Mindfuck

by Christopher Wylie

  Dancing a delicate ballet among multiple jurisdictions, intelligence agencies, legislative hearings, and police authorities, I have given more than two hundred hours of sworn testimony and handed over at least ten thousand pages of documents. I found myself traveling around the world, from Washington to Brussels, to help leaders unpack not only Cambridge Analytica but also the threats social media poses to the integrity of our elections.

  Yet, in my many hours of giving testimony and evidence, I came to realize that the police, the legislators, the regulators, and the media were all having a difficult time figuring out what to do with this information. Because the crimes happened online, rather than in any physical location, the police could not agree on who had jurisdiction. Because the story involved software and algorithms, many people threw up their hands in confusion. Once, when one of the law enforcement agencies I was dealing with called me in for questioning, I had to explain a fundamental computer science concept to agents who were supposedly specialists in technology crime. I scribbled a diagram on a piece of paper, and they confiscated it. Technically, it was evidence. But they joked that they needed it as a crib sheet to understand what they were investigating. LOL, so funny, guys.

  We are socialized to place trust in our institutions—our government, our police, our schools, our regulators. It’s as if we assume there’s some guy with a secret team of experts sitting in an office with a plan, and if that plan doesn’t work, don’t worry, he’s got a plan B and a plan C—someone in charge will take care of it. But in truth, that guy doesn’t exist. If we choose to wait, nobody will come.

  CHAPTER 2

  LESSONS IN FAILURE

  -

  It was eight years before this that I moved to England, where the story of my entanglement with Cambridge Analytica began. I’d worked in Canadian politics for a few years, but the irony is that I moved to London to escape politics. In the summer of 2010, I moved into a flat on the south bank of the River Thames, near the Tate Modern, the modern art museum housed in the colossal old Bankside Power Station. After several years in Ottawa, I had decided, at age twenty-one, to leave politics and move across the Atlantic to attend law school at the London School of Economics and Political Science (LSE). No longer in politics, I was unshackled from my old responsibilities to the party. It didn’t matter anymore who I might be seen with, and I no longer had to watch what I said or think about who might be listening. I was free to meet new people and I was excited to make a new life for myself.

  When I arrived it was still summer, and the first thing I did after unpacking was head out to sit with the sunbathers, tourists, and young couples in Hyde Park. I took full advantage of London, spending Friday and Saturday nights in Shoreditch and Dalston, and Sundays in Borough Market, London’s oldest food market, which is crammed into an outdoor hall alive with a cacophony of shouting traders, visitors, and cooking stalls. I started making friends with people my age, and for the first time, I felt young.

  But a few days after I arrived, still feeling fuzzy from the jet lag, I got a call that made it clear that it wouldn’t be so easy to leave politics behind. Four months earlier, a man named Nick Clegg had become the country’s deputy prime minister.

  First elected to the European Parliament in 1999, he worked his way up to become, in 2007, the leader of the Liberal Democrats. This was back in the days when the Lib Dems were the radical third party of British politics—the first to support same-sex marriage, and the only party to oppose the war in Iraq and call for abandoning Britain’s nuclear arsenal. In the 2010 general election, after more than a decade of Labour’s now tired “third way,” “Cleggmania” swept across Britain. At his peak, Clegg was polling as highly as Winston Churchill and positioned himself as Britain’s answer to Barack Obama. After the election, he became part of a coalition government that made the Conservative David Cameron prime minister. The call was from his office: They’d heard about my data work in Canada and America from mutual connections in liberal politics, and they wanted to know more.

  At the appointed time, I arrived at Liberal Democratic Headquarters (LDHQ), which was then still located at No. 4 Cowley Street in Westminster. Located only a few blocks from the Palace of Westminster, the converted neo-Georgian mansion stood handsomely adorned in crimson brickwork and flanked by large stone chimneys on either side. It was rather outsize for the tiny, winding street, so I had no problem spotting it. Because it housed the offices of a party in Her Majesty’s Government, an armed unit of the Metropolitan Police stood guard nearby, strolling up and down the small side street. After being buzzed in, I heaved open the weighty wooden doors and walked up to reception, where I was greeted by an intern who would show me to the meeting. Still adorned with the manor’s original chandeliers, oak paneling, and fireplaces, the place revealed the faded elegance of a once grand residence, which felt oddly fitting for this once grand party.

  Cowley Street, as they all called it, was unlike anything else I had seen in Canada or the United States. I wondered how all the party staffers waddling past one another in the cramped and creaking hallways could get anything done. Old bedrooms were jammed with desks, and cables connecting to servers were taped up on walls and around doorframes. In one converted closet, a man apparently with sleep apnea was snoring loudly on the floor, but no one paid him much attention. Looking around, I got the impression that this place operated more like an old boys’ clubhouse than a party in government. I walked up a large staircase with ornately carved railings and was shown into a large boardroom that must have once been the main dining room. After I’d waited several minutes, a small cadre of staffers filed in. When the obligatory British small talk concluded, one of them said, “So tell us about the Voter Activation Network.”

  After Obama’s 2008 victory, parties all over the world were becoming interested in this new “American-style campaign,” powered by national targeting databases and big digital operations. Behind the campaign was the emerging practice of microtargeting, where machine-learning algorithms ingest large amounts of voter data to divide the electorate into narrow segments and predict which individual voters are the best targets to persuade or turn out in an election. The Lib Dems wanted to talk to me because they were unsure whether they could translate this new school of campaigning into the British political system. What was so interesting for them about the project I had worked on with the LPC—setting up the same kind of voter-targeting system used by the Obama campaign—was that it was the first of its kind and scale outside the United States. And Canada, like Britain, uses the same first-past-the-post “Westminster model” electoral system and has a diverse array of political parties. In this conversation, the staffers realized that half of the localization work would have already been done if they imported the Canadian version of the technology. At the end of the meeting, they were almost giddy after learning about what the system could do. After leaving, I ran back to school to catch the tail end of a lecture on the rules of statutory interpretation and thought that was the end of it.

  But the Lib Dem advisers called again the next day, asking if I could come back and tell my story to a bigger group. I was in the middle of a lecture, so I didn’t pick up initially, but after four missed calls from a random number, I stepped out to see what was so urgent. There was a senior staff meeting that afternoon, and they asked if I could do an impromptu presentation on microtargeting. So after class, I walked from LSE back to Cowley Street with my backpack filled with textbooks. With such short notice, I didn’t have time to change, so I headed to meet the deputy prime minister’s advisers in a Stüssy graphic print T-shirt and camo sweatpants.

  Walking into the same boardroom, I was met this time with the humming cacophony of a packed room. I was ushered to the front without any prompt, so, after apologizing for my slightly ridiculous attire, I proceeded to just wing it. I told them how the Lib Dems could use microtargeting to overcome the disadvantages that come with being such a s
mall party. And as I continued, I couldn’t help but get more passionate. I hadn’t spoken about this since I left the LPC, and my heart simply poured out. I told them about what I had seen on the Obama campaign, what it was like to see so many people vote for the first time, what it was like to see African Americans at rallies filled with hope. I told them that this was not just about data; this was about how we could reach the people who had given up on politics. This was how we would find them and inspire them to turn out. But, most important, this was about how technology could be the vehicle that this party, which now found itself in the corridors of power, deployed to upend the entrenched class system that underpins so much of British politics.

  A few weeks later, the Lib Dems asked me to come work for them and implement a voter-targeting project in Britain. I was just starting my degree at LSE, and, as a twenty-one-year-old student, I was finally finding my feet in London. I hesitated about whether it was really a good idea to distract myself with politics again. But here was a chance to take the same technology—the same software, and essentially the same project—and finish what I had started in Canada. But it was what I saw hanging so casually on the wall in one of the offices at Cowley Street that finally pulled me in. It was an old yellowing card, with slightly curled corners, with an excerpt of the Liberal Democrats’ constitution that read NO ONE SHALL BE ENSLAVED BY POVERTY, IGNORANCE OR CONFORMITY.

  I said yes.

  * * *

  —

  AFTER THE 2008 PRESIDENTIAL ELECTION, I returned to Ottawa and wrote a report about the Obama campaign’s new technology strategies. It landed with a thud. Everyone was expecting me to tell them about the campaign’s flashy branding, graphics, and viral videos. Instead I wrote about relational databases, machine-learning algorithms, and how these things connected to one another through software and fundraising systems. When I recommended that the party invest in databases, people thought I’d lost my mind. They wanted sexy answers—not this. Obama was their benchmark for a “model” campaign, and they were taken with high cheekbones and pouty lips, not the skeleton and backbone that made it all possible.

  Most campaigns can be boiled down to two core operations: persuasion and turnout. The turnout, or “GOTV” (get out the vote), universe is those people who likely support the candidate but do not always vote. The persuasion universe is the inverse, representing those who likely will vote but do not always support the party. People who are either very unlikely to vote or very unlikely to ever support us are put into an exclusion universe, as there is no point in engaging them. Voters who are both very likely to support the candidate and very likely to vote—these are the “base” voters, and they are typically excluded from contact, but they might be prioritized for volunteer or donor recruitment. Finding the right set of voters to contact is the name of the game.

  In the 1990s American voters were generally targeted using data provided by local or state offices, which typically contained each voter’s party registration (if they had one) and their voting history (which elections they came out to vote in). However, the limitation to this approach is that not all states provide this information, voters change their mind more frequently than they change their party registration (or would not register a party), and this information would tell you nothing about the issues that actually motivate the voter. What microtargeting did was find extra data sets, such as commercial data about a voter’s mortgage, subscriptions, or car model, to provide more context to each voter. Using this data, along with polling and the statistical techniques, it’s possible to “score” all of the voter records, yielding far more accurate information.

  What Obama’s campaign did was mainstream this technique and put it at the heart of its campaign operations. This is important because the organized chaos of campaign activity is typically not what one sees on TV, such as speeches or rallies. Rather, it is the millions upon millions of direct contacts made by volunteer canvassers or via direct mail to individual voters throughout the country. Although less sexy than a beautifully crafted speech or amazing branding, it is this unseen machinery that provides the critical horsepower of a modern presidential campaign. When everyone else is focused on the public persona of the campaign, strategists are focused on deploying and scaling this hidden machinery.

