Mindf*ck: Cambridge Analytica and the Plot to Break America

by Christopher Wylie

The story of Cambridge Analytica shows how our identities and behavior have become commodities in the high-stakes data trade. The companies that control the flow of information are among the most powerful in the world; the algorithms they’ve designed in secret are shaping minds in the United States and elsewhere in ways previously unimaginable. No matter what issue you care about most—gun violence, immigration, free speech, religious freedom—you can’t escape Silicon Valley, the new epicenter of America’s crisis of perception. My work with Cambridge Analytica exposed the dark side of tech innovation. We innovated. The alt-right innovated. Russia innovated. And Facebook, that same site where you share your party invites and baby pictures, allowed those innovations to be unleashed.

Everything we do, he said, is predicated on understanding exactly who we need to talk to, and on which issues.

Once I understood the basics, I switched from petals to people. VAN was filled with information on age, gender, income, race, homeownership—even magazine subscriptions and airline miles. With the right data inputs, you could start to predict whether people would vote for Democrats or Republicans. You could identify and isolate the issues that were likely to be most important to them. You could begin to craft messages that would have a greater chance of swaying their opinions.

BUT BY DIRECTLY COMMUNICATING select messages to select voters, the microtargeting of the Obama campaign had started a journey toward the privatization of public discourse in America. Although direct mail had long been part of American campaigns, data-driven microtargeting allowed campaigns to match a myriad of granular narratives to granular universes of voters—your neighbor might receive a wholly different message than you did, with neither of you being the wiser.

When campaigns were conducted in private, the scrutiny of debate and publicity could be avoided. The town square, the very foundation of American democracy, was incrementally being replaced by online ad networks. And without any scrutiny, campaign messages no longer even had to look like campaign messages. Social media created a new environment where campaigns could now appear, as Obama’s campaign piloted, as if your friend was sending you a message, without your realizing the source or calculated intent of that contact. A campaign could look like a news site, university, or public agency.

Tech companies began making money from their ability to map out and organize information. At the core of this model was an essential asymmetry in knowledge—the machines knew a lot about our behavior, but we knew very little about theirs. In a trade-off for convenience, these companies offered people information services in exchange for more information—data. The data has become more and more valuable, with Facebook making on average $30 from each of its 170 million American users. At the same time, we have fallen for the idea that these services are “free.” In reality, we pay with our data into a business model of extracting human attention.

More data led to more profits, and so design patterns were implemented to encourage users to share more and more about themselves. Platforms started to mimic casinos, with innovations like the infinite scroll and addictive features aimed at the brain’s reward systems. Services such as Gmail began trawling through our correspondence in a way that would land a traditional postal courier in jail. Live geo-tracking, once reserved for convicts’ ankle bracelets, was added to our phones, and what would have been called wiretapping in years past became a standard feature of countless applications.

The newly formed Cambridge Analytica became that arsenal. Refining techniques from military psychological operations (PSYOPS), Cambridge Analytica propelled Steve Bannon’s alt-right insurgency into its ascendancy. In this new war, the American voter became a target of confusion, manipulation, and deception. Truth was replaced by alternative narratives and virtual realities.

Seemingly out of nowhere, an uprising erupted in America with manic cries of MAGA! and Build the wall! Presidential debates suddenly shifted from policy positions into bizarre arguments about what was real news and what was fake news. America is now living in the aftermath of the first scaled deployment of a psychological weapon of mass destruction.

What became clear in these communications was how early the Russian government had identified the Anglo-American alt-right network, and that it may have groomed figures within it to become access agents to Donald Trump. It showed the connections among the major events of 2016: the rise of the alt-right, the surprise passage of Brexit, and the election of Trump.

Facebook is no longer just a company, I told them. It’s a doorway into the minds of the American people, and Mark Zuckerberg left that door wide open for Cambridge Analytica, the Russians, and who knows how many others. Facebook is a monopoly, but its behavior is more than a regulatory issue—it’s a threat to national security. The concentration of power that Facebook enjoys is a danger to American democracy.

five-factor model of personality, which represents personality as a set of ratings on five scales: openness, conscientiousness, extraversion, agreeableness, and neuroticism. With time and testing, the measurement of these five traits has proven to be a powerful predictor of many aspects of people’s lives.

Insurgencies, by nature, are asymmetric, in that a few people can cause large effects. So catalyzing an insurgency within the belligerent’s organization requires first concentrating resources on a few key target groups. This is optimized by good profiling and identifying the types of people who are both susceptible to new ways of thinking and connected enough to inject our counternarrative into their social network.

