23 December 2018

The Search for Anti-Conservative Bias on Google

By Sue Halpern

Last Wednesday, a day after Google’s C.E.O., Sundar Pichai, sat before the House Judiciary Committee to answer questions about the company’s search engine, Donald Trump’s 2020 campaign manager, Brad Parscale, wrote a post on Twitter that ended with the hashtag #StopTheBias. Parscale, the digital director of Trump’s 2016 campaign, is considered a master of online marketing—the nearly ninety million dollars that he spent on Facebook, YouTube, Instagram, and other digital media is widely credited for playing a major part in Trump’s unlikely victory—so it might have seemed strange that he was complaining about Silicon Valley’s anti-conservative bias. “They make bias decisions all the time to tweak search results when Democrats are damaged from results,” Parscale tweeted about Google, “but leave negative Republican results because ‘that’s how it works.’ Bias is in their corporate DNA.” It’s a claim we’re going to hear a lot in the months to come.

During the hearing, Republican members of Congress kept coming back to this question of anti-conservative bias, only it wasn’t much of a question. Steve Chabot, a Republican congressman from Ohio, explained that, when he searched for information about the Republican health-care bill or tax cuts, the articles that came up first were all negative. “How do you explain this apparent bias on Google’s part against conservative points of view, against conservative policies?” he asked Pichai. Lamar Smith, a Republican from Texas, also seemed certain that Google was gaming the results against conservatives, telling Pichai at one point, “Those who write the algorithms get the results they must want. Apparently management allows it.” Meanwhile, Zoe Lofgren, the Democratic congresswoman from California, who was fully aware, as she said, that “it’s not some little man sitting behind the curtain figuring out what we’re going to show the user,” slyly asked why photographs of Donald Trump were returned when one Googled the word “idiot.”

Pichai, who grew up in India before coming to the United States for graduate school at Stanford and later at Wharton, has worked for Google since 2004, when it was fielding about forty-four per cent of Internet search queries. Google’s market share is now ninety-two per cent, which has led some lawmakers, including President Trump, to invoke the specter of antitrust regulation, which may have had something to do with Pichai’s appearance in Congress. He had noticeably skipped the Senate Intelligence Committee hearing in September that was attended by Twitter’s C.E.O., Jack Dorsey, and Facebook’s C.O.O., Sheryl Sandberg, which happened to be held shortly after Trump himself called out Google for anti-conservative bias. In his opening remarks last Tuesday, the congressman Kevin McCarthy, who is the incoming House Minority Leader, offered a pleasantly snide welcome, thanking Pichai for “finally testifying on Capitol Hill.”

During the hearing, as members of Congress spent their five-minute speaking allotments mostly asking somewhat inane questions—like when the congressman Steve King, a Republican from Iowa, wondered why his seven-year-old granddaughter’s iPhone was acting strangely, and Pichai had to explain that his company does not make iPhones—Pichai sat attentively, betraying no annoyance, even after repeating, in various ways, that Google’s search algorithm did not discriminate against one party or the other. “Our products work without any bias,” he said. “We build our products in a neutral way.”

This was to be expected. Algorithmic neutrality is a common Silicon Valley refrain. But an algorithm built without favoring one political party or another, or constructed without intentionally championing a particular ideology, is actually designed to deliver culturally biased results. A search engine runs on algorithms and artificial intelligence to instantaneously sift through the Internet’s nearly two billion Web sites. Google’s engineers have embedded something they call “authoritativeness” into their search algorithm to deliver its results, though what this is, exactly, is challenging to understand, because it appears to be based on a tautology: an authoritative source is a source that a lot of other sources consider to be authoritative. When Representative Chabot searched Google for information about Republican tax cuts, for example, and the top results were all negative, it was not because Pichai or others at Google were opposed to tax cuts—it’s because, in part, a lot of people were interested in reading critiques of the Republican plan. A Google spokesperson told me that there are limited circumstances in which Google would remove anything; for the most part, the company’s philosophy is, if it’s legal, there is value in being able to see it.

It is also true that search results can be gamed through search-engine optimization to push sites to the top of the queue, as both the Observer and the Guardian reported two years ago, when they found Google prioritizing anti-gay, anti-Semitic, anti-science, and alt-right search results. “The key has traditionally been connected to influencing the algorithm with a high volume of biased search terms,” Brittan Heller, of the Anti-Defamation League, said at the time. Safiya Noble, the author of “Algorithms of Oppression” and a professor at U.S.C., discovered something similar in her research as well. Through six years of studying Google search results, looking at the ways they obliquely embed and promote racism, she found that people of color “were the most likely to have their identities co-opted and misrepresented by pornography or sexually graphic content, or by other stereotypical, racist, or sexist images or ideas,” she told me in an e-mail. The reason, she added, is that “Google search is a search tool that is subsidized by advertising. Those with the most capital and technical skill in optimizing content are the most likely [to have] their content be more visible on the first few pages of search results. For this reason, it’s important that people understand that vulnerable communities are often the least likely to have the resources to purchase keywords and least likely to technically optimize content in their own interests.”

