Rise of the Machines: How AI-Generated Streamers Are Revolutionizing Chinese E-Commerce | by Kieran Bush | Oct, 2023

The Midnight Enigma: Unmasking China’s Mysterious AI Streamers

Imagine scrolling through live streaming videos on Taobao, China’s premier ecommerce platform, at 4 a.m. The most curious thing you’ll notice is the bustling activity. While most people are sound asleep, diligent streamers are hard at work, showcasing products and offering discounts to their audiences. But look closer, and you’ll see something peculiar — many of these livestream influencers appear somewhat robotic. Their lip movements almost match their words, but there’s a subtle unnaturalness that gives them away.

Here’s the twist — these streamers aren’t real people. They’re AI-generated clones of actual streamers. As technologies for creating lifelike avatars, voices, and movements become more advanced and accessible, the popularity of these deepfakes has surged across China’s e-commerce platforms.

The Live Streaming Phenomenon in China

In the realm of marketing in China, live streaming reigns supreme. Whether it’s on Taobao, Douyin, Kuaishou, or other platforms, influencers can strike massive deals within hours. Top names can rake in over a billion dollars’ worth of sales in a single night and enjoy a status akin to that of big movie stars.

However, for smaller brands, the process of training and retaining live stream hosts, along with managing the technical aspects of broadcasting, can be prohibitively expensive. The solution? Automation.

The Economic Impact of AI-Generated Streamers

Since 2022, a wave of Chinese startups and tech giants has been offering services to create deep-fake avatars for e-commerce live streaming. With just a few minutes of sample video and around $1,000 in costs, brands can clone a human streamer who can work around the clock.

One Xiaoice streamer even managed to bring in over 10,000 RMB ($1,370) in revenue in just one hour, demonstrating the economic potential of these AI-generated streamers. When the livestream focuses on a single product, Xiaoice’s AI streamer can interact with it in front of the camera, engaging the audience and enhancing the shopping experience.

From Deep Fake to E-Commerce: The Evolution of Synthetic Media

The concept of synthetic media has been making headlines since the late 2010s. It gained notoriety when a Reddit user named “deepfake” swapped faces in adult content. While the technology has evolved, the core idea remains the same — it enables faces to be generated or manipulated to resemble specific real individuals and perform actions the real person has never done.

Initially, deepfake technology was notorious for its misuse in revenge porn, identity scams, and political misinformation. Attempts to commercialize it for more benign purposes remained somewhat of a novelty. However, Chinese AI companies have discovered a new use case that’s gaining traction.

The Birth of AI-Generated Streamers

Founded in 2017, the Nanjing-based startup Silicon Intelligence initially focused on natural-language processing, particularly text-to-speech technologies. However, in 2020, the company started to explore AI’s potential in the realm of live streaming. At first, it required 30 minutes of training videos to create a digital clone capable of human-like speech and actions. Over the years, this time dramatically reduced to one minute of video.

As the technology advanced, the costs decreased. Generating a basic AI clone now costs around $1,100, covering a year of maintenance. The generated avatar synchronizes its mouth and body movements with scripted audio, and the scripts themselves are often generated by large language models.

AI Streamers in Action

Once created, these AI streamers perform seamlessly. All human operators need to do is provide basic product information, proofread the generated script, and watch the digital influencer go live. In advanced versions, the technology can interact with live comments, providing matching responses from its database. It can also adjust its marketing strategy based on viewers’ numbers.

These AI clones are trained using common scripts and gestures typically seen in e-commerce videos, ensuring their language and body movements match. For example, if a human streamer says, “Welcome to my livestream channel. Move your fingers and hit the follow button,” their avatar points upwards, mirroring where the “Follow” button is located in most mobile livestream apps.

Embracing Emotion and Expanding the Workforce

Several Chinese tech giants, including Alibaba, Tencent, Baidu, and JD, have launched their own AI-generated live streaming services. These generated streamers, though they might not surpass the star influencers, are well-suited to replace mid-tier ones, leading to reduced job opportunities and lower average salaries for human streamers.

The cost-effectiveness of AI streamers has made them increasingly attractive to brands, especially as they strive to lower costs and improve efficiency. AI-generated streamers may not fully replace humans, but they are well on their way to becoming an essential part of the livestreaming ecosystem. As the technology continues to evolve, it remains to be seen how regulators will address its implications for synthetic media and generative AI.