This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
Independent audits of Facebook's ad-delivery system revealed that its algorithms were skewing the delivery of housing and employment advertisements along gender and racial lines. Despite Facebook's claims of taking steps to address discrimination, researchers found that the system continued to reproduce demographic skews even when advertisers used neutral targeting parameters. This behavior resulted in disparate exposure to job and housing opportunities, raising significant concerns regarding algorithmic fairness and compliance with equal employment opportunity laws.
Facebook's housing and employment ad delivery process allegedly resulted in skews in exposure for some users along demographic lines such as gender and racial identity.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
Unequal treatment of individuals or groups by AI, often based on race, gender, or other sensitive characteristics, resulting in unfair outcomes and unfair representation of those groups.
Also classified as
The ad-delivery system failed to perform reliably and fairly under neutral targeting conditions, exposing it to discriminatory errors.
The ad-delivery system systematically treated individuals unequally by withholding job and housing advertisements based on gender and race, resulting in unfair outcomes.
AI system
Due to a decision or action made by an AI system
Unintentional
Due to an unexpected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed
The discriminatory skew was an unexpected and unintended outcome of Facebook's ad-delivery algorithms optimizing for relevance and engagement after deployment.
Independent audits confirmed that Facebook's autonomous ad-delivery algorithms systematically discriminated by gender and race in housing and employment advertisements. While representing a significant civil rights concern and violating US equal opportunity laws, the incident poses minor national security risk and is manageable through standard legal and regulatory frameworks.
The report describes Facebook's ad-delivery system making real-time decisions about which users see which ads based on 'market and financial optimization effects as well as the platform's own predictions about the relevance of ads'. This represents autonomous operation based on learned patterns without explicit human planning for each ad delivery.
High Risk: The system is used for 'Employment and worker management systems, such as AI used in recruitment' and housing, which have significant implications for fundamental rights and access to livelihoods. The report explicitly details how the system delivers 'employment and housing opportunities' differentially.
The primary function of the AI system is to deliver targeted advertisements to users on Facebook's platform, optimizing for relevance and engagement.
The reports are based on rigorous, peer-reviewed independent academic audits from USC and Northeastern University, which controlled for qualifications and demonstrated clear statistical evidence of algorithmic skew. The findings are highly consistent across multiple audits and legal actions (HUD lawsuit).
None that are plausible, as the audits carefully controlled for qualifications to rule out non-discriminatory explanations.
Technology
Algorithmic Relevance Optimization
System optimizes for predicted relevance, reproducing historical gender skews.
Engagement-Driven Delivery
Algorithms prioritize user engagement, skewing ad delivery along gender lines.
Market Optimization Effects
Financial and market dynamics in ad delivery auctions skew demographics.
Data Inputs
Biased Historical Data
Training algorithms on historical job data that reflects existing gender skews.
Lack of Qualification Data
Algorithms lack explicit qualification data, using demographics as proxies.
Human Factors
User Engagement Bias
Historical user click patterns reinforce algorithmic bias during delivery.
Overreliance on Neutral Parameters
Advertisers assume neutral settings prevent bias, unaware of delivery skew.
Process and Methods
Black-Box Ad Delivery
Lack of transparency in how algorithms work prevents external auditing.
Deficient Auditing Methods
Internal testing failed to detect non-qualification gender skew in delivery.
Management
Prioritizing Growth Over Safety
Facebook prioritized engagement and revenue over resolving algorithmic bias.
Neglecting Algorithmic Bias
Responsible AI team neglected delivery bias due to narrow focus.
Lack of Political Will
Management refused to disable algorithmic targeting for sensitive categories.
Regulatory Environment
Weak Enforcement of Laws
Regulatory bodies slow to enforce equal opportunity laws on digital platforms.
Lack of Clear Standards
Absence of clear legal standards for algorithmic bias in ad delivery.
AI system involved
Identified by automated extraction; can differ from the alleged developers above.
