Content moderators report that the introduction of artificial intelligence, rather than simplifying their work, has made it more challenging. Algorithms frequently fail to accurately identify violent and harmful content, increasing the workload for specialists and raising the risk of dangerous material spreading online.
According to Bloomberg, which interviewed 13 industry professionals, AI systems are still unable to reliably detect violent or unsafe content. A TikTok employee, speaking under the pseudonym Kevin, revealed that since he began using AI in his daily work over a year ago, the system has been wrong roughly 70% of the time. For example, the algorithm may focus on a fuel gauge in a video while ignoring a speedometer reading 200 km/h, which is critical for assessing risk.
Moderators note that AI is being deployed faster than it can learn to effectively filter dangerous content. Experts warn of a future internet increasingly controlled by algorithms, where hate speech, propaganda, threats to children, and other harmful material could spread freely.
“If AI is implemented simply to reduce the number of safety specialists, it will only lower protection levels. Real humans are necessary,” said Lloyd Richardson, CTO of the Canadian Centre for Child Protection.
Content moderation is already emotionally demanding, and hiring sufficient staff is extremely difficult: YouTube alone sees over 20 million videos uploaded daily. Companies are increasingly relying on AI, but it often complicates the work instead of helping.
Of the 13 moderators interviewed by Bloomberg, 12 said their workflow has become more difficult since neural networks were introduced. Kevin explained that much of his time is spent correcting AI errors manually, labeling content accurately. He believes that these corrections are essentially training the system that may eventually replace human moderators, even if management does not explicitly acknowledge this