Google’s search engine is proprietary, and the company is very careful to insure that it remains, in the words of the University of Maryland law professor Frank Pasquale, “a black box.” In my own conversations with Google employees, I was told that location and “freshness” are both crucial to a search result, that the title of an article is more important than the words in the article, that personalization—culled from one’s search history—has a minor effect when doing a general search and none at all when searching for news, and that the company welcomes “white hat” search-engine optimization efforts that help pages rise to the top of the results. But these factors are among some two hundred that determine what appears at the top of a search result—and the nature of these hundreds of factors, including what they are are, how they are valued, and on what basis, remains a mystery. “That, to me, is the most important problem,” Pasquale told me. “We really have very little sense of how the decisions are made, who’s responsible, or how they’re reasoning about these things. They’re almost like a government. You know, they’re a very, very, powerful company making very powerful decisions.”

Google’s enormous power seemed to be the only thing that the representatives of our actual government agreed upon. Even so, they squandered the opportunity to grill Pichai about Google’s leaky data-protection practices—and this was a day after Google announced that it was expediting the shutdown of its social-networking site, Google+, after revealing that the personal data of 52.5 million users was inadvertently exposed to third-party app developers between 2015 and 2018. It was a breach that the company—with Pichai’s blessing—chose not to report to consumers when it was first discovered, apparently out of fear of alerting regulators. And only one member of Congress, Jamie Raskin, a Democrat from Maryland, pointed out that YouTube has been co-opted by white supremacists and hate groups. They use YouTube not only to spread their messages and recruit members but to make money through ads—money that then funds those groups. Because these videos are widely shared within the alt-right ecosphere, they also spill over to search, as I found out when I typed the words “Hillary Clinton pedophile” into the Google search bar, and a YouTube video that has been seen over 65,500 times, titled “Pedophile Hillary Clinton to be Arrested Before Nov 11, 2018,” was one of the top results.

It is, perhaps, unsurprising that none of the Congress members complaining about Google’s anti-conservative bias appeared overly concerned that hate groups had hijacked YouTube. Instead, they kept hammering at the bias they claimed the company was directing against them.This is because, as Pasquale told me, the Republicans are very good at “working the refs” to get what they want. What they want here is to bring Google to heel, as they’ve done to Facebook, which conspicuously hired the conservative politician Jon Kyl to investigate its anti-conservative bias, and added an avowedly conservative publication, the now defunct Weekly Standard, to its fact-checking team, giving the magazine’s staff the ability to down-rank sources with which it disagreed ideologically.

But the #StopTheBias campaign has a more pernicious goal: it is yet another way for Trump and his minions to undermine the credibility of the mainstream media. In August, when the President issued a series of tweets claiming that Google search was biased against conservative sources, it was reported that he was getting his “facts” from a chart published by a right-wing news blog, PJ Media. The chart divides the American media landscape into outlets on the left—the New York Times, NBC News, Variety, the Associated Press, and the New York Daily News, among them—and those on the right, including Fox News, the New York Post, Breitbart, Infowars, and the Daily Caller. Parsed this way, a search result that leads with the Times and the A.P. can be called left-leaning. It’s a distortion wrapped in a redundancy. It also pretends an equivalence between media based on reporting and media based on dogma.

But here’s the thing: that PJ Media chart is a bastardization of the “Media Bias Chart” compiled by the nonpartisan company Ad Fontes Media. And that chart places the A.P., ABC, and CBS squarely in the middle—which is to say that they demonstrate no bias—and the New York Times and CNN so close to the middle that their liberal predilections only register marginally. In contrast, the chart places Breitbart and Infowars as far from the middle as possible, in a category called “Most Extreme Right,” and characterizes both as “Nonsense, damaging to public discourse.” It also describes PJ Media itself as “hyper-partisan right.”

The distinctions between these two charts became especially salient the day before Pichai’s trip to the Hill. To illustrate Google’s apparent anti-conservative bias and give #StopTheBias more oxygen, Breitbart published internal Google e-mails from 2016 that showed Google employees, concerned that Breitbart was a source of hate speech and fake news, discussing the possibility of removing the site from its advertising program, which would cut off a source of Breitbart’s revenue. It should tell us everything we need to know about Google’s anti-conservative bias that they decided not to do so.

No comments: