TikTok's "Suggested Accounts" Algorithm Allegedly Reinforced Racial Bias through Feedback Loops

Marc Faddoul, an AI researcher from UC Berkeley specializing in algorithmic fairness, conducted an experiment on TikTok's user recommendation system. When following a new account, TikTok suggests other accounts to follow, and Faddoul observed that these recommendations appeared to be based on physical appearance similarities including race, gender, hair color, age, body profile, and even visible disabilities. He repeated the experiment with a fresh account and documented similar results, which he shared on Twitter calling them 'face-based filter bubbles.' BuzzFeed News replicated similar experiments with comparable outcomes, such as following a hijabi creator leading to recommendations of other women wearing hijabs. TikTok denied that their algorithm uses race or profile pictures for recommendations, stating they use collaborative filtering based on user behavior patterns where users who follow account A also follow account B. However, Faddoul noted that collaborative filtering could still reproduce existing biases in user behavior and create feedback loops that make it harder for underrepresented creators to gain followers. This was not TikTok's first controversy, as they had previously admitted to suppressing content from queer, fat, and disabled users in December.