The most effective form of perspecticide is one that first mutates the concept of self. In this light, the manipulator attempts to “steal” the concept of self from his target, replacing it with his own. This usually starts with attempting to smother the opponent’s narratives and then dominating the informational environment around the target. Often this involves gradually breaking down what are called psychological resilience factors over several months. Programs are designed to create unrealistic perceptions in the targets that result in confusion and damage self-efficacy. Targets are encouraged to begin catastrophizing about minor or imagined events, and counternarratives attempt to remove meaning, creating an impression of confusing or senseless events. Counternarratives also attempt to foster distrust in order to mitigate communication with others who might hamper the target’s evolution. It is much harder to stay loyal to an existing hierarchy or group when you begin to think that you are being used in some unfair way, or when events seem senseless or purposeless. You become less willing to accept setbacks, take risks, or comply with commands.

But simply degrading morale is often not enough. The ultimate aim is to trigger negative emotions and thought processes associated with impulsive, erratic, or compulsive behavior. This moves a target from mild or passive resistance (e.g., less productivity, taking fewer risks, rumors, etc.) into a re...

People high on the narcissism scale are susceptible because they are more prone to feelings of envy and entitlement, which are strong motivators of rule-breaking and hierarchy-defying behavior. This means these targets will be more likely to develop an exaggerated suspicion of harassment, persecution, victimhood, or unfair treatment. This is the “low-hanging fruit” for initiating the subversion of a larger organization. Later, this learning would serve as one of the foundations for Cambridge Analytica’s work catalyzing an alt-right insurgency in America.

Let’s be clear: These operations are not some kind of therapeutic counseling; they are a form of psychological attack. It’s important to remember that, in a military context, the target’s personal agency or consent is not a concern. The target is the enemy. The choice for the military is often either to send in a drone to incinerate the enemy or to mess with the enemy’s unit to such an extent that they begin to fight among themselves or get sloppy and make exploitable mistakes. If you are a military commander or an intelligence officer, psychological manipulation is the “light touch” approach.

With the advent of social media, suddenly military and security agencies had direct access to the minds and lives of guards, clerks, girlfriends, and runners of criminal and terrorist organizations all around the world. What the social data offered was a trail of detailed personal information that previously would have taken months of careful observation to gather. The targets were in effect creating their own dossiers with rich data that could quicken a psychologist’s assessment of their disposition. This spurred a host of research into psychological profiling that could be automated with machine-learning algorithms. These al...

Although there was substantial underdevelopment of traditional telecommunications infrastructure, largely stemming from corruption and the neglectful legacies of colonial administrations, some of the world’s poorest countries had leapfrogged generations of technology, achieving impressive advances in mobile networks.

One unintended consequence of having large pluralities of citizens connected via mobile phone networks was that everybody could be traced, tracked, profiled, and communicated with. Jihadist networks such as ISIS, AQAP, and Boko Haram had already figured this out, taking advantage of easy access to the minds of future conquests. And that turned the rules of warfare upside down.

If we can measure or infer certain traits in individuals using personal data, and then use those same traits to describe a culture, we can chart a distribution, creating an approximate metric for that culture. This framework made it possible for us to propose how we could use personal data found on social media, in clickstreams, or from data vendors to identify, for example, who the most extroverted Italian people are through their patterns of behavior as individual consumers and users. Then, if one wants to shift the culture to make it slightly less extroverted, this data gives us a list of actual named Italians, ordered by their degree of extroversion, whom we could track and target over time, trying to chip away at their extroversion. In other words, culture change can be thought of as nudging the distribution curve of culture up or down. What the data allowed us to do was to disaggregate that culture into individuals, who became movable units of that society.

THE GOAL IN HACKING is to find a weak point in a system and then exploit that vulnerability. In psychological warfare, the weak points are flaws in how people think. If you’re trying to hack a person’s mind, you need to identify cognitive biases and then exploit them. If you walk up to a random person on the street and ask, “Are you happy?” the chances are high that she will say yes. If, however, you walk up to that same person and first ask, “Have you gained weight in the last few years?” or “Are any people from your high school more successful than you?” and then you ask “Are you happy?”—that same person will be less inclined to answer yes. Nothing about her personal situation or history has actually changed. But her perception of her life has. Why? Because one piece of information in her mind was weighted more than the others.

What we played with as the questioner was how she was weighting that information, which in turn affected her judgment of that information. We biased her mental model of her life. So which is true? Is she happy or not happy? The answer depends on which information is being pulled to the front of her mind. In psychology, this is called priming. And this is, in essence, how you weaponize data: You figure out which bits of salient information to pull to the fore to affect how a person feels, what she believes, and how she behaves.

