Can an advertising algorithm lead to gender-based discrimination?

The answer is "Yes," even when no bias was intended.

WHAT YOU NEED TO KNOW

Computer algorithms are used to automate decision-making in an increasing number of areas, including the delivery and distribution of advertising. However, there are concerns that algorithmic decision making could lead to outcomes that may disadvantage certain populations, especially minority groups. So what if an algorithm exhibits say, a gender-based bias? Algorithmic bias is troubling because it raises the potential that an algorithm could make decisions that reinforce bias that already exists in society. Authors Anja Lambrecht and Catherine Tucker investigated the potential for algorithmic bias in the display of ads promoting careers in STEM (science, technology, engineering and math). They found that the advertising algorithm showed the ads to 20 per cent more men than women, and this disparity was worse for people in younger age ranges starting out on their careers.

MORE DETAILS

What Did the Researchers Do?

Authors Lambrecht and Tucker performed a field experiment with millions of ad impressions, where the underrepresented group were women in STEM. The study ran ads on multiple tech platforms including popular social media and search platforms like Facebook, Instagram, Google and Twitter. The ad promoted careers in the STEM sector and linked to a website providing information on this topic. The authors set up ad campaigns to run across 191 countries with ads that were designed to be gender-neutral, however the authors were able to see the share of male and female users to which the ads were shown.

What Did the Researchers Find?

Their results showed that STEM ads were shown to 20 per cent less women than men, more so if they were young. Further investigation revealed that this wasn’t because women click less on the ads, use social media less, or because the algorithm learned biases from revealed preferences in each country. Instead, it happened because advertising algorithms optimize the display of ads, given the cost of showing ads to specific consumers. It turns out that females – especially young females between 25-34 – are a prized demographic by many advertisers. This means that it is more costly for any advertiser, including the researchers’ campaigns, to show ads to this group. Put differently, it was cheaper for the campaign to show the ad to male eyeballs than to female eyeballs. In sum, this study shows that even though algorithms may not be designed to be discriminatory, they may yet end up discriminating in the pursuit of a non-discriminatory goal.

WANT TO KNOW MORE?

Contact: Catherine Tucker; cetucker@mit.edu

Lambrecht, A., & Tucker, C. (2019). Algorithmic bias? an empirical study of apparent gender-based discrimination in the display of stem career ads. Management Science, 65(7), 2966-2981.