Cambridge Analytica: Microsoft’s Exploitative Ad-Tech
Trump & Brexit’s infamous data firm was allegedly grown in the Microsoft-funded advertising research labs at Cambridge University.
Steve Bannon-owned Cambridge Analytica LLC, a US subsidiary of UK defense contractor SCL Group, has been increasingly faulted for its dual role in both Donald Trump and Brexit’s victories. SCL is staffed by a number of former UK military psychological operations specialists, and both Brexit’s Leave.EU & Trump’s GOP campaign data teams frequently boasted of Cambridge Analytica’s claimed ability to predict the behaviors of hundreds of millions of voters. Despite all this, the syncretic backstory behind its evolution and rise to power — especially with regard to the creation of its most discussed toolkit — has not been properly examined by any other news outlets, possibly because it involves well-respected heavyweights of tech and academia.
Henceforth referred to here as CA or CambAnal, Cambridge Analytica has garnered a large amount of public interest over the last year and a half for its use of big data in elections. Its hypertargeted advertising and customer profiling database technology is exactly what was being sought back in 2012, when Bannon advised Rebekah and Robert Mercer to invest $5m in SCL Group. That influx of cash led to the formation of CA’s UK cousin and predecessor, SCL Elections.
The primary subject of CambAnal’s received media attention has been debate over the controversial merits of CA’s claimed use of Five Factor Model or BIG5 personality profiling — aka OCEAN, an acronym derived from Openness, Conscientiousness, Extraversion, Agreeableness, and Neuroticism — with much less of a focus on either the sourcing of the identifying datapoints or the origination of the tools & methodology CA deployed. As it turns out, it seems the parties involved in that backstory would like to keep it that way.
A December 2016 article in Das Magazin — a sensation in Germany at the time, which spread virally upon its English translation published by VICE’s Motherboard in January — told a story relayed by Michal Kosinski, formerly Director of Operations for Cambridge University’s Psychometrics Centre and Leader of the e-Psychometrics Unit. Kosinski seemed to be suggesting SCL and CA might have stolen (or at least repurposed without permission) his team’s research into third-party use of millions of Facebook interactions as data point items, which Kosinski et al collected using a Facebook quiz app and then ran through OCEAN profiling for personality cataloguing purposes.
Cambridge University played a major role in an earlier TEXTFIRE article, as it has become a hotbed of international spy chatter over the last several months, encompassing rumors about Gen. Michael Flynn and Russian influence at the renowned spycraft coffee klatch, the Cambridge Intelligence Seminar (where key UK military psychological operations expert Stephen Jolly is said to be a frequent attendee).
Nearly every article about CambAnal across the media spectrum has mentioned Kosinski, yet no reporters have investigated Cambridge University any further. Not only has Cambridge escaped scrutiny, but so has its longtime partner organization on its Facebook & psychometrics projects: Microsoft.
At the time of his key March 2013 journal article, Kosinski was also a research advisor to the Online Services and Advertising group at Microsoft Research Cambridge. One of his two article co-authors was Dr. Thore Graepel, who “is a researcher at Microsoft Research Cambridge leading the Online Services and Advertising and Applied Games group, where their work is focused on the application of large scale machine learning and probabilistic modelling techniques to a wide range of problems including online advertising, web search, and games,” according to his still-active Cambridge online profile.
Perhaps Cambridge simply needs to update its website, for Dr. Graepel currently works as Research Lead at Google DeepMind and Professor of Computer Science at University College London. DeepMind is a company Google purchased in 2014, which seeks to conquer quantum computing but now primarily develops “the best techniques from machine learning and systems neuroscience to build powerful general-purpose learning algorithms,” and which has come under fire for its mishandling of millions of UK patients’ health data while under contract from Britain’s NHS.
