Driving Sentiment Metrics with AI

How Robert Mercer financed the transition from targeting demographics to exploiting psychographics, and why the 70-year-old presidential candidate with the fun hair and Twitter Tourette’s had the dominant cyber game. Expect more of the same on the road to 2020.

OCEAN’s 2016

If you follow the world of finance, and particularly investment, you’re no doubt familiar with Renaissance Technologies’ Medallion portfolio, the most profitable hedge fund in history. Open to employees only, its returns are unmatched: an average of 71.8% during the 20 years between 1994 and 2014; included in that average was a massive 98% return for 2008 alone, in the wake of the global financial crisis and the same year that the S&P dropped 38%. Founded by former NSA cryptanalyst and all-around math genius Jim Simons in 1982, it was the first company to deploy mathematical statistical modeling and quantitive analysis applications as the primary way to predict the stock market. Simons would later be joined by Robert Mercer, a computational finance wizard fresh out of IBM. The company is extremely secretive — their website homepage consists of one sentence and two addresses — and its PhDs hold doctorates in math, physics, and computer science. Statisticians and scientists are preferred employees, prototypical Wall Street-types are met with disdain, and the company has made a ton of money embracing big data analysis decades before the current explosion of integrating AI with fintech began to take hold.

For Simons, who stepped down as CEO of Renaissance Technologies in 2009 (he was followed by Mercer, who served as co-CEO alongside Peter Brown until late 2017), a chunk of that money is spent on many philanthropic endeavors via the Simons Foundation and Math for America. When he has contributed to politicians, they’ve been primarily Democratic. Robert Mercer has different, somewhat quirkier appetites when it comes to distributing his wealth, the most notable being the person more responsible than any other single human for Donald Trump’s victory over Hillary Clinton in the 2016 election. (Though an argument can be made that that distinction should go to Clinton herself.) Mercer’s quirky side is worth examining, but with respect to installing our 45th president, Mercer relied on a three-prong strategy:

1) Pour millions into Cambridge Analytica, where Steve Bannon was a vice-president. Employ the power of data analytics.

2) Pour millions into Breitbart News, where Steve Bannon served as executive chairman. Exploit the power of social media.

3) Pour millions into the campaign to elect Trump, where Steve Bannon served as head of that campaign, and later as chief strategist to President Trump.

Happily lurking in plain sight orchestrating the Trump bombast, everyone got to see Bannon as the bulldog face of this covert agenda cloaked in revolutionary rhetoric, with all the dark lord bravado actually masking Bannon’s essential job: that of being the best girl Friday Robert Mercer would ever hire. Eventually, Bannon’s toxicity would define his expendability quotient after the main objective had been realized, but before being shown the door(s), his staunch dedication to implement Mercer’s bidding in getting Trump elected was pure, blind loyalty to the mission, and required one more moving part. Enter Facebook.

When Mark Zuckerberg testified before Congress in earlier this month, he was well-prepared to point out how Facebook was assured all the compromising data in the Kogan / Cambridge Analytic data breach had been neutralized. When he found out that Facebook was essentially lied to, he immediately leapt into action and waited two years before doing anything about it. Nevertheless, he did come across as reasonably earnest to a room full of people who were clearly doing their level best to understand this visitor from foreign lands. The whistleblower on the other side of this equation, Christopher Wylie, recounts the series of events in a slightly different light: tools he helped build to harvest and exploit the data profiles of tens of millions of American citizens at the direction of Steve Bannon and Cambridge Analytica were known to Facebook by late 2015. Wylie was sent a letter in August of 2016 requesting he immediately delete any of the material obtained. He didn’t. Facebook didn’t follow up. Bannon then became Trump’s campaign manager, which effectively meant Cambridge Analytica became Trump’s campaign manager.

What do you mean, “trying to”?

While it’s easy to be cynical about an extremely wealthy (and cynical) hedge fund manager deciding he wants to be a political puppet master later in life, let’s not give the extremely wealthy coder who swapped out the hoodie for a suit while visiting D.C. a pass just yet.

A leaked document from Facebook Australia detailed one down under research project it would have preferred kept quiet: the company’s use of algorithms to pinpoint and then exploit the sentiments of children and young adults in real time, allowing marketeers to capitalize accordingly along the emotional spectrum of youthful experiences, and not just day to day, but minute by minute. This was research done in 2017, long after Facebook pulled the Aleksandr Kogan app, and on over 6 million people, some barely teenagers.

