TikTok is planning on building out its artificial intelligence (AI) tools that can moderate content, CEO Shou Zi Chew said at TED2023 yesterday (April 20). The company currently employs “tens of thousands” of workers to moderate user content, in addition to using AI, he said.
“The tech will become more precise, more specific and handle a larger scale with content moderation,” he said.
Chew took the stage in Vancouver, British Columbia in Canada for TED’s annual event. He spoke for 20 minutes with Chris Anderson, head of the conference organization.
Content moderation was top of mind for U.S. legislators when Chew testified in front of Congress last month. Rep. Cathy McMorris Rodgers, a Washington Republican, accused the company of presenting harmful content to users, including content that promotes self-harm, eating disorders and dangerous challenges. At the hearing, Chew repeated how that content violates TikTok’s guidelines, which didn’t seem to appease legislators. But at the TED conference, he shed light on how TikTok’s content moderation actually works and what the company is doing to improve it.
TikTok’s Irish content moderation team is one of the company’s “most important cost items,” Chew said in the interview. He didn’t elaborate on how much the company spends on content moderation, which as a private company, it isn’t required to disclose.
Chew’s five-year vision for TikTok includes increasing the number of small businesses that profit from the app and promoting the discovery of new ideas, he said.
How TikTok’s algorithm works
At the conference, Shew also explained the basics of TikTok’s algorithm, which Anderson pointed out is unlike any other social media recommendation engine. The technology learns what users are interested in based on what they’ve liked in the past, and it finds others who have liked the same videos, or who have the same “interest signals,” Chew said. The algorithm then recommends what those users individually like to each other, which is how the app can predict what users are interested in before they might know it themselves.
The algorithm isn’t unlike Meta’s, which owns Facebook and Instagram. But TikTok’s quick growth and high viewership could be attributed to the vision it was founded with, Chew said. It always intended to recommend posts based on what users like rather than who they know, while other apps were built on something different, he said. Meta (META) began as a company to connect friends online before pivoting to recommend content based on interests.
Some critics suggest Instagram has lost touch with its purpose. Users want to see posts from their friends and an algorithm that favors photos over videos, but their feeds resemble the interest-based videos they see on TikTok, according to a petition entitled “Make Instagram Instagram Again” that has 2.3 million likes on the app.
“You can’t just shift it halfway,” Chew said. “The community comes in and expects something different.”
But that’s what Instagram seems to be doing. “We’re also going to need to evolve, because the world is changing quickly, and we’re going to have to change along with it,” said Adam Mosseri, head of Instagram, in a video posted on Twitter last year.