Facebook's ad delivery algorithm systematically showed job advertisements predominantly to one gender despite advertisers not specifying gender targeting, with Global Witness testing revealing discriminatory patterns in ad distribution across multiple countries.
Global Witness conducted an investigation into Facebook's ad delivery system by posting job advertisements in France and the Netherlands. All ads were configured using Facebook's mandatory 'traffic' campaign objective, which shows ads to users Facebook's algorithm determines are most likely to click on the linked website. The advertisers specified only that ads should be shown to adults living in or recently in either country, with no gender targeting specified. However, Facebook's algorithm autonomously decided which users to show the job ads to, resulting in systematic gender-based discrimination in ad delivery. The organization expressed concern that Facebook's ad delivery algorithm was not just replicating but exacerbating societal biases, narrowing opportunities for users and frustrating workplace equity progress. This builds on previous findings by Global Witness in the UK in 2021, and mirrors issues that led to a US Justice Department lawsuit against Meta over housing ad discrimination. The investigation also extended to India, Ireland, South Africa and the United Kingdom. Global Witness has filed complaints with data protection authorities and human rights organizations in France and the Netherlands, arguing that the discriminatory ad delivery violates European Convention on Human Rights, EU law, and national constitutions that prohibit sex-based discrimination.
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.
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
No population impact data reported.