How Chinese influencers use AI digital clones of themselves to pump out content – The Guardian Feedzy

 

In September, Chen Yiru, a Taiwanese influencer with nearly nine million fans on Weibo, livestreamed footage of himself eating chicken feet for a jaw-grinding 15 hours.

His followers were suitably wowed – until some started to question if such a feat was humanly possible. The small print on the video stream confirmed their suspicions: “For display purposes only, not a real person.”

Many of Chen’s fans were outraged, and he reportedly lost more than 7,000 followers between 24 and 26 September. Even the legal community weighed in. Quoted in Chinese media reports, Dong Yuanyuan, a senior partner at Tiantai, a Beijing law firm, said that AI avatars could not be “completely untied from the celebrity himself” and that “virtual live broadcasts … do not exempt celebrities from legal liability”.

But Chen is hardly alone in outsourcing his duties to an AI avatar. Chinese influencers, or key opinion leaders (KOLs), particularly in the e-commerce industry, are increasingly turning to digital clones to pump out content 24/7. For some stars, like Chen, this enables them to take their content and earnings to even greater heights. But for lesser-known livestreamers, AI may put their jobs at risk, as media companies pivot towards cheaper digital stars.

Livestreaming is big business in China. The industry employed more than 1.23 million people in 2020, according to iResearch, and there are more than 700 million internet users who follow their channels, according to Daxue Consulting. While the phenomenon started out with livestreams of people talking, singing or going about their day, the industry has become closely intertwined with the world of e-commerce. Livestreamers are expected to rake in 4.9tn yuan (£0.5tn) in sales in 2023, more than 11% of the total e-commerce sector.

Livestream shopping channels show influencers talking about, or trying out, products for hours on end. They can respond to viewer questions about the products and push discounts and sales for brands.

Now AI startups are getting in on the trend by selling digital avatars to influencers and media companies. Silicon Intelligence, based in Nanjing, can generate a basic AI clone for as little as 8,000 yuan, although the price can increase for more complicated programming, according to MIT Technology Review. The company only needs one minute of footage of a human being to train a virtual livestreamer.

A recent survey of 10,000 young people on Weibo found that more than 60% would be interested in working as influencers or livestreamers. But it is these up-and-coming influencers whom the AI bots are most likely to displace.

“This trend may place more pressure on lower-tier livestreamers as they are more dispensable to brands,” says Yaling Jiang, an independent analyst and founder of Following the Yuan, a newsletter about Chinese consumers.

Bigger fans like Chen rely on their off-camera profiles to boost their status and bankability. The most difficult part of becoming a successful KOL “is to be part of the hype and media cycle”, says Jiang. “The AI influencers do not have gossip, aren’t seen in reality shows, on the streets or in the stadium like Taylor Swift is. If they aren’t in the public eye, what media value do they have?”

Then there is the issue of authenticity. On 11 October the Chinese government published draft guidelines for firms using generative AI technology. The proposed regulations said that individuals to be cloned using AI should provide written consent for their biometric data to be used in that way, but they did not elaborate on how such content should be labelled to the public. Some platforms, like Douyin, have their own requirements, but they are not widely applied and, according to Jiang, “there are still a lot of grey areas”.

The world of deepfake livestreamers may soon catch the attention of Chinese regulators. But until then, video platforms are featuring an increasing number of clones – and videos of clones advertising clone-making services.

Additional research by Chi Hui Lin