Gender-based Algorithmic Bias
December 2019: We are compiling summaries of state-of-the-art research in ethics at the frontier of technology, following the theme of our 2019 Susilo Symposium. Today, we review insights on gender-based algorithmic bias from Professors Anja Lambrecht (London Business School) and Catherine Tucker (MIT).
Lambrecht and Tucker developed a field experiment to test how the Facebook algorithm delivered an ad promoting job opportunities in Science, Technology, Engineering and Math (STEM). The ad was tested in 190 countries and was explicitly intended to be gender-neutral in its delivery. However, the authors found that fewer women saw the Facebook ad than men.
The authors explored the reasons why the advertising algorithmic behaved in this biased way. They found that it was not due to engagement: if a woman ever saw the ad then she was actually more likely to click on it than a man was.
Instead, it turned out that the algorithm aimed at minimizing cost to the advertising organization. In online advertising, showing ads to different individuals is associated with different costs. It is more expensive to deliver an ad to a woman that a man, because women are more desirable to advertise to. This is because women control more household expenditures and are more likely to purchase after seeing an ad. Hence, the algorithm decided to show this ad to cheaper males rather than expensive females.
The authors argue that some of the commonly advocated policies, such as algorithm transparency, would not be effective in eliminating this bias in the display of ads. Looking into this algorithm would simply reveal that it was designed to minimize cost.
Reference
Lambrecht A and 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.