The other significant co-author on that publication (and its several predecessors, starting in Feb 2012) was Dr David Stillwell, Deputy Director of The Psychometrics Centre, whose Cambridge profile labels the barely-30y/o as the “Mark Zuckerberg of Psychometrics.” Stillwell originally wrote the myPersonality app which gathered the studies’ data from millions of Facebook profiles, doing so while preparing for his graduate studies in 2007. Just one year after Facebook first became available to the general public, myPersonality began hoarding user profiles.
Stillwell also created a Facebook app called You Are What You Like, a product released no later than October 2013. “By linking the myPersonality database to Facebook Likes we have created an application that is able to predict your personality simply by logging in to your Facebook account,” boasted its deactivated Cambridge webpage.
Another tool developed from myPersonality (and its cousin myIQ) was LikeAudience, described in the 2010–11 annual Cambridge report as “a breakthrough research and marketing tool developed by Michal Kosinski that provides detailed psycho-demographic profiles of brands, products, ideas, and all the other things people like on Facebook.”
The Psychometrics Centre was working on “gender difference in intelligence” and “perceptions of intelligence in Facebook images and their relationship to actual IQ and to personality, the validity of social network likes as predictors of product preference and the relationship between personality and product preference,” the report went on to say. “Michal Kosinski established a continuing collaboration with the Microsoft Laboratory at Cambridge, on several projects including an exploration of crowd sourcing of intelligence in the Amazon Mechanical Turk.”
“Never before have we had access to such comprehensive behavioural data about consumers,” said Kosinski in a 22 Apr 2011 article by Cambridge. “We think [LikeAudience] will revolutionise marketing, because it introduces a completely new dimension by adding scientifically robust personality tests to other demographic information.”
The same article announced Kosinski and Stillwell believed that LikeAudience would be “of particular value to marketers, who will be able to uncover new potential audiences for their advertising campaigns, and exploitable niches based on the fans of their closest rivals. The potential significance for politicians, particularly when on the election trail, is also clear.”
Identifying politics and elections as potential use case scenarios has been a common theme with these products. The “How Does It Work?” page on YouAreWhatYouLike.com posited, “If most of the things you like are liked by liberal people (e.g. Quentin Tarantino) — it is quite likely that you are also liberal.”
“Sarah Palin appeals to a rather different personality type. Her followers are likely to be more traditional in mindset, disciplined, dutiful, and older than the average Obama fan.” Further analysis of Palin-approving LikeAudience users found those with similar profiles “still track the fortunes of the last President, George W. Bush, and also enthuse about Pizza Hut and the Seattle Seahawks.”
PreferenceTool, the successor of the deprecated LikeAudience, seems quite sure of its efficacy in applying all this knowledge to behavioural modification. “PreferenceTool enables marketers to significantly improve targeting and reduce the cost of marketing campaigns,” said its splash page. “It is used by leading online marketing agencies and is proved to have increased campaign effectiveness by up to 140%. [emphasis theirs]”
Despite an outward veneer of academically pure intention, this research is clearly being conducted for specific purposes. To get a better sense of their work, Cambridge has helpfully provided us with a litany of online demonstrations of what appears to be its currently most hyped [and administratively sanctioned] product, the online psychometric assessment creation tool Concerto. We can quickly see the difference between Cambridge Psychometric’s historical topics of focus, such as preschool activities and mental disorders, as compared to the names of quizzes performed in Concerto, including several titled “Advertisement Test”.
Both the Psychometrics Centre and Microsoft Research websites plainly hawk their myPersonality-derived services to interested clients, using keywords of ‘advertising’ and ‘machine learning’. Cambridge offers the use of myPersonality data as a service, on both the Psychometrics homepage and and its flagship Apply Magic Sauce site, and also offers open-source data to hundreds of active researchers at myPersonality.org.
Apply Magic Sauce hosts a sample personality quiz, which likely distracts rubberneckers more interested in the self-help quizzes found on Cambridge’s DiscoverMyProfile.com, but its business2business page boasts of services which will sound very familiar if you’ve read this far. “We enable instant psychological assessment of your users based on their online behaviour,” it excitedly exclaims, by using a range of tools “from multi-channel keyword targeting to psycholinguistic tailoring.”
“In a fraction of a second, you can now adjust the presentation, delivery and content of your message to suit the distinct psychological make-up of the person viewing it,” if you are simply willing to pay a few hundred or thousand quid per month, depending on the extent of your customised needs. To be clear, Cambridge University’s publicly available Apply Magic Sauce product is openly offering to perform high-frequency advertisement creation based on psychographic profiling data culled from Facebook, in what is likely a very lucrative setup for the school.
On the private sector side, “Online Services and Advertising and Applied Games group” sounds rather self-explanatory, so it is relatively clear what Microsoft has intended to accomplish through its Cambridge funding. Microsoft Research Cambridge (MSRC) research is being funneled into Microsoft’s Cortana AI tool and its Azure Cognitive Services AI platform, as well as “voting games” and internet ads. An MSRC study from Sept 2014 is titled “Efficient Advert Assignment”, focusing on algorithmic pay-per-click schemes in online search results.
“Only in the recent literature have computationally efficient methods been considered for market and auction design,” the paper reminds us. “In the context of electronic commerce and specifically sponsored search auctions, these computational considerations are of critical importance given the increased diversity and competition associated with online advertising.”
Cambridge and Microsoft’s team-up has produced algorithm-to-algorithm ‘high-frequency trading’ of consumers’ personalities and their custom-targeted online adverts. No wonder the Psychometrics Centre changed departments in 2016, moving from ‘Social and Developmental Psychology’ to ‘Judge Business School’ (where UK ex-Ministry of Defence psyops expert Stephen Jolly is also currently a Fellow).
TAKING THE CAMBRIDGE OUT OF CAMBANAL
With fingers pointing at CambAnal and CambUni as a result of all the recent media coverage, blame has begun to be shifted around. School, tech company, and professors are all turning on each other. In an attempt to uncover their true relationships, data protection advocate and researcher Paul-Olivier Dehaye has filed a number of Freedom Of Information requests with Cambridge, but has thus far only received refusals; instead, the University has appeared to be attempting to distance itself from the research in question. Their stance seems to be that they have no claim to the data or results, that Michal Kosinski owns it all, and that Microsoft is solely responsible for its existence.
In the aforementioned Das Magazin article, Kosinski absolved himself and Cambridge of any responsibility, saying, “This is not my fault. I did not build the bomb. I only showed that it exists.” He instead laid nearly all the blame at the feet of a young Cambridge professor, then known as Dr Aleksandr Kogan. Kosinski said Kogan approached him in late January 2014, attempting to purchase access to myPersonality on behalf of a deep-pocketed company, and it was not until some time later that Kosinski remembered it was SCL. Kogan went on to start his own company and harvested Facebook information from unsuspecting Amazon Mechanical Turk workers, paying the going rate of pennies per full profile download, and then turned that info over to SCL in some sort of licensing deal.
In a public Facebook conversation, Kogan denies almost all of this. He says he and Kosinski knew each other and continued to stay in touch after SCL entered the frame; SCL also met and had direct interactions with Kosinski; and SCL was by that time heavily into machine learning and microtargeting applications of psychological data for political purposes.
As he noted in his lengthy Facebook rebuttal, Kogan legally changed his married name to Aleksandr Spectre in 2016, making him Dr Spectre. He only has a small handful of published papers to his name(s), focused exclusively on emotionality neuroscience, though that didn’t stop him from becoming Director aka “beloved commander” of the Cambridge Prosociality and Well-Being Lab as early as July 2013. His full birth name was Aleksandr Borisovich Kogan, receiving the name of his grandfather, the chief of the physiology school in Rostov-on-Don, Russia, located across from Crimea on the Sea of Azov and a 2 hour drive from Ukraine. In fact, what is now Southern Federal University established the A.B. Kogan Research Institute for Neurocybernetics — the study of machine-to-human brain interfaces, at a lab which has produced the FaceIdent facial recognition software — in honor of the late Kogan.
