No Easy Explanation for Racial Bias Found in Google Ads
Feb 7, 2013 12:15 PM PT
Google ads appear to have a racial bias, according to a study conducted by a Harvard professor. Google AdSense ads relating to the word "arrest" tend to appear more often in the search results for names commonly identified with black people than for those more often associated with white people, Latanya Sweeney found.
The ads hint that the subject of the search -- or someone with the same name -- has been arrested. Sweeney searched 2,184 racially associated personal names across two websites in carrying out her research.
"Perhaps you are in competition for an award, an appointment, a promotion or a new job," Sweeney said in her study. "Appearing alongside your list of accomplishments is an advertisement implying you may have a criminal record, whether you actually have one or not. Worse, the ads don't appear for your competitors."
Instant Checkmate's Role
Instant Checkmate.com, a fee-required database of arrest records and criminal histories, had more ads on AdSense than its competitors, Sweeney found.
It was only the Instant Checkmate ads that included the word "arrest," she noted, while ads from the company's competitors did not.
Twenty-one percent of 432 Instant Checkmate ads on Google.com included an additional 10 templates, all using the word "criminal."
Ninety percent of Instant Checkmate ads on Google.com were suggestive of arrest regardless of race. Breaking this down, 92 percent of the 366 ads that appeared for black-identifying names and 80 percent of the 66 ads for white-identifying names were suggestive of arrest.
Targeted Marketing or Racial Profiling?
"At this point, we know the discrimination occurs," Sweeney told the E-Commerce Times. "We do not yet know why."
One possible explanation is that Instant Checkmate is conducting targeted marketing rather than racial profiling, and that society's inherent bias is at fault.
"There may be some way in which the reaction to those ads is perpetuating them," Greg Sterling, senior analyst at Opus Research, told the E-Commerce Times.
Such ads are "based on crude assumptions about race," he said, but "there's a very complicated cycle where preexisting stereotypes are reflected back to us by movies and ads. I'd say there are individuals behind the campaign that are making assumptions, and it's a reflection of society at large."
Google AdWords does not conduct any racial profiling, Google spokesperson Aaron Stein told the E-Commerce Times.
"We also have a policy which states that we will not allow ads that advocate against an organization, person or group of people," he continued. "It is up to individual advertisers to decide which keywords they want to choose to trigger their ads."
Sweeney used lists from prior studies that employed black-identifying and white-identifying names, adding two black female names, "Latanya" and "Latisha," from her earlier observations. A total of 63 distinct first names were culled from those lists.
She then harvested the Web to get the last names of real people who had one of the 63 first names selected earlier. In all, she came up with 2,184 racially associated full names of people with an online presence. Using images associated with those names, Sweeney confirmed that the racially associated first names initially obtained were predictive of race.
She conducted searches on Google.com and Reuters.com for each full name and recorded which ads were displayed with the results. The searches were conducted at different dates and times of the day, using different IP addresses and browsers at several U.S. locations.
In brief, this is how AdSense works: Advertisers provide Google with search criteria, copies of possible ads to deliver when they match a search, and bids on the ad. Google operates a real-time auction across bids for the same search criteria. It places ads next to search results in descending order starting with those with the highest bids.
It's not clear whether the perceived racial bias can be laid at Instant Checkmate's door or at Google's, or if society at large is to be blamed, study author Sweeney said. Advertisers serve up multiple variations for a search string, and Google's algorithm selects the one that gets the most clicks from viewers.
Google's cloud caching strategies to deliver ads quickly may also bias ad delivery toward templates previously loaded in the cache.