Questions about Google autocompl
Last week, YouTube channel SourceFed released a video accusing Google of intentionally manipulating its autocomplete function in favor of Democratic presidential candidate Hillary Clinton.
Using screenshots comparing the autocomplete function to Google Search Trends, SourceFed journalist Matt Lieberman claimed that Google suppressed the autocomplete function for Hillary because the search engine did not autocomplete the words “criminal” or “crime” when used in conjunction with Clinton’s name. As further proof he offered other search term pairs for other candidates that could be perceived as offensive — for instance, “Bernie Sanders” and “socialist” (an affiliation about which Sanders has been fairly forthright) and “Donald Trump” and “racist.” The story quickly went viral. As of right now, the video has over 800,000 views on YouTube.
SourceFed’s thesis — that Google had been intentionally manipulating Google autocomplete for Clinton’s name — was, of course, wrong. Within hours, a number of commenters began weighing in on SourceFed’s flawed logic and method. Timothy B. Lee at Vox noted that the word “crimes” is never returned in autocomplete when paired alongside an individual’s name. Rhea Drysdale from Outspoken Media did her own investigation, finding that other potentially defamatory or libelous terms, such as “rape,” are also not included in a Google autocomplete (though specific terms like “email charges” or “lawsuit” did return results for presidential candidates). Over the weekend, Google responded to SourceFed’s claim in an email to the Washington Post explaining that the “Autocomplete algorithm will not show a predicted query that is offensive or disparaging when displayed in conjunction with a person’s name.”
Google has not, however, explained why their algorithm reflects this policy or who defines what is offensive or disparaging, though there are several potential explanations.
Google’s autocomplete algorithm has been tweaked to reflect company and local policies in the past. University of Maryland professor Nick Diakopolous has investigated how different search engines, like Google and Bing, have exercised this editorial power with regards to adult or offensive content. Google’s US autocomplete algorithm may also have been shaped by lawsuits around the world. As early as 2011, a court in Italy ruled that Google needed to filter out potentially libelous search suggestions. There have been several other cases where Google has been sued due to its autocomplete, in Germany, Hong Kong, Australia, and France (where Google was sued for allegedly promoting “unsolicited and systematic associations between famous people and their Jewishness”). In many of these cases, Google has been asked to block libelous words that appear next to names through the autocomplete function.
What the journalists from SourceFed may have stumbled upon was not an instance in which search results were intentionally being manipulated in favor of a candidate, but how algorithms can reflect complex jurisdictional issues and international policies that can, in turn, govern content. Though in the US Google is covered under Section 230 of the Communications Decency Act (which limits liability for online intermediaries that host, republish, or, in this case, aggregate speech), these other rulings may have led Google to feel the need to modify its autocomplete algorithm from showing potentially defamatory or libelous search term pairs. Internal decisions regarding what is ethical and socially acceptable, combined with a desire to please customers by abiding by certain social norms, also most likely play a significant role in determining what is or isn’t allowed.
What’s at stake then is not whether or not Google has policies that govern content, but how those policies are negotiated within the company and then communicated. Though the SourceFed journalists were off in their reasoning (a fact they themselves have not yet rectified honestly), they did point out something interesting: We’ve known that policies made in the US affect information flows abroad, but very seldom do we address how policies made abroad might affect the governance of algorithms here.
Points/spheres: What drives Google’s policy of “not show[ing] a predicted query that is offensive or disparaging when displayed in conjunction with a person’s name?” Good question! Robyn Caplan provides some context in the vein of the Who Controls the Public Sphere in the Era of Algorithms? workshop, which Data & Society hosted in February 2016 as part of our developing Algorithms and Publics project. More posts from workshop participants:
- Ben Franklin, the Post Office and the Digital Public Sphere by Ethan Zuckerman
- No More Magic Algorithms: Cultural Policy in an Era of Discoverability by Fenwick McKelvey
- Facebook Must Be Accountable to the Public by danah boyd
Robyn Caplan is the Algorithms and Publics project lead at Data & Society and a PhD Candidate at Rutgers University’s School of Communication and Information.
