TikTok’s AI Bots Will Destroy the App

TikTok unveiled new artificial intelligence tools last week that will allow brands and creators to make digital avatars to help sell things online and reach a global market.

To highlight the new offerings, the company released a video featuring a rather uncanny AI clone of its Head of Global Operations, Adrienne Lahens, touting how the new features enable her digital duplicate to speak Spanish, French, German and Japanese. The only problem: Her line of AI-generated Japanese was a load of gibberish.

This push to flood TikTok with what are essentially deep-fake clips of actors and creators to promote online shopping threatens the integrity of the app. It also comes at a time when TikTok is fighting for its future in the US due to perceived national security concerns over its Beijing-based parent, Bytedance Ltd. Rushing to bombard “For You” pages with AI bots at a time like this is unwise.

The company has since quietly deleted the demo video, replacing it with a version that at least uses “nihongo” as the accurate Japanese word for the language. But as we’ve all learned, it’s hard to completely remove something from the internet, and the original video remains embedded in tech news outlet Endgadget’s write-up of the announcement. A TikTok representative in Tokyo blamed a “technical fault” for the blip in the original clip.

It’s hardly the first time a tech firm has had an eyebrow-raising rollout of a new generative AI tool. Google’s parent Alphabet Inc.’s stock briefly plunged last year after it showed the public a first look at its answer to OpenAI’s ChatGPT and the bot spit out a factually inaccurate response to a question about NASA’s James Webb Space Telescope. TikTok’s fumble is just the latest, and shows that despite the massive global hype, generative AI technology is still loaded with glitches and hardly ready for the myriad magical applications industry proponents like to claim.

The internet is already flooded with AI-generated videos that make it harder than ever to tell who is real and who is fake. The threat of these becoming more widespread points to a dystopian and annoying future online experience. And their proliferation can be used for much more nefarious purposes, like creating nonconsensual deep-fake porn.

TikTok, for its part, hasn’t publicly released the suite of tools it teased last week that will let creators and advertisers build custom AI avatars or use its offering of pre-built avatars based off real people, and has invited community feedback. It also promises to require brands that do eventually employ the new features to label the content as AI-generated, offering a morsel of transparency to users.

This foray into livestreamed shopping experiences that use AI avatars may be new to the US market, but the concept has been widely embraced by tech companies in Asia, and especially China, for years now — to the tune of billions of dollars.

More than 10% of China’s retail purchases last year came via livestreaming, Bloomberg News reported this month, skyrocketing from just 1% in 2019. Douyin, the Chinese version of TikTok with the same parent company, was used by 88% of frequent live-commerce consumers, a McKinsey analysis last year found.

The AI element here isn’t new, either. E-commerce giant JD.com even created an AI avatar of its founder and Chairman Richard Liu to hawk electronic devices, as well as steak and blueberries, on a livestream earlier this year.

There are fresh signs, however, that even Chinese users are tired of AI clones trying to sell them stuff. Tencent Holdings Ltd.’s WeChat signaled earlier this month that it plans to ban the use of AI on its livestreaming commerce platforms, as Chinese internet regulators have asked local providers to take measures to prevent the spread of false information and discriminatory content. Citing people familiar with the matter, the South China Morning Repost reported that WeChat’s motive for the crackdown was to encourage more live anchors to have real-time interactions with viewers as AI versions of livestreamers proliferate.

Ultimately, TikTok’s powerful algorithm drives engagement and has helped make it one of the fastest-growing apps in the US, with more than 150 million users. But what really makes the app special is the mega-viral clips of human experiences that cannot be created by AI. My current favorite TikToker is an American trucker who found a stray kitten under his trailer and now posts daily updates about his new pet’s adventures, while still pretending not to like cats. My least favorite? The ones I can tell are clearly AI-generated and those trying to sell me things.

TikTok’s stated mission is to “inspire creativity and bring joy.” Flooding the app with creepy AI clones of people for e-commerce is the antithesis of this. At a time when it can’t afford any missteps in the US, the company should think again before rushing to do this.

More from Bloomberg Opinion:

• China’s AI Strength Shows Limit of US Curbs: Catherine Thorbecke
• AI Doesn't Have to Drain So Much of Our Energy: Parmy Olson
• Tim Cook Controls iPhone, So He’s the New AI Kingmaker: Dave Lee

Want more from Bloomberg Opinion? OPIN <GO>. Or subscribe to our daily newsletter.