Facebook Admits AI Failure During New Zealand Terror Livestream
Facebook acknowledges that its artificial intelligence failed to detect and block the horrific livestream of the New Zealand mosque attack. The company also notes that users shared over 800 distinct variants of the video across multiple platforms before they could stop the spread.
Facebook admits that its artificial intelligence completely fails to detect and block the livestreamed video of the New Zealand mosque terror attack. The shooter broadcasts the horrific incident live on the platform, and the automated systems designed to catch violent content do not flag the footage as it happens. Facebook's vice president of product management, Guy Rosen, shares these details in a blog post as the company faces intense global criticism for its role in the tragedy.
The social media giant also blames users for not reporting the graphic video quickly enough to human moderators. After the initial broadcast, individuals around the world rapidly re-share the footage across a porous digital ecosystem, including YouTube, Twitter, Reddit, and 8chan. People capture the video in various ways, such as filming television screens and recording computer monitors, which makes it incredibly difficult for any single company to track and remove the content.
Despite these challenges, Facebook manages to find and block over 800 visually distinct variants of the violent clip. The company explicitly states that implementing a simple time delay on livestreams would not effectively prevent similar broadcasts in the future. Facebook promises to review its products and look for ways to improve its detection systems going forward, though this incident highlights the severe limitations of current AI moderation technology.