Americans are rightly concerned that Big Tech companies in Silicon Valley have an outsized influence over the news and opinions we see on social media platforms and in Google searches. The fact that most of these platforms originate in some of the most liberal strongholds in the country — Mountain View, San Francisco, and Menlo Park, California — raises questions about whether they're either intentionally or subconsciously putting their oversized thumbs on the scales to promote points of view they agree with and hide views they find odious or dangerous. A study released today from GovPredict shows that more than 90 percent of political donations by Alphabet employees went to Democrats. This news will only amplify fears that everything we read, see, and hear is being controlled by tech industry employees with a left-wing political bent.
In the first of a series of articles examining the political preferences of major American companies, GovPredict looked at the political donations of employees at Alphabet — the parent company of Google and many of its subsidiaries, including YouTube, Nest, Google Ventures, Calico, Adsense, Google Ventures, and Verily. The analysis used Federal Election Commission (FEC) data on contributions to federal candidates and causes.
"The question is simple: what are the political preferences of Alphabet employees, as revealed by their political giving histories, and how have these preferences evolved over time?" GovPredict explained in the report.
"Our analysts and machines first had to identify the variants of employer name that Alphabet employees used when filing election contributions," GovPredict said. "The final list had 233 variants, including 'Google Ventures,' 'Nest Labs,' 'Nest at Google,' 'Verily (Google Life Sciences),' and the like." The researchers also had to categorize as either Democrat or Republican "the 1,105 unique committees to which Alphabet employees have contributed over the past decade and a half."
The majority of the party tags were supplied by the FEC; others were categorized by hand. "Organizations that might not explicitly identify with a political party but which ideologically are synchronized were issued with a party label," they said. For example, the League of Conservation Voters Action Fund was categorized as a Democratic cause. In another example, "A contribution in 2007 to Arlen Specter was categorized as a contribution to a Republican, since he changed party affiliation in 2009."
The findings were astounding, but not at all surprising to those of us who have been paying attention to this issue: "Since 2004, Alphabet employees have contributed a little over 90% of their political dollars to Democratic candidates and causes," GovPredict discovered.
(Credit: GovPredict)More than $15 million in Alphabet employee political contributions went to Democrat candidates and causes while a mere $1.6 million went to Republicans between 20014 and 2017. The largest donation disparity came in 2016 when 94 percent of contributions went to Democrats — $5.8 million vs. $403,000, suggesting a strong reaction to President Trump's election that prompted a flurry of donations.
Google has insisted over and over that they're not playing favorites. Maggie Shiels from Google's corporate communications department recently insisted to PJM that "Google does not manipulate results."
PJM's Charlie Martin recently explained how human factors could skew an algorithm that determines what we see and don't see.
An algorithm is nothing more than a procedure — a series of steps that lead to a result. The word is mostly used with reference to computer programs, but not necessarily — the way you learned to do long division is an algorithm.
The specific algorithms that are used come from the category of "machine learning" or more broadly "artificial intelligence." These phrases sound science-fictional and cool but the reality is that all of these are doing something conceptually simple: the programs get inputs, process them in various ways, and present them to a person who says "you're getting warmer" or "you're getting colder." This trains the program to get warmer as often as possible.
The potential weak spot here is the person in the loop. Imagine you're Twitter and you have a machine learning algorithm you're training to identify Nazis on Twitter. You put a person in the loop who thinks Trump is a Nazi and anyone who says anything favorable about Trump is a Nazi sympathizer. (They exist: I lost a couple of friends when they called me a Nazi sympathizer for just that reason.)
The algorithm spots someone liking the tax cuts: the person says, "he's a Nazi." The algorithm soberly notes that. It doesn't know any better, it has no more understanding than an old-fashioned tabulating machine understood why it put the A and B cards into different bins.
Endorsed Brett Kavanagh? "He's a Nazi." In Congress with an (R) after your name? "Oh yeah, definite Nazi."
Pretty quickly the algorithm will confidently identify any Republican, any Trump fan, or any independent who says #MAGA, as a Nazi.