Facebook algorithm serves racial and sexist bias in job and real estate ads
Facebook automatically discriminates in job and real estate ads. This is the result of a study published in advance on "arXiv" by researchers from Northeastern University, the University of Southern California and the non-profit organization.
Even if advertisers actually want to address a broad audience with their targeting parameters, Facebook's algorithm seems to be geared towards racist and sexist stereotypes when placing ads.
Lumberjacks are white men, only women sit at the supermarket checkout and cab drivers belong to minorities anyway. These are classic stereotypes that Facebook's ad targeting apparently serves automatically. As the researchers report, the social network displayed ads with targeting options that were actually the same to extremely different users in tests. A job in the timber industry was suggested to 72 percent of white people, and nine out of ten were men. On the other hand, 72 percent of women were shown an ad for a checkout job in a supermarket and three quarters of black people were shown jobs in the cab industry.
This difference therefore appears to be due to the content of the ad, i.e. that the algorithm automatically shows a racial or gender bias based on text elements such as the job title or image properties. The latter is also supported by the fact that, according to the study, a housing offer is automatically shown less often to white users if the photo simply shows a family of color instead of a white family. However, the advertising budget also seems to play a role. In tests, more money resulted in a higher proportion of women in the audience reached - over 55% with a very high budget compared to under 45% with a low budget.
If Facebook's advertising algorithm really does automatically discriminate, this could ultimately lead to legal problems. Because as the magazine New York emphasizes, non-inclusive job and real estate ads are illegal in the USA. For Facebook, however, another problem may be much more immediate. "Our results reveal previously unknown mechanisms that can lead to potentially discriminatory ad delivery, even when advertisers have chosen their targeting parameters to be highly inclusive," the researchers write. If advertisers felt their interests were at risk, this could jeopardize advertising revenues. (pte)
To the paper: "Discrimination through optimization: How Facebook's ad delivery can lead to skewed outcomes"