We are all affected with cognitive biases, which are the commonly occurring errors in our thinking that generate flawed subjective interpretations of information. It is completely normal for people to process information with bias—in fact, everyone does—and oftentimes these biases are harmless in day-to-day life. These biases are not random in each person. Rather, they are systematic errors, meaning they create patterns in common forms of irrational thinking. In fact, thousands of cognitive biases have been identified in the field of psychology. Some biases are so common and seemingly intuitive that it can be hard for people to even recognize that they are actually irrational.

For example, the psychologists Amos Tversky and Daniel Kahneman conducted a study that asked participants a very simple question: “Suppose you sample a word at random from an English text. Is it more likely that the word starts with a k, or that k is the third letter?” Most people responded with the former, that words that start with k (e.g., kitchen, kite, or kilometer) are more likely. However, the opposite is true, and one is actually twice as likely in a typical English text to encounter words where the third letter is a k, such as ask, like, make, joke, or take. They tested for five letters (k, l, n, r, and v) like this. It is easier for people to think of words by first letter because we are taught to organize (or alphabetize) words by their first letter. However, people conflate this ease of recall with frequency or probability, even when this is far from the truth. This cognitive bias is called the availability heuristic, and is just one of many biases that affect our thinking. The bias is why, for example, people who see more news reports of violent murders on the news tend to think that society is becoming more violent when in fact global murder rates have been declining overall during the last quarter century.

But Mercer, it seemed, wanted us to attempt an even more ambitious version of profitable disruption. By profiling every citizen in a country, imputing their personalities and unique behaviors, and placing those profiles in an in silico simulation of that society (one created inside a computer), we would be building the first prototype of the artificial society. If we could play with an economy or culture in a simulation of artificial agents with the same traits as the actual people they represented, we could just possibly create the most powerful market intelligence tool yet imagined. And by adding quantified cultural signals, we were verging on a new area of something akin to “cultural finance.” We thought that if we got it right, we could run simulations of different futures of whole societies. Forget shorting companies; think about entire economies.

The obvious first step was to visit the state and talk to people, to learn how they perceived the world and what mattered to them. We couldn’t generate questions until they had introduced themselves to us, in their own way and in their own environment. Once we had a better feel for what Virginians cared about and how they approached things, we could then structure specific questions for quantitative research. Politics and culture are so intertwined that one cannot usually study one without the other.

In the five-factor model of personality, conservatives tend to display a combination of two traits: lower openness and higher conscientiousness. In the most general way, Republicans aren’t likely to seek out novelty or to express curiosity about new experiences (with closet cases being the obvious exception). At the same time, they favor structure and order, and they don’t like surprises. Democrats are more open but also often less conscientious. This is in part why political debates often center around behavior and the locus of personal responsibility.

It turns out that Republicans can accept a batshit insane candidate, so long as it’s consistent insanity.

In the 1990s, political strategists started buying personal information to use in campaigns. Think about it: If you know what kind of car or truck a person drives, whether they hunt, what charities they give to, and what magazines they subscribe to, you can start to form a picture of that person. Many Democrats and Republicans have a look. And their look was captured in this data snapshot. You can then target potential voters based on that information.

We also got access to census data. Unlike developing nations with less stringent privacy controls, the U.S. government won’t provide raw data on specific individuals, but you can get information, down to the county or neighborhood level, on crime, obesity, and illnesses such as diabetes and asthma. A census block typically contains six hundred to three thousand people, which means that by combining many sources of data, we could build models that infer those attributes about individuals. For example, by referencing risk or protective factors for diabetes, such as age, race, location, income, interest in health food, restaurant preference, gym membership, and past use of weight-loss products (all of which are available in most U.S. consumer files), we could match that data against aggregated statistics about a locality’s diabetes rates. We could then create a score for each person in a given neighborhood measuring the likelihood that they had a health issue like diabetes—even if the census or consumer file never directly provided that data on its own.

I once met the primatologist Jane Goodall, and she said something that always stuck with me. Mingling at a reception, I asked why she researched primates in the wild instead of in a controlled lab. It’s simple, she said: Because they don’t live in labs. And neither do humans. If we are to really understand people, we have to always remember that they live outside of data sets.

Fox News, and all I could think about was how the network was conditioning people’s sense of identity into something that could be weaponized. Fox fuels anger with its hyperbolic narratives because anger disrupts the ability to seek, rationalize, and weigh information. This leads to a psychological bias called affect heuristic, where people use mental shortcuts that are significantly influenced by emotion. It’s the same bias that makes people say things they later regret in a fit of anger—in the heat of the moment they are, in fact, thinking differently.

With their guards down, Fox’s audience is then told they are part of a group of “ordinary Americans.” This identity is hammered home over and over, which is why there are so many references to “us” and direct chatting to the audience by the moderators. The audience is reminded that if you are really an “ordinary American,” this is how you—i.e., “we”—think. This primes people for identity-motivated reasoning, which is a bias that essentially makes people accept or reject information based on how it serves to build or threaten group identity rather than on the merits of the content. This motivated reasoning is how Democrats and Republicans can watch the exact same newscast and reach the opposite conclusion. But I began to understand that Fox works because it grafts an identity onto the minds of viewers, who then begin to interpret a debate about ideas as an attack on their identity. This in turn triggers a reactance effect, whereby alternative viewpoints actually strengthen the audience’s resolve in their original belief, because they sense a threat to their personal freedom.

One of the problems I noticed with the topical political debates on American cable news channels is the lack of nuance in labeling voter constituencies. White voters, Latino voters, women voters, suburban voters, etc., are all frequently discussed as unidimensional and monolithic groups, when in fact the salient aspects of many voters’ identities do not actually reflect the labels that pollsters, analysts, or consultants use to describe them. And this in turn alienates certain people. If you’re a white man living in a trailer, for example, you’re probably going to get angry when you are shown people on TV who are insisting that white people are super-privileged in this country. If you grew up using an outhouse, you probably don’t have much tolerance for a big discussion about whether trans people should be able to use the toilets of their choice. If you’re lower middle class and you see a black person on welfare, it’s not surprising that your attitude would be “Well, what about my welfare?” if you live in a state that has continually cut your support. This is not to defend these views, but if we want to understand them, we have to remain open to other perspectives, even ugly ones.

Harvesting social media data to profile users with an algorithm was just the beginning. Once their behavioral attributes were inferred, simulations could be run to map out how they would communicate and interact with one another at scale.

Moldbug’s views on “truth” influenced Bannon and what Cambridge Analytica would become. Moldbug has written that “nonsense is a more effective organizing tool than the truth,” and Bannon embraced this. “Anyone can believe in the truth,” Moldbug writes. “It serves as a political uniform. And if you have a uniform, you have an army.”

When it comes to what’s happening in other places, Americans will talk about “tribes,” “regimes,” “radicalization,” “religious extremists,” “ethnic conflicts,” “local superstitions,” or “rituals.” Anthropology is for other people, not Americans. America is supposedly this “shining city upon a hill,” a term Ronald Reagan famously popularized, adapting it from the biblical story of the Sermon on the Mount. But when I would see evangelists prophesying the end times and woe unto the nonbelievers, when I watched a Westboro Baptist Church demonstration, when I saw a gun show with bikini-clad ladies carrying semi-automatics, when I heard white people talk about “black thugs” and “welfare queens,” I saw a country deep in the throes of ethnic conflict, religious radicalization, and a bubbling militant insurgency. America is addicted to its own self-conception, and it wants to be exceptional. But it’s not. America is just like any other country.

The United States has its own origin myths, its own extremist groups. At SCL, I had the displeasure of watching countless propaganda videos disseminated by ISIS and deadbeat wannabe African warlords. The way members of jihadist cults fetishize their guns is no different from the way members of the NRA fetishize theirs. I knew that if we were going to truly study America, we needed to do it as if we were studying tribal conflict—by mapping out the country’s rituals, superstitions, mythologies, and ethnic tensions.

For example, people kept talking, unprompted, about the seemingly obscure issue of congressional term limits. They’d say over and over that the big problem in Washington is that the politicians stay in office too long and get bought by special interests. At one focus group in North Carolina, a couple of people used the phrase “Drain the swamp,” so he included that in the notes he sent back, too. CA would later study that phrase using multivariate tests on online panels of target voters, to see whether it resonated with voters.

The evidence was clear: The patterns of a social media user’s likes, status updates, groups, follows, and clicks all serve as discrete clues that could accurately reveal a person’s personality profile when compiled together.

One of the challenges for social sciences like psychology, anthropology, and sociology is a relative lack of numerical data, since it’s extremely hard to measure and quantify the abstract cultural or social dynamics of an entire society. That is, unless you can throw a virtual clone of everyone into a computer and observe their dynamics.

They told me, essentially, that Facebook simply let them take it, through apps the professors had created. Facebook wants people to do research on its platform. The more it learns about its users, the more it can monetize them. It became clear when they explained how they collected data that Facebook’s permissions and controls were incredibly lax. When a person used their app, Stillwell and Kosinski could receive not only that person’s Facebook data, but the data of all of their friends as well. Facebook did not require express consent for apps to collect data from an app user’s friends, as it viewed being a user of Facebook as enough consent to take their data—even if the friends had no idea the app was harvesting their private data. The average Facebook user has somewhere between 150 and 300 friends.

On Facebook, you could “like” things—pages of brands or topics, along with the posts of friends. The purpose of liking was to allow users a chance to curate their personas and follow updates from their favorite brands, bands, or celebrities. Facebook considers this phenomenon of liking and sharing the basis of what it calls a “community.” Of course, it also considers this the basis of its revenue model, where advertisers can optimize their targeting using Facebook data. The site also launched an API (application programming interface) to allow users to join apps on Facebook, which would then ingest their profile data for a “better user experience.” In the early 2010s, researchers quickly caught on that entire populations were organizing data about themselves in one place. A Facebook page contains data on “natural” behavior in the home environment, minus the fingerprints of a researcher. Every scroll is tracked, every movement is tracked, every like is tracked. It’s all there—nuance, interests, dislikes—and it’s all quantifiable. This means the data from Facebook has increasingly more ecological validity, in that it is not prompted by a researcher’s questions, which inevitably inject some kind of bias.

Previously, the only way one could have acquired such data would have been from your bank or phone company, which are strictly regulated to prevent access to that sort of private information. But unlike a bank or telecom company, social media operated with virtually no laws governing its access to extremely granular personal data.

Although many users tend to distinguish between what happens online from what happens IRL (in real life), the data that is generated from their use of social media—from posting reactions to the season finale of a show to liking photos from Saturday night out—is generated from life outside the Internet. In other words, Facebook data is IRL data. And it is only increasing as people live their lives more and more on their phones and on the Internet. This means that, for an analyst, there’s often no need to ask questions: You simply create algorithms that find discrete patterns in a user’s naturally occurring data. And once you do that, the system itself can reveal patterns in the data that you otherwise would have never noticed.

Facebook users curate themselves all in one place in a single data form. We don’t need to connect a million data sets; we don’t have to do complicated math to fill in missing data. The information is already in place, because everyone serves up their real-time autobiography, right there on the site. If you were creating a sys...

In fact, a 2015 study by Youyou, Kosinski, and Stillwell showed that, using Facebook likes, a computer model reigned supreme in predicting human behavior. With ten likes, the model predicted a person’s behavior more accurately than one of their co-workers. With 150 likes, better than a family member. And with 300 likes, the model knew the person better than their own spouse. This is in part because friends, colleagues, spouses, and parents typically see only part of your life, where your behavior is moderated by the context of that relationship. Your parents may never see how wild you can get at a 3 A.M. rave after dropping two hits of MDMA, and your friends may never see how reserved and deferential you are in the office with your boss. They all have slightly different impressions of who you are. But Facebook peers into your relationships, follows you around in your phone, and tracks what you click and buy on the Internet. This is how data from the site becomes more reflective of who you “really are” than the judgments of friends or family. In some respects, a computer model can know a person’s habi...

Psychologists and university researchers soon discovered that MTurk was a great way to leverage large numbers of people to fill out personality tests. Rather than have to scrounge for undergraduates willing to take surveys, which never gave a truly representative sample anyway, researchers could draw all kinds of people from all over the world. They would invite MTurk members to take a one-minute test, paying them a small fee to do so. At the end of the session, there would be a payment code, which the person could input on their Amazon page, and Amazon would transfer payment into the person’s account. Kogan’s app worked in concert with MTurk: A person would agree to take a test in exchange for a small payment. But in order to get paid, they would have to download Kogan’s app on Facebook and input a special code. The app, in turn, would take all the responses from the survey and put those into one table. It would then pull all of the user’s Facebook data and put it into a second table. And then it would pull all the data for all the person’s Facebook friends and put that into another table.

Users would fill out a wide battery of psychometric inventories, but it always started with a peer-reviewed and internationally validated personality measure called the IPIP NEO-PI, which presented hundreds of items, like “I keep others at a distance,” “I enjoy hearing new ideas,” and “I act without thinking.” When these responses were combined with Facebook likes, reliable inferences could then be made. For example, extroverts were more likely to like electronic music and people scoring higher in openness were more likely to like fantasy films, whereas more neurotic people would like pages such as “I hate it when my parents look at my phone.” But it wasn’t simply personality traits we could infer. Perhaps not surprisingly, American men on Facebook who liked Britney Spears, MAC Cosmetics, or Lady Gaga were slightly more likely to be gay.