As easy as it would be to cast the Russian mad scientist’s grandson Dr Spectre as the villain in this British spy story, the problem is Kosinski’s timeline doesn’t make a whole lot of sense.
The word on the street is CambAnal hired researchers from Cambridge University as part of its original staff, which is allegedly where the “Cambridge” came from when naming the company on its first 31 Dec 2013 Delaware registration, along with the perceived instant name cachet. But that this would be less than a month before Spectre allegedly approached Kosinski on behalf of SCL — or before SCL met Kosinski — is an enormous coincidence, especially since the Psychometric Centre’s Facebook-related research publications and marketing products had been making the rounds for 3–5 years at that point. Why would SCL send young Spectre to the Psychometrics Centre at all, since he didn’t work directly with Kosinski, and how/where/why did SCL & Spectre initially cross paths?
Not to mention, the first pre-CA “unpaid pilot project to show how SCL might work” for electioneering purposes was in the 2013 Virginia gubernatorial race, which culminated on 5 Nov 2013. What was their ‘value add’ in mid/late 2013 if SCL was merely offering UK military psyops techniques (as described in our companion article), if Cambridge Analytica and its fancy-sounding name wouldn’t be officially established until two months later, and if they weren’t testing out any BIG5 methodology or offering up psycho-demographic profile data from Facebook?
Steve Bannon — the current White House Chief Strategist who still attends National Security Council meetings — has owned at least a hefty chunk of Cambridge Analytica, the Facebook-powered, international military psyops-based, Mercer-funded, Microsoft-enabled, finance-loopholed, Amazon-trained, far-right-affiliated, foreign-born, algorithmic-advertising-driven media machine which participated in both the victorious Brexit and Donald Trump campaigns.
In fact, Bannon is the only publicly documented shareholder/owner of Cambridge Analytica LLC to date. Every SCL subsidiary in the UK has reported their shareholders on a regular basis for public dissemination, as is the lawful standard, but similar information is generally not available in the US. As a result of Bannon’s required White House financial disclosures, however, we learned he owned between $1m-$5m in CambAnal, though in defiance of legality he had not sold it as of at least late April 2017.
Now, there is no shortage of marketing & PR agencies looking to exploit technology for political or financial gain, as evidenced by a private psychological operations firm-for-hire like Bell Pottinger, British Army psyops-based Verbalisation, or any number of boutique advertising startups. Even frequent GOP candidate vendor Campaign Solutions did so back in 2014 for the radically right wing John Bolton SuperPAC, which received $4m of Robert Mercer’s money between Apr 2014-Sep 2016.
Their job then was to help microtarget DirectTV advertisements by cross-checking state voter files with consumer data such as magazine subscriptions, so the voters could then be sorted into 100 personality types, arranged into 28 clusters, and sent a specific version of a political ad tailored to each of those clusters through their satellite TV. Cambridge Analytica was actually hired to work closely with Campaign Solutions on the 2014 campaigns to create those varying TV ads. In 2015 & 2016, CambAnal was repeatedly contracted by John Bolton SuperPAC to work with Connell Donatelli, which shares an owner with Campaign Solutions, to produce even more targeted ads.
Even though big data is nigh ubiquitous, when Bannon-SCL’s CambAnal later joined Trump’s GOP data team in employing the same high-speed AI advertising and algorithmic purchasing strategies laid out by Microsoft & Cambridge University while partnering with Facebook & Google’s ad deployment platforms during the 2016 elections, it was a very big deal.
A foreign defense contractor, specializing in military-grade behavioral modification techniques being run through artificial intelligence from the brightest minds in the world, has for years been embedded directly inside domestic political campaigns at the highest level. The campaigns themselves deserve increased scrutiny, but perhaps the researchers powering the electronic brains behind this scheme — and the advertising industry as a whole — should be examined and held accountable.
The preceding is an edited excerpt from TEXTIFIRE’s behemoth report: