I’m 31 years old, liberal and artistic, impulsive and spontaneous, contemplative, and a tiny bit competitive. I’m also male, but I don’t repress my feminine side.
That’s what Magic Sauce infers from my Twitter feed. It gets my age and gender wrong, but so does Twitter itself, which explains the futile ads I see for beard care products and baldness cures.
Apply Magic Sauce is the algorithm developed by researchers at University of Cambridge Psychometrics Centre, which does not work with the company called Cambridge Analytica, and was not involved in political campaigns during the US Presidential election or the Brexit referendum.
The Psychometrics Centre hit the headlines in 2013 with a paper called ‘Private traits and attributes are predictable from digital records of human behavior’ by Michal Kosinski et al. Their work used Facebook ‘Likes’ to predict personal traits. You may recall some of the odder correlations: they found liking curly fries to be a predictor of intelligence, and linked “being confused after waking up from naps” to male heterosexuality.
The study showed they could predict categories like gender and race well (correlation coefficients of 0.93 and 0.95 respectively) though results on personality traits including openness and intelligence were weaker (0.43 and 0.39 at best). The authors also raised the ethical dilemma of inferring private information from publicly available data. Shortly after the paper’s publication, Facebook ‘Likes’ became private.
Again, Kosinski has no connection with Cambridge Analytica, the company claiming to have used big data and voter profiling to win for Trump, among others.
How did we get from a quirky, if disquieting, piece of academic research to fears that an unholy combination of big data and psychological warfare has “conquered democracy”?
Reading recent press coverage, you might think that Cambridge Analytica were the first to apply big data in political campaigning. But no, targeting potential voters by combining masses of data from diverse sources goes back at least to Barack Obama’s first election campaign.
Daniel Scarvalone, Associate Director of Research and Data at Bully Pulpit Interactive, has worked on many political campaigns for the Democrats and other liberal causes. In O’Reilly publication Data And Democracy, he outlines three stages of digital campaigning.
“Let the audience be your guide. Leverage the sophistication of political analytics programs to define the precise audience you want to reach, and then build strategies matching the consumption and behavioral patterns of that audience.”
As voting day approaches, campaigners focus their finite resources on two groups of people: the undecided, and those who already support you, but need a push to vote.
In the U.S., voter information, often including previous voting behaviour and party affiliation, is easily obtainable. “An overwhelming majority of Democratic campaigns use … Acxiom, Experian, or Infogroup to supplement their voter files with consumer records,” says Scarvalone.
In the U.K. an individual’s data is more protected by law, but all the major British political parties use commercially-available databases to profile potentially winnable voters. Since the 2015 general election, they’ve borrowed technology (and people) from Obama’s successful campaigns.
LA-based NationBuilder offers a software system to organise supporters who give you their email address. If you tried the Labour Party’s “How Many Of Me?” interactive in 2015, they’re probably still tracking your social media feeds.
Next, says Scarvalone:
“Tailor the creative. Digital advertising allows campaigns to tailor creative elements of advertising to specific audiences at scale. Audiences can be shown the customized messages that will move them the most, instead of speaking to everyone with one megaphone.”
It’s telling that Scarvalone dismisses the megaphone. Political parties today are less likely to campaign on big, distinctive principles, and more likely to aim policies at specific voters they need to win over. Abolishing tuition fees? Capping fuel bills? Via Google ads or Facebook, it’s very easy to target students, or people who search for cheaper energy suppliers.
This isn’t new. The UK Labour Party was using market-research-style focus groups to develop policy back in the 20th century. You may not like it – I know I don’t – but it pre-dates the technology. Where digital media do offer something new is Scarvalone’s third stage:
“Don’t just measure how much, measure how well. … Campaigns should be consistently integrating attitudinally based, experimentally-informed programs (EIPs) to measure the efficiency and effectiveness of every dollar they spend.”
How many people watched that video right through? How many clicked the link? Google, Facebook, whoever you pay to place your ads, can give you exact data. And A/B testing of alternative messages isn’t limited to commercial advertisers.
The UK’s Behavioural Insights Unit (Nudge Unit), for example, tested 8 variants of a message encouraging us to register as organ donors. The message that produced the most registrations asks, “If you needed an organ transplant, would you have one? If so, please help others.” By deploying this message, they expect to register 96,000 more potential organ donors in a year.
These are the techniques that Cambridge Analytica claims to use for their clients, political and commercial: combine databases to identify and profile persuadable people, then test different messages till you find the one that best changes their behaviour.
But before you panic about the unseen manipulations of digitally targeted and tested propaganda, take a step back, and think about how it works in the real world. Despite Twitter’s best advertising efforts, I have yet to buy a single beard care product (I’m not a big fan of beards on men, let alone on me).
If there’s one lesson to draw from recent political events, it should be that people are stubbornly difficult to profile, predict or nudge, on important issues at least. Polls based on demographic sorting repeatedly fail to foresee how we will vote. Foregone conclusions are confounded by the willful electorate. Who can blame political leaders for seeking psychometric snake oil and big data voodoo to find out what, or how, we’re thinking?
To my mind, the problem here is not the immense power of technology and science. That’s exaggerated, especially by businesses with a service to sell. The problem is lack of trust in democracy, in politics, and in voters.
Political leaders who cynically try to press our buttons only when they need our votes willingly treat us like passive data points, or rats in behavioural experiments. They’d love to nudge us into approving their policies once every few years before lapsing back into passivity.
But those who think millions of voters were swayed by cleverly placed online advertising are also underestimating our capacity to think for ourselves. The fear of Cambridge Analytica rides on a deeper fear: that most of us are idiots who can be bamboozled by big data and simplistic psychology.
When a vote goes the way you hoped it wouldn’t, it’s tempting to look for something to blame. But it wasn’t big data what won it. It was the lack of convincing arguments on opposing sides that lost it. This is a problem that won’t be solved by more controls on electoral campaigning, but by better political debate. And that starts with a bit of respect for the electorate, your opponents as well as your allies.
Timandra Harkness is the author of Big Data: does size matter? She will be chairing a debate about the future of data in public life at the British Library on 19 September and speaking at the Battle of Ideas on 29 October.