Discrimination through optimization: How Facebook's ad delivery can lead to skewed outcomes (1904.02095v5)

Abstract: The enormous financial success of online advertising platforms is partially due to the precise targeting features they offer. Although researchers and journalists have found many ways that advertisers can target---or exclude---particular groups of users seeing their ads, comparatively little attention has been paid to the implications of the platform's ad delivery process, comprised of the platform's choices about which users see which ads. It has been hypothesized that this process can "skew" ad delivery in ways that the advertisers do not intend, making some users less likely than others to see particular ads based on their demographic characteristics. In this paper, we demonstrate that such skewed delivery occurs on Facebook, due to market and financial optimization effects as well as the platform's own predictions about the "relevance" of ads to different groups of users. We find that both the advertiser's budget and the content of the ad each significantly contribute to the skew of Facebook's ad delivery. Critically, we observe significant skew in delivery along gender and racial lines for "real" ads for employment and housing opportunities despite neutral targeting parameters. Our results demonstrate previously unknown mechanisms that can lead to potentially discriminatory ad delivery, even when advertisers set their targeting parameters to be highly inclusive. This underscores the need for policymakers and platforms to carefully consider the role of the ad delivery optimization run by ad platforms themselves---and not just the targeting choices of advertisers---in preventing discrimination in digital advertising.

Analysis of Discriminatory Dynamics in Facebook's Ad Delivery

The paper "Discrimination through Optimization" explores the complex and often opaque mechanisms of digital advertising platforms, specifically focusing on how the ad delivery process can inadvertently result in discriminatory outcomes. Through an empirical paper on Facebook's advertising platform, the authors provide a critical analysis of how both market effects and algorithmic optimizations during ad delivery contribute to unintended biases along demographic lines such as race and gender.

Key Findings

1. Market Effects and Budget Influence: It is observed that market dynamics significantly affect ad delivery. For instance, ads with lower budgets tend to reach a disproportionately male audience. This is attributed to economic factors where certain demographics become more expensive to target, thereby skewing reach when constrained by budget limitations.
2. Ad Creative Impact: The paper reveals that the content of the ad itself (headline, text, image) plays a pivotal role in determining the demographics of the ad audience. Ads stereotyping gender interests, such as those related to cosmetics or bodybuilding, exhibited stark gender biases in their delivery.
3. Automated Content Classification: The authors identify that Facebook's platform likely employs automated image classification, influencing initial ad delivery skew. This is exemplified by experiments using invisible but classifiable image data leading to pronounced skew in audience demographics from the onset.
4. Observations in Employment and Housing Ads: Real-world applications of these insights were demonstrated through employment and housing advertisements. Despite identical audience targeting parameters, substantial skew was observed, with ads showing bias along gender and racial lines due to the content of the ad creative alone.

Implications

The implications of these findings are multifaceted, impacting both the practical and theoretical domains of digital advertising systems:

• Regulatory Considerations: The research underscores the necessity for regulatory bodies to scrutinize not just the targeting features available to advertisers, but also the delivery algorithms that platforms utilize. These algorithms can autonomously create discriminatory outcomes, independent of advertisers' intentions.
• Ad Platform Accountability: The paper questions the applicability of existing legal protections, such as Section 230 of the U.S. Communications Decency Act, which shields platforms from liability for third-party content. Given that platforms play an active role in shaping ad delivery outcomes, these legal frameworks might require reevaluation.
• Transparent Reporting: The results highlight the pressing need for increased transparency in advertising platforms' operations. Current ad transparency tools do not provide sufficient granularity for researchers to dissect the mechanisms leading to skewed delivery, especially regarding protected categories.

Future Directions

This research opens several avenues for future investigation and development:

• Algorithmic Fairness: There is a need for developing algorithmic techniques that reconcile platform optimization goals with fairness criteria, possibly drawing from notions like individual fairness or preference-informed fairness.
• Interdisciplinary Exploration: Further exploration in collaboration with legal scholars could illuminate the intersection between digital advertising practices and anti-discrimination laws.
• Expanded Research Scope: Extending this analysis across different platforms and broader demographic attributes could provide a more comprehensive understanding of ad delivery biases in digital ecosystems.

In conclusion, this paper provides a compelling examination of how ad delivery systems, through their inherent market and algorithmic optimizations, contribute to unintended discrimination. This calls for robust discussions among technologists, policymakers, and legal experts to develop frameworks and innovations that ensure equity in digital advertising.