According to a story doing the rounds, psychometric big data pushed Britain into Brexit and Trump on to America. The winning sides adopted a method developed at the University of Cambridge to psychometrically profile people by using publicly available data including Facebook "likes". They used these to create devastatingly effective digital advertising and targeted millions of voters' psychological traits.
The work was undertaken by Cambridge Analytica, a data analysis company backed by a Donald Trump-supporting billionaire, which acquired the Cambridge method from under the nose of its co-inventor Michal Kosinski. The story ends with Kosinski haunted by having revealed the existence of this digital, election-winning "bomb".
It's a cracking yarn, first told by Swiss publication Das Magazin, translated samizdat-style by a blog then republished by Vice's Motherboard. Its explanation of how Brexit and Trump may have used data to manipulate votes has been followed up by The Observer and The Times among others.
However, Kosinski's work develops ideas that have been around for decades. Cambridge Analytica, which offered to answer questions for this article but has not yet responded, has elsewhere denied using the methodology. And Kosinski believes micro-targeting based on online data could strengthen democracy, not cripple it.
In 2013, when working at the University of Cambridge's Psychometrics Centre, Kosinski co-authored a scientific paper showing that Facebook likes successfully predicted whether someone voted Democrat or Republican 85 per cent of the time, sexual orientation in 88 per cent of men and whether someone was African American or Caucasian American in 95 per cent of cases. Using data from more than 58,000 volunteers who provided their likes and took standard personality tests, the authors wrote that likes were nearly as good a guide to someone's openness as the test itself.
Theresa May – liberal and artistic
A version of this test is online here, with another that analyses language. The first 1,000 words of Conservative Prime Minister Theresa May's Brexit speech from January generates a 67 per cent openness rating, making her "liberal and artistic" rather than "conservative and traditional", and a 99 per cent score for her being a man.
To be fair, a political speech is a controlled and stylised piece of writing, and such predictions are much more likely to work based on what we do and say online day in, day out. Even though Facebook likes are no longer public, there are plenty of social media alternatives. "We're willingly leaving a lot of digital footprints that we don't really want to hide," Kosinski, now an assistant professor of organisational behaviour at Stanford Graduate School of Business, tells The Reg.
"What people don't realise is that this information that we willingly want to share is already more than enough for a good algorithm – and algorithms are quickly getting better – to reveal your intimate traits that perhaps you may not want to reveal, like your political views, religiosity, personality, IQ, sexual orientation and so on.
"What I think people have to understand is that going forward there's going to be less privacy and perhaps no privacy whatsoever. Instead of assuming that somehow we can get our privacy back, we should focus on shaping culture, societies and law to accommodate for the post-privacy future."
Using psychology to group people is not new.
Jon Morris, professor of advertising at the University of Florida, says that the concept of "early adopters" was first described by Everett Rogers in 1962, and electrocardiogram tests have been used since the 1950s. "Psychological profiling, using any kind of a device or any kind of questioning, may have become a little more sophisticated, but it's certainly been around forever in the life of advertising," he says.
He and other academics are similarly comfortable that people's votes are swayed by emotions and that you can get psychological insights from online data. Morris's research on voting has found emotions are about twice as important as facts, and that comparing data from FMRI brain-scanning and online surveys shows the latter do a good job in assessing traits.
Data's dirtier than you think
Andrew McStay, reader in advertising and digital media at Bangor University, has his doubts about the real-world applicability. "There are always huge promises of accuracy, but advertisers themselves tend to treat it – not sceptically as such, because they use it and they do buy into it, but I think they're slightly cautious," he says, with disappointing effectiveness ratings for supposedly targeted adverts.
"The technology industry might depict it as a perfect science, in reality the data is much dirtier than they sometimes suggest."
Kosinski acknowledges that advertising has used psychological grouping for decades, but believes the technique he demonstrated allows greater accuracy. "The resolution is higher. Instead of targeting people on a group level, the algorithms allow you to target people on the individual level."
Politicians can use it to bad ends, he agrees, but adds: "Lying to people and trying to scare them about others, or misinforming people using propaganda or discouraging people from voting – those are bad examples of political communication. Those bad and good examples of political communication are completely independent of the channel. You don't need digital targeting of your political messages to lie to people or discourage them voting. You can use traditional TV or a newspaper."
This is not quite the dystopian message of the Das Magazin article. "This article presents its authors' point of view," says Kosinski. "I share most of their opinions but not all. When it comes to the quotes from me, I think they're all accurate."
As to whether Cambridge Analytica drew on his work, he believes the company redeveloped ideas from his papers, noting the company's use in presentations of examples he used and its hiring of his PhD examiner, who had close knowledge of his work.
In the Motherboard article, Cambridge Analytica said it had no dealings with Kosinski, uses a different methodology, hardly used psychographics in the US election and did not attempt to discourage any Americans from voting.
Kosinski adds: "It doesn't really matter much if they were inspired by my work or stumbled on these ideas independently, I think. What matters is what they were doing and how they were going about it."
It's not all doom and gloom
Outside politics, he thinks that analysis of digital footprints could be used to match people with a job or an interesting book or to automate psychological diagnosis, making it accessible to many more people. Universities and employers could use it to search for scholarship and apprenticeship candidates who lack qualifications, looking for potential rather than how much families have spent on their children's education, opening access to "positions that before were given only to people who could show they graduated from Cambridge," he says.
Regardless of its other uses, some are worried by politicians using psychological profiling. Peter Kinderman, professor of clinical psychology at the University of Liverpool, notes that politicians have used psychological tricks throughout history. But if they want to use psychological profiling they should obtain informed consent, as scientific researchers and clinicians are required to do. "They should be honest and we should be aware."
Bangor's McStay says that informed consent for online psychological profiling is currently the exception, given that most web browsers accept and retain cookies by default. Martin Sykora, a lecturer in information management at Loughborough University whose research using Twitter predicted Trump's success, points out that many apps demand wide-ranging access to personal data as a condition of use. They don't explain why, so it can't be called informed consent. "Companies often don't seem to really bother with that too much and maybe should give it a bit more thought," Sykora says.
Kosinski argues it would be hard to ban politicians from using psychologically targeted messages, as companies could operate from other countries or find other ways around it. "Instead of banning particular methods of doing the bad thing, we should ban the bad thing itself," he says.
"If people are really concerned about discouraging people from voting, let's talk about how to enact policies or regulations or build technologies that protect people from being discouraged from going to vote, or let's just make voting obligatory."
As for Cambridge Analytica's work for Donald Trump, Kosinski says: "I think people are getting upset about it because they need a scapegoat. Back when Obama used similar methods just calling them different names, no liberals were losing their sleep. They also did not care when Hillary was spending way more money on personalised political marketing delivered by people way more competent than those working for Trump."
He believes that online personalised political messages can be more interesting and relevant than general ones, and the recipients are more competent to judge their quality. It also makes political communication far cheaper because it is highly efficient.
"One can enter the political race at much lower cost, and we could see that with Brexit, but also Trump and Bernie Sanders," Kosinski says. "It completely changes the economics of political targeting."
And it could allow politicians to reach groups who have previously been ignored and may not have bothered voting as a result. "With both Donald Trump and Bernie Sanders, they attracted huge followings among groups of people who previously were not politically active," Kosinski says.
"It's creating turbulence as we speak in different countries. But I think in the long term, if we can survive the turbulence now, it's actually great news for democracy that more people enter the dialogue and are being engaged." ®