So, how do TikTok’s algorithms actually work?
According to Andrew Kaung, the algorithms’ fuel is engagement, regardless of whether the engagement is positive or negative. That could explain in part why Cai’s efforts to manipulate the algorithms weren’t working.
The first step for users is to specify some likes and interests when they sign up. Andrew says some of the content initially served up by the algorithms to, say, a 16-year-old, is based on the preferences they give and the preferences of other users of a similar age in a similar location.
According to TikTok, the algorithms are not informed by a user’s gender. But Andrew says the interests teenagers express when they sign up often have the effect of dividing them up along gender lines.
The former TikTok employee says some 16-year-old boys could be exposed to violent content “right away”, because other teenage users with similar preferences have expressed an interest in this type of content – even if that just means spending more time on a video that grabs their attention for that little bit longer.
The interests indicated by many teenage girls in profiles he examined – “pop singers, songs, make-up” – meant they were not recommended this violent content, he says.
He says the algorithms use “reinforcement learning” – a method where AI systems learn by trial and error – and train themselves to detect behaviour towards different videos.
Andrew Kaung says they are designed to maximise engagement by showing you videos they expect you to spend longer watching, comment on, or like – all to keep you coming back for more.
The algorithm recommending content to TikTok’s “For You Page”, he says, does not always differentiate between harmful and non-harmful content.
According to Andrew, one of the problems he identified when he worked at TikTok was that the teams involved in training and coding that algorithm did not always know the exact nature of the videos it was recommending.
“They see the number of viewers, the age, the trend, that sort of very abstract data. They wouldn’t necessarily be actually exposed to the content,” the former TikTok analyst tells me.
That was why, in 2022, he and a colleague decided to take a look at what kinds of videos were being recommended to a range of users, including some 16-year-olds.
He says they were concerned about violent and harmful content being served to some teenagers, and proposed to TikTok that it should update its moderation system.
They wanted TikTok to clearly label videos so everyone working there could see why they were harmful – extreme violence, abuse, pornography and so on – and to hire more moderators who specialised in these different areas. Andrew says their suggestions were rejected at that time.
TikTok says it had specialist moderators at the time and, as the platform has grown, it has continued to hire more. It also said it separated out different types of harmful content – into what it calls queues – for moderators.
