TikTok’s AI-Run Video Removal System Boasts a 5% False Positive Rate
Updated: Aug 21, 2021
On 9 July 2021, TikTok announced that it will use an artificial intelligence-based system to automatically detect and delete videos that go against its policies and safety strategies. In an official statement, Tiktok’s Head of US Safety Eric Han clarified that the change was made with moderators’ well-being in mind.
Credit: Solen Feyissa
“In addition to improving the overall experience on TikTok, we hope this update also supports resiliency within our Safety team by reducing the volume of distressing videos moderators view and enabling them to spend more time in highly contextual and nuanced areas, such as bullying and harassment, misinformation, and hateful behavior,” the post read.
But the company will only apply such technology to content categories where it can remove malicious videos at high accuracy levels. Currently, automated processes are helping take down clips disobeying Tiktok’s rules on minor safety, adult nudity and sexual activities, violent and graphic content, illegal activities and regulated goods.
Digital Information World reported that TikTok’s artificial intelligence (AI) wrongfully flags videos for a breach in guidelines five per cent of the time. When a user has supposedly committed a Community Guidelines violation, TikTok will now send that person a warning. In times when a zero-tolerance policy has been overstepped, an automatic account ban will follow.
Credit: Digital Information World
Should someone be charged with more than one violation, they will experience at least one of the four consequences, depending on the seriousness and frequency of their actions: a 24 or 48-hour ban preventing them from uploading content, commenting or editing their profile; a 72-hour view-only experience making them unable to post videos or engage with others; a message cautioning them that their account is close to being banned; a permanent ban.
Content Creators can appeal undue video removals and get their posts back on the platform if there was a misreport. Users can also inform TikTok of any violations on the app.
Automating content moderation can present an advantage to both moderators and users as it will let the former focus their efforts on eliminating bullying, harassment, fake news and other distressing user activity. In return, this will make TikTok a safer space.
Tiktok Pte. Ltd. is one of the most dominant app publishers in Singapore, with a global total of over 2.6 billion downloads across its five apps in December 2020. According to market research firm 42matters, TikTok has amassed more than 100 million downloads locally, ranking fifth in the “highest ranked apps in Singapore” list.
Written by Sophia Lopez