As algorithmic decision-making touches more and more aspects of our lives, questions about the underlying rules that route these bits and bytes — and ultimately determine the digital content and opportunities we are exposed to — are growing.
A study by researchers from Carnegie Mellon University and the International Computer Science Institute, reported this week in the MIT Technology Review, has highlighted potential gender bias in Google’s ad-targeting algorithms. The researchers found that male job seekers were more likely than equivalent female job seekers to be shown ads for high-paying executive jobs when they visited a news website.
The researchers used a tool of their own making, called AdFisher, to gain intel on how Google’s ad-targeting works, recording the ads served by Google on third party websites after created scores of carefully curated web user profiles — giving them a benchmark to compare how ads were being served to users based on their interests and gender.
However they caution it’s difficult to definitively determine how ads are being targeted because the process is so complex and opaque. Ad buyers can make demographic decisions about the targeting themselves and also incorporate their own data sources on people’s online behavior to do additional targeting for certain types of ads. So where exactly any gender bias in ad serving is coming from is difficult to determine without more data.
The researchers write:
We cannot determine who caused these findings due to our limited visibility into the ad ecosystem, which includes Google, advertisers, websites, and users. Nevertheless, these results can form the starting point for deeper investigations by either the companies themselves or by regulatory bodies.
Another finding was that Google’s ad settings transparency tool, which lets users view and edit the interests the company has inferred for them (based on tracking their digital behavior), does not always reflect potentially sensitive information that is being used to target them with ads.
“Browsing sites aimed at people with substance abuse problems, for example, triggered a rash of ads for rehab programs, but there was no change to Google’s transparency page,” reports the MIT Technology Review. It goes on to quote Anupam Datta, an associate professor at Carnegie Mellon University who helped develop the AdFisher tool, criticizing the lack of transparency around ad targeting.
“I think our findings suggest that there are parts of the ad ecosystem where kinds of discrimination are beginning to emerge and there is a lack of transparency. This is concerning from a societal standpoint,” said Datta.
Asked for its response to the research, Google provided the following statement to TechCrunch: “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 Ad Settings, as well as the ability to opt out of interest-based ads.”
Google’s policy for interest-based advertising apparently includes restrictions on advertisers targeting based on a variety of “sensitive” categories, such as health and medical, sexual orientation and negative financial status. However, based on the research findings, it appears that substance abuse data is being used to target Google users with ads for rehab clinics — which surely strays into the health/medical category targeting exclusion. So it’s possible advertisers are relying on their own data for such sensitive targeting.
The researchers note that since being contacted by them about their research Google has added a disclaimer to its ad settings page to note that the information shown on the page reflects only “some of the Google ads that you see”. Datta argues that limits the usefulness of the transparency tool, adding that he believes Google could reveal a fuller picture around interest-based ad-targeting if it chose to. “They are serving these ads, and if they wanted to they could reflect these interests,” he said.
Other issues that might be influencing ad targeting could include aspects such as the relative value of different demographics and genders to advertisers — and the resulting volume of ad inventory to serve each. The sites where ads are being served are another influencing variable.