While advertisements are an everyday experience for Internet users, how Google and other companies convert personal web activity to serve ads remains secret. A recent study unlocked some unsettling revelations about Google’s ad-targeting algorithms — that they may be discriminating between Internet users.
Men browsing job sites were more frequently shown ads with higher salaries than were women, say researchers from Carnegie Mellon University and the International Computer Science Institute (ICSI).
The researchers built AdFisher, a custom that simulates the browsing activities of web users, and discovered the differences between ads served to fake male and female users.
The same study also found users were shown ads for rehab program if they had visited sites related to substance abuse. According to Google’s transparency tool called “ad settings,” users are able to view and edit the interests the company has predicted for the user.
However, those transparency settings do not always reflect the actual and potentially sensitive information that is being used to target web users. Therefore, the option to edit one’s profile may only offer an illusion of control over web ads.
The researchers offer an unsettling explanation for the study’s results. “We cannot claim that Google has violated its policies,” the researchers told Verge. “In fact, we consider it more likely that Google has lost control over its massive, automated advertising system.” They add that “even without advertisers placing inappropriate bids, large-scale machine learning can behave in unexpected ways.”
Responding to the study results, Google’s Andrea Faville told the MIT Technology Review:
“Advertisers can choose to target the audience they want to reach, and we have policies that guide the type of interest-based ads that are allowed. We provide transparency to users with ‘Why This Ad’ notices and Ads Settings, as well as the ability to opt out of interest-based ads.”
Other recent instances of algorithmic discrimination have made headlines. Google apologized after computer programmer Jacky Alciné posted pictures on Twitter showing proof that the new Photos app tagged his African-American girlfriends as “gorillas.”
In April, researchers found that Google images search for “CEO” returned only 11 percent female images, but in real life, 27 percent of a U.S. chief executives are female.
While Google refuses to offer authentic transparency, there seem to be cracks in the system, which enable researchers and citizens to peer behind the curtain.
