Facebook AI Misclassifies Video of Black Men as 'Primates': An Analysis of Racial Bias in Algorithmic Behavior

Facebook has issued an apology after its AI mistakenly labeled a video featuring Black men with an inappropriate prompt. Users who viewed the Daily Mail video were prompted to "[k]eep seeing videos about Primates," prompting outrage over the "unacceptable error." In response, Facebook has disabled the recommendation feature responsible for this incident and is investigating the underlying cause to prevent future occurrences.

Company spokeswoman Dani Lever acknowledged the issue, stating, "While we have made improvements to our AI, we know it's not perfect, and we have more progress to make. We apologize to anyone who may have seen these offensive recommendations."

This incident highlights a broader issue of gender and racial bias in artificial intelligence. Facial recognition technology continues to struggle with accuracy, often misidentifying people of color and women. Last year, false matches in facial recognition led to the wrongful arrests of two Black men in Detroit. Additionally, in 2015, Google Photos infamously misidentified images of Black individuals as "gorillas," subsequently opting to censor the term from its searches and tags.

In a bid to address these challenges, Facebook created a dataset aimed at improving AI understanding of diverse demographics. This dataset features over 40,000 videos with 3,000 paid actors, who provided information on their age, gender, and skin tones. Despite these efforts, the recent incident demonstrates that much work remains for the AI community to eliminate bias and improve the accuracy of these technologies.