Digital People in the Colosseum: Can Technology Master the Violence of Traffic?

Why are digital people being suppressed by domestic new media platforms? Can technology really control traffic? This article will show you how digital people perform on short video platforms. This article is also recommended for companies or individuals to read.

In an announcement released on the 19th, Kuaishou announced that it would not provide additional traffic support for content created with the assistance of AIGC, such as digital human live broadcasts. At the same time, the e-commerce live broadcast room of Kuaishou's official digital human capabilities (Nuwa products) will no longer enjoy additional e-commerce traffic support, and recorded live broadcasts generated by AIGC digital humans outside the site will be controlled according to the "play recorded content" rules.

Within a week, Video Account and Kuaishou successively launched heavy blows against digital humans. At first glance, it may be just an unexpected tacit understanding, but behind Kuaishou's action of "even hitting its own people", the conflict between the platform and digital humans is surfacing.

1. Hunting for digital people: Beijing-Alibaba and Tencent-Kuaisan go left, while Tencent-Kuaisan and Tencent-Kuaisan go right

Although both Video Account and Kuaishou have sent out “yes” signals to digital humans, in the current Chinese Internet, different platforms have completely different attitudes towards digital humans.

To be more precise, video accounts, Kuaishou and other content platforms are restricting and tending to be strict. Traditional e-commerce platforms represented by Alibaba and JD.com have not only vigorously promoted and supported the application of digital human live broadcast in the platform ecology in the past period of time, but also regarded digital human live broadcast as an important focus of platform AI empowerment.

Two months ago, Liu Qiangdong's digital person "Purchasing and Sales Dong Ge" made his "personal" live broadcast debut in JD.com's purchasing and sales live broadcast room. Later, the media reported that during the live broadcast, Liu Qiangdong's digital person was not only able to explain the products fluently, but also to conduct lucky draws and distribute red envelopes in the live broadcast room. In addition, even the habit of rubbing fingers and occasionally nodding when speaking was completely copied by "Purchasing and Sales Dong Ge".

Digital person of "Purchasing and Sales Dongge"

At that time, industry analysts commented that traditional e-commerce was increasingly facing growth bottlenecks, and JD.com had already fallen behind in the current mainstream live streaming e-commerce track. Digital human live streaming, as a new productivity empowered by AI, is a starting point for JD.com to rejoin the live streaming competition.

JD.com has also been trying to prove the value of digital human live streaming.

Just one day before Kuaishou announced the limit on digital human live broadcasts, the head of JD Cloud Yanxi’s innovation business promoted its own products at the "618 Media Open Day", emphasizing that the cost of its digital human live broadcasts is only 1/10 of that of real people, and the effect is "close to real people."

Compared with JD.com, which entered the market late, Taobao's layout in digital human live streaming is faster.

In July last year, Taobao tried AI live streaming for the first time, creating six AI "Lier babies" with different looks to live stream with real anchors. At the Alibaba M Summit held in September, Alibaba launched the "AI full-time live streaming capability", which allows merchants to use hundreds of AI virtual anchors in the AI ​​full-time live streaming library to customize their images according to actual needs, and also enable multi-language live streaming interactions.

According to media reports, at that time, more than 500 merchants had used Alimama's AI capabilities to achieve 24/7 non-stop "AI full-time live broadcast" on the Taobao system, effectively reducing the cost of live broadcast by 90% - this figure is quite consistent with JD.com's statement that "the cost of digital human live broadcast is only 1/10 of that of real people."

The digital people that traditional e-commerce platforms have vigorously promoted are now being "banned" by Video Account and Kuaishou. The reason for