A study conducted by researchers at Carnegie Mellon University has found that women are exposed to significantly less online advertisements promoting high paying jobs than their male counterparts, raising questions about the fairness of tech giant’s ad targeting program.
The study of Google ads was done using AdFisher, a CMU-developed tool that runs experiments with simulated profiles. While the program established that gender discrimination was definitely happening, what’s unknown is who or what is responsible. Specifically — is it a preference of the advertisers to recruit men for these jobs? Or is this just simply an unintended consequence of machine learning algorithms driving online recommendation engines down the wrong path?
“This just came out of the blue,” said Anupam Datta, associate professor of computer science and of electrical and computer engineering, in reference to the gender discrimination finding. He went on to explain that they were using AdFisher as part of a larger study on the operation of Google's Ad Settings Web page. “Many important decisions about the ads we see are being made by online systems,” he explained. “Oversight of these 'black boxes' is necessary to make sure they don't compromise our values.”
AdFisher ran a total of 21 experiments evaluating Ad Settings. Once the gender discrimination trend was noticed, researchers used the CMU program to create 1,000 simulated users, half of which were designated males, and the other half female. They then had these profiles visit 100 top employment sites, after which AdFisher reviewed and recorded the ads that were shown to the simulated users.
The site most strongly associated with the male profiles was a career coaching site service for executive positions paying more than $200,000.
“The male users were shown the high-paying job ads about 1,800 times, compared to female users who saw those ads about 300 times,” said Amit Datta, a Ph.D. student in electrical and computer engineering.
Worth pointing out: the ads most associated with female profiles were for a generic job posting service and an auto dealer.
Now, it should be noted that the researchers did not find any evidence pointing to wrongdoing on Google’s end. While AdFisher can identify discrepancies, it cannot explain why they occur without looking inside the black box.
Another experiment showed that some changes in ads presented to users based on their browsing activity are not clearly explained on Google’s Ad Settings page; that is, when one of the simulated users visited a web site associated with substance abuse, Google began sending them more ads for a drug rehabilitation center. This change was not reflected in Ad Settings, thereby making it all but impossible for the user to change preferences and stop the substance abuse-related ads.
Datta explained that this is likely because the change in ads was the consequence of remarketing; that is, Google allows advertisers to reach users who’ve already visited their page. What AdFisher found was that Ad Settings doesn’t provide a complete picture of the inferences Google has made about the use (Datta notes that Google started highlighting this limitation a few weeks ago, and is working to adjust it).
One other thing worth pointing out from the experiments — by adjusting the Ad Settings, users are able avoid some classes of ads they might dislike. The team found that if the simulated users visited online dating sites, they were able to remove that interest from Ad Settings to reduce their exposure to online dating ads, thereby giving users some choice over the ads Google shows them.
The entire study was published in the journal, Proceedings on Privacy Enhancing Technologies, and it is entitled Automated Experiments on Ad Privacy Settings: A Tale of Opacity, Choice, and Discrimination.
Via Carnegie Mellon University
Advertisement
Learn more about Electronic Products Magazine