  Eventually some of us in the Opposition Leader’s Office, where I was working in Parliament, realized we could show the party how useful the Voter Activation Network would be if we created a parliamentary version of it for the leader’s interactions with constituents and citizens. The party was unwilling to foot the bill for something so extravagant as a new database, but we realized we had room for it in the leader’s official parliamentary budget. The only problem was that these were technically public funds, and any pilot database we created could not be used for political purposes. But we were not too concerned. A parliamentary version would contain the records of constituents and citizens who had contacted the leader, and since constituents are simply voters with a different hat on, it would allow us to highlight all the same functionalities to the party without needing them to spend anything. Surely, after seeing such a system firsthand, the Liberal Party of Canada would begin to understand the potential of data. We asked Mark Sullivan and Jim St. George if they’d ever thought about expanding VAN internationally—to Canada. They hadn’t done any big projects outside the United States at that point, but they jumped at the chance to work with us. With the help of Sullivan and St. George, we were able to create a Canadianized VAN infrastructure in six months. To the party’s delight, VAN even worked in both English and French. There was only one problem: There was no data to actually fuel the system.

  Computer models are not magical incantations that can predict the world—they can make predictions only when there is an ample amount of data to base a prediction upon. If there was no data in the system, then there could be no models or targeting. It would be like buying a race car but skimping on gasoline—no matter the car’s sophisticated engineering, it just wouldn’t start. So the next step was to procure data for VAN. But data was going to cost money, and because it would be used for campaigns, by law the party had to pay for it, not the leader’s parliamentary office. But almost immediately there was hard pushback from the party, which was not eager for change. I turned to the MP who first brought me into politics, Keith Martin. He had given me my first internship when I was still in school, and later my first real job, in the Canadian Parliament. Martin was often called the “maverick” of Canadian politics, and he staffed his office with mavericks, too. For me, he was a perfect fit. Martin trained as an ER doctor and spent his early medical career in African conflict zones, treating everything from land-mine injuries to malnutrition. This guy was legit cool and lived an amazing life before politics—on the wall in his office he had photos of himself looking like Indiana Jones in a khaki overshirt, sitting with leopards. As an ER medic, he was trained to not waste time, but in politics you survive by wasting time. He was once so incensed at the mechanistic procedures of Parliament that, mid-debate, he picked up “the Mace”—the gold-plated medieval weapon we inherited from Britain that lies in the aisle of the House of Commons.

  In 2009, Jeff Silvester, Martin’s senior adviser, a former software engineer who’d turned his attention to politics, was one of the few people in the party who understood what I was trying to do. He was my mentor and my rock throughout my time at Parliament. I explained that even though the party hadn’t authorized me to move forward with the data targeting program, we needed to do it. And that meant we needed funding. With Martin’s approval, Jeff agreed to help me raise money without telling the party’s national office about it. We started holding secret events where I could explain to would-be donors that we needed this program if the LPC had any hope of being competitive in the twenty-first century. We did all this on the down-low, persuading people on the ground to put in money while the party staff weren’t paying us any attention. In short order,
we raised several hundred thousand Canadian dollars, which was enough to get the program started. Dissatisfied with the national office, the British Columbia wing of the party agreed to be the guinea pigs in our experiment.

  It was not clear whether any of this would work. In the United States, there are only two major parties, whereas in Canada there are five. This means that the dimensionality of what you are predicting for is no longer binary (Democrat or Republican) but multivariate (Liberal, Conservative, New Democratic Party, Green, or Bloc Québécois). With more options, you get many different kinds of swing voters (e.g., Lib-Con vs. Lib-NDP vs. Lib-Green, etc.) who can float their support around in many directions. There was also a far less developed market for consumer data in Canada and Europe, so many of the standard data sets in the USA either were not available or had to be pieced together from many sources. Finally, parties in other countries often have strict donor or spending caps. A lot of people were skeptical that microtargeting could even be deployed outside of America, but I wanted to try nonetheless.

  I called up Ken Strasma, who ran Obama’s targeting operation in 2008, and asked if he’d be willing to help us set up a program in Canada. Strasma’s team in Washington, D.C., then built the models. The B.C. office in Vancouver pieced together useful data sets, such as old polling and canvassing data, and Strasma worked out how to deal with the additional complexities of multi-party politics. Volunteers all over the province were given either these new canvass lists to try out or old lists that would serve as controls. A sigh of relief came to the B.C. party staff as the results came in. As with many campaigns, the party uses persuasion canvasses to reach out to voters who have not yet made up their minds which party they will be supporting. By comparing the successful conversion rates (where a previously undecided voter declares support for your party) of the old lists with those of the new microtargeting lists, the B.C. operation was able to establish that the new targeting approaches had higher conversion rates. It was very exciting. We proved that what Obama accomplished in America would be possible in different political systems around the world. But when the national party in Ottawa found out what we had done, they objected to a national project. They wanted to run campaigns like Obama’s, but when shown how to do it, they refused.