But the clinical trial perhaps most ominous in the Facebook catalog of sentiment metrics analysis is its attempt at driving those sentiments toward predetermined objectives and observing the results. In 2012, Facebook and Cornell University studied almost 700,000 FB users for a week to gauge responses when one group had positive content removed from their feeds while the another group had negative content removed from their feeds, and both were measured against a control. The objective — to observe if “emotional contagion” was indeed a thing, and would indeed spread, like anything viral — achieved its desired, and unsurprising, result: sad shit makes you sad, and happy shit makes you happy. What was significant was that those leading the experiment neither told the participants what was happening nor obtained their consent to participate in a research study specifically engineered to affect their emotional state. Amazingly, the results were published in the Proceedings of the National Academy of Sciences before everyone involved had to duck and cover for such an wanton breach of trust.

With the Australian case from last year, a wholly unsurprising twist: Facebook claims their internal data about your public data is not for public consumption. This aligns well with points congressman Ben Lujan (D-NM) made to Mark Zuckerberg before his allotted four minutes ran out during congressional testimony: first, he queried Zuckerberg on why a major data scraping vulnerability that Facebook was made aware of in 2013 didn’t get resolved until a week before Zuck’s 2018 arrival in Washington; second, he sought clarity on how many average data points Facebook collects on “shadow profiles,” or people that don’t use Facebook. (FB users, the congressman claims, have an average of 29,000.) Finally, Rep. Lujan asked for an explanation as to why a non-Facebook user who wants to continue being a non-Facebook user, has never signed a consent form with Facebook, and simply wishes to not have their identity tracked and contained within Facebook’s data analytics engines is instructed to sign into their Facebook page to turn off those permissions. It was one of the more informative exchanges of the day, with Zuckerberg’s silent stares speaking volumes.

He’s totally checking you out

It’s important to keep perspective here: Facebook, as Mark Zuckerberg explained to Senator Orrin Hatch through a muffled, I-can’t-believe-these-are-the-dudes-in-charge-of-oversight giggle, makes money through advertising. So of course they use targeted data to sell things, and of course they’re going to let others use targeted data to sell things, and of course they’re going to apologize for it once they get caught doing something unethical. It’s become a facet of their business model. When Zuckerberg was asked by congressman John Sarbanes (D-MD) about the number of Facebook-approved ads for the two presidential candidates in the 2016 election, he didn’t have the numbers handy, but Rep. Sarbanes did: roughly 66,000 ads approved for Clinton, and 5.9 million approved for Trump. Which brings us back to Cambridge Analytica.

There has been a lot of press covering Cambridge Analytica, it’s successful use of psychometric profiling and personality-centric, micro-targeted advertising at the behest of the Trump campaign, and the fact that in three states easily carried by Obama in 2012, Pennsylvania, Michigan, and Wisconsin, Trump won each state’s electoral votes by less than one percentage point. This is the number that should resonate when considering the power of an effectively deployed AI system: yes, the schematic for a personality quiz developed by two doctoral candidates was hijacked by Aleksandr Kogan, a professor doing academic-that-was-actually-commercial research sold to Cambridge Analytica, and the resulting data harvest of over 80 million Facebook users has got the company some well-deserved scrutiny; however, Trump won those three aforementioned states with less than 80,000 votes. Total. This percentage is smaller than 1/1000th of the number of FB profiles exposed, and Cambridge Analytica obviously had access to more than Facebook intel when using data analytics to divide the US populace into 32 personality types. But Facebook users were targeted on a massive scale, and not just by an infamous troll farm in St Petersburg. It took a highly — and artificially — intelligent operation to correlate quantitative, qualitative, and non-linear data-driven insights to position the Trump campaign heavily in Michigan and Wisconsin during the final weeks of the 2016 election. Trump took Michigan by 0.2%. This will look less random as time allows emotional humans to catch up to what machine-learning frameworks can already comprehend.

If the data science behind such an operation is extremely intricate, the layout for this optimization loop would present as the most rudimentary Mark Lombardi sketch ever: a circle of speech bubbles tied together describe Bannon / Breitbart > Bannon / Cambridge Analytica > Facebook > Target Demo > Bannon / Breibart > Bannon / Cambridge Analytica > Facebook > Target Demo, etc., repeating endlessly and cyclically until desired feedback has been actualized. The first three have direct, connective tissue to the bubble with Robert Mercer’s name in it; the fourth, Target Demo, generally has no idea who Mercer is but does identify with having perceptions validated and reinforced, even if the shaping of those perceptions goes unnoticed.

How to Shag an Election

Robert Mercer funded Cambridge Analytica to influence U.S. elections the way he saw fit. Operating under the umbrella of the SCL Group, a British strategic communications firm, Cambridge Analytica was set forth “to address the vacuum in the US Republican political market.” It’s behavioral modification by way of digital engagement, according to its recently unemployed founder and hot mic celebrity, Alexander Nix. And the personal ROI for Mercer has surely mirrored any financial ones he’s received from Renaissance’s Medallion Fund over the years.

It’s important to note, however, that Mercer initially deployed Cambridge Analytica to work its magic for Ted Cruz, his first choice in the 2016 election. When Trump won the primary, Mercer capitalized on that, and certainly not because Trump is also an introverted, socially awkward, gifted data scientist and successful billionaire (Trump is none of those things); rather, it’s because Robert Mercer wanted to buy the largest number of shares — controlling stock — in the American presidency, which he appears to have done.

As the secretive king maker who is reportedly cripplingly shy in social situations, Robert Mercer would probably not appreciate his personal data being mined and available for purchase, yet he appears highly motivated to do just that to others, as it gels with his world view in general: people are worth what they earn, and if he earns thousands of times what a schoolteacher earns, he is therefore that much more valuable than a thousand teachers. It is blatantly obvious that Mercer cares not a whit about Trump supporters or their values, other than as fodder to suit his interests. Unlike Facebook / Cambridge Analytica publicity that has erupted in recent weeks, opensecrets.org has unearthed another funnel of neoconservative dark money that infiltrated the waning days of the 2016 election with fear-mongering advertising: a 501(c)(4) named Secure America Now. Most Americans didn’t see the ads dripping with near-comedic xenophobia, but a few in swing states did, which was the intent. Facebook and Google coordinated to get those ads where they were directed, and helping underwrite it all was a cool $2 million from Robert Mercer.

Using Citizens United to unite citizens

In Mercer’s eyes, those on welfare are regarded as having negative value, but not cats, because they provide humans pleasure. And it only gets more uncomfortable from there: Mercer considers the Civil Rights Act of 1964 a major mistake, thinks white racism no longer exists, and considers climate change a hoax. He also really, really, really likes guns, and that includes playing dress up with them. His other hobbies include a $2.7 million dollar toy train set in his basement and financing a massive stockpile of urine in the mountains of Oregon. (Yep. The latter, it should be noted, has been with the help of chemist Arthur Robinson, founder of the Oregon Institute of Science and Medicine, who conducts research on the 14,000-and-counting specimens when not rejecting global warming or promoting intelligent design. Unknown if any pee pee vials have labels denoting surnames of Russian call girls, or — as The Intercept was quick to notice — what influence this may have had on naming the 2016 Mercer-funded Super PAC to get Trump elected: “Make America Number 1.”)

The bottom line for Robert Mercer is the bottom line: he and everyone at Renaissance made a lot of money, and continue to do so. And they are a creative lot, to be sure: Renaissance has succeeded (so far) in not paying $6.8 billion in taxes the IRS claims it owes for tens of millions of trades made in the 2000s by classifying them as long-term, and therefore lower-taxed, “basket options” it purchased through banks and held for at least a year. Renaissance appeared before the Senate in 2014, defending its position, and the banks named no longer engage in the practice of selling basket options to the firm. The IRS is steadfastly demanding the billions in back taxes be paid. It may not require a room full of quants or an artificial neural network to predict with reasonable accuracy how the Trump administration will view Robert Mercer’s plight with the taxman.

To conclude on a less-than-apocalyptic note regarding artificial intelligence and the power of social networks, when assessing all the gloom and doom a manipulative, power-hungry individual like Robert Mercer can wield deploying AI systems across any and all media, it’s important to recall how it was also just one person, a retired lawyer living in Hawaii, who thought “I should do something,” on November 9, 2016, and did. She used Facebook’s platform to instigate one of the finest examples of American democracy and global protest ever recorded: the 2017 Women’s March. Mercer and his ilk will never have the proper currency to purchase that away.

Driving Sentiment Metrics with AI was originally published in Hacker Noon on Medium, where people are continuing the conversation by highlighting and responding to this story.