CS Colloquium - Discrimination through optimization: Measuring fairness and bias in social network ad delivery

CS Colloquium - Discrimination through optimization: Measuring fairness and bias in social network ad delivery promotional image


Alan Mislove


The enormous financial success of online advertising platforms is partially due to the precise targeting and delivery features they offer. These platforms routinely claim to be able to identify which platform users are most likely to meet advertisers’ objectives, and steer ads towards those users via their ad delivery algorithms. This is typically accomplished by estimating the “relevance” of ads to users, but raises a number of concerns about bias, discrimination, and impacts on historically disadvantaged groups and society as a whole. Unfortunately, the manner in which platforms calculate such relevance estimates is opaque and challenging to study, and platforms are often resistant to sharing information publicly.

In this talk, I discuss work my group has done to address this challenge. I first develop a measurement methodology using Facebook’s advertiser interface that can measure the influence of Facebook’s choices about how to deliver ads. I then demonstrate that ad delivery can be significantly skewed by protected classes on Facebook, due to the platform’s own predictions about the “relevance” of ads to different groups of users. I show significant skew in delivery along gender and racial lines for “real” ads for employment and housing opportunities despite neutral targeting parameters. I then describe two additional contexts in which we have studied how ad delivery algorithms make decisions and are impacting users: how Facebook chooses to delivery ads with political content, and how Facebook delivers ads with images of people from different racial, gender, and age groups. Overall, our findings demonstrate previously unknown mechanisms that can lead to potentially discriminatory ad delivery, even when advertisers set their targeting parameters to be highly inclusive.


Alan Mislove is a Professor and Senior Associate Dean for Academic Affairs at the Khoury College of Computer Sciences at Northeastern University, which he joined in 2009. He received his B.A., M.S., and Ph.D. in computer science from Rice University in 2002, 2005, and 2009, respectively.

Prof. Mislove's research is on algorithmic auditing.  He develops methodologies and study the real-world systems that millions of users interact with every day, focusing on issues of algorithmic discrimination, fairness, and privacy.  His aim is to enable regulators, policymakers, and society at large to better understand how these systems work, how they are used and abused, and what impacts they are having on end users and society at large.  He is a recipient of an NSF CAREER Award (2011), a Google Faculty Award (2012), a Facebook Secure the Internet grant (2018), the ACM SIGCOMM Test of Time Award (2017), the IETF Applied Networking Research Prize (2018, 2019), the USENIX Security Distinguished Paper Award (2017), the NDSS Distinguished Paper Award (2018), the IEEE Cybersecurity Award for Innovation (2017), and a Facebook Secure the Internet Grant. He has testified before Congress and his work has been covered by the Wall Street Journal, the New York Times, and the CBS Evening News.

Friday, October 14, 2022 4:00pm to 5:00pm
Seamans Center
103 South Capitol Street, Iowa City, IA 52240
View on Event Calendar
Individuals with disabilities are encouraged to attend all University of Iowa–sponsored events. If you are a person with a disability who requires a reasonable accommodation in order to participate in this program, please contact Computer Science Dept. in advance at 319-335-0713 or matthieu-biger@uiowa.edu.