that person on Facebook? The challenge is still the same. Of course, if, if you have voter rolls, if, if you have other, you know, either third-party or first-party data, you can bring that to Facebook. But Facebook’s data itself, I, I think people tend to overestimate how much value there actually is in their targeting data.
BOB GARFIELD: I guess I don’t have to take your word for that. This is Aleksandr Kogan, the PhD researcher at the heart of this Cambridge Analytica story, the guy who [LAUGHS] was accused of misappropriating the Facebook data, speaking to the BBC this week.
ALEKSANDR KOGAN: In practice, my best guess is that we were six times more likely to get everything wrong about a person as we were to get everything right about a person. Like, I personally don't think micro-targeting is an effective way to use such data sets.
QUESTIONER: You don’t think it could sway an election.
ALEKSANDR KOGAN: I think it could have only hurt the campaign.
QUESTIONER: The Trump campaign, you mean?
ALEKSANDR KOGAN: Any campaign. I think what Cambridge Analytica has tried to sell is magic and it made claims that this is incredibly accurate and it tells you everything there is to tell about you. But I think the reality is it’s not that. If you sit down and you really work with the statistics, those claims quickly fall apart.
ANTONIO GARCÍA MARTÍNEZ: I mean, obviously, Kogan might be kind of backpedaling there, given all the attention his, his work has, has garnered, but the reality is, particularly in the political space, it’s often very difficult to figure out if something actually works, right? When you’re selling shoes online it’s relatively easy to show you an ad for shoes and then if you go and buy those shoes, to actually connect those two things together and actually look at the return on your investment in advertising. No campaign, no matter how well funded or sophisticated, can actually really tell if their campaign drove voters to the ballot box and made them vote or not.
BOB GARFIELD: So the last few days you’ve been making the media rounds explaining why the secret sauce wasn't either secret or particularly saucy but, you know, at least a year ago the New York Times ran a story saying that very same thing. We also learned that the Ted Cruz presidential campaign ditched the models because they were unreliable. So why do you suppose that you were at pains to go through this [LAUGHS], yet again?
ANTONIO GARCÍA MARTÍNEZ: To be honest, people would rather believe that dark mysterious forces using some diabolical technology made evil happen, rather than, you know, a more mundane story. You know, I actually made a sort of joke tweet about that, you know, what are we gonna believe, that the center-left parties in the US and the UK were pathetically complacent in reacting to global capitalism and spurred a populist revolt that caused Trump and Brexit to happen? Or, or do we believe --
-- the protocols of the elders of Facebook story, where somehow ad targeting came out of nowhere and elected a -- you know, an autocrat? And everyone’s just like two, two all the way, right, two is the one we’re, we’re going to go with. I, I think there’s other forces here more than the, the reality of it that’s causing all this media attention.
BOB GARFIELD: Okay, that’s one of the week’s journalistic misdirections, the red herring. Part two is what we’ve missed about Facebook that the Trump campaign [LAUGHS] absolutely did exploit to run what turned out to be an astonishingly efficient social media campaign. It has to do with the Facebook algorithm, and this is really kind of remarkable, the actual cash value of being inflammatory. Please explain, for starters, the auction bidding system.
ANTONIO GARCÍA MARTÍNEZ: Right, this is a key part of how this whole media-scape works, that the public doesn’t understand and, frankly, should. So how does the auction process work? I, I think the best way might be explaining via metaphor with the regular feed, which I think most people understand. In News Feed, you’re, obviously, not seeing all the stories and posts and shares that all your friends have put out there. Facebook actually ranks them for you because otherwise there would be a torrent of information that you can’t parse. And so, Facebook uses a model, an, an algorithm that calculates how likely you are to engage with that piece of content. And by engage, we mean share it yourself, click on it, comment on it; anything you interact with, they will tend to show you more of.
What I think people don’t understand is that that same dynamic plays out on the advertising side. In other words, Facebook also calculates, based on what it knows about you, which ads you’re likely to interact with. Ads on Facebook are, are not just like ads in a, in a newspaper or on TV. They often appear as a regular Facebook story, right? So you can share them, you can comment on them. It, it appears like a regular post inside your feed. What I think people don’t realize is that that engagement metric affects whether that ad’s going to show up in your feed and also how much that ad ends up costing to the advertiser.
BOB GARFIELD: And, weirdly, it costs the advertiser less if people click on it more.
ANTONIO GARCÍA MARTÍNEZ: Exactly because at the end of the day Facebook is optimizing for what in the trade is called CPM, that’s cost per mil, cost per thousand ads. After all is said and done, a politician or a product or whatever that embraces a rhetorical strategy that, for, for better or worse, causes lots of engagement, they will either get more media for the same amount of money or get the same media for less money than they otherwise would if they weren't sort of clickbait-y.
BOB GARFIELD: So, if I understand this right, the business model of valuing engagement actually offers an advantage to incendiary ads and so-called “fake news,” right?
ANTONIO GARCÍA MARTÍNEZ: Yes, to the extent that negative rhetoric drives more engagement than, say, more nuanced or tempered rhetoric, then, indeed, it would fare better on the Facebook ad system, yes.
BOB GARFIELD: Okay, that’s one thing that the Trump campaign exploited about Facebook. Another is something that you were in charge of, a feature called Custom Audiences and later something called Lookalike Audiences. Explain how they leveraged their investment with them.
ANTONIO GARCÍA MARTÍNEZ: Right. Custom Audiences are sort of a, a fancy marketing name for Facebook joining to the outside world of data. I mean, a simple example of this that all of your listeners have probably had, if you go browse the internet and you go shop for something or whatever and you go back to Facebook and you see that same pair of shoes or that same handbag you were shopping for appear inside Facebook, that's what I mean. There’s this outside world of data that historically didn’t enter Facebook but, starting in 2012, there was a couple of different technologies -- one of them was Custom Audiences -- that allowed that to happen. Effectively what happens is the advertiser quite literally uploads a spreadsheet with names, email addresses, phone numbers, etc., and then Facebook tries to find those users inside Facebook. It’s probably the most common tool that savvy advertisers with lots of data use on Facebook.
So what would that mean in the political context? Well, voter files, for example, that I understand have things like names and addresses could be uploaded to Facebook. Donors who actually donated money to Trump, all those people, like at an individual granular level, would be targetable on Facebook. On top of that comes the second product you mentioned, which is Lookalike Audiences, and this addresses a really deep need for most advertisers. Most advertisers have some set of customers or voters that they know are into their brand or into their products. The question is, how do I find more of them? And so, after I’ve uploaded a Custom Audience and, again, that's based on my email list, people who’ve come to my site. Facebook then finds people who are like that.
And here is where the true value of Facebook’s data comes in. It turns out, you know, what they care about is not your photos or your conversations with your coworkers or your partner. What they really care about and what they use their data for is figuring out other people who are like you. That’s a key thing. I mean, the reality is if you’ve got five friends, more than likely they have the same consumption patterns or the same political views that you do. And via the magic of Facebook data, you can expand an initial targeting list into a yet bigger one.
BOB GARFIELD: Okay, so is this just marketers and politicians using finer and finer targeting, as they’d be expected to or is all of this data overlaying, the data joining, whether by businesses or candidates or governments, putting us all at a high risk of utter loss of privacy?
ANTONIO GARCÍA MARTÍNEZ: I think you put your finger on it. You know, even I, former ad tech mercenary that I am or was, I’m willing to concede that I think politics is definitely special, right? Selling you on Brexit is different than selling you on a pair of shoes. And if the government sees fit to regulate political advertising differently than it does regular advertising, that’s not the craziest thing in the world.
And so, if what comes out of this whole Cambridge blowup is that the rules that typically apply to conventional media and political advertising also apply to Facebook, I think that would be just fine. But, you know, once you’ve taken this genie out of the bottle or this toothpaste out of the tube, it’s hard to put back in. And so, I, I don’t see the politicians who, by the way, are themselves the ones who would cook up the regulation, regulating the very thing that put them in power.
[MUSIC UP & UNDER]
BOB GARFIELD: Antonio, thank you very much.
ANTONIO GARCÍA MARTÍNEZ: No, thanks for trying to clarify this. It’s amazing how the, the press’s coverage of this has gone in so many directions.
BOB GARFIELD: Antonio García Martínez is a contributor to Wired, a former Facebook product manager and author of Chaos Monkeys: Obscene Fortune and Random Failure in Silicon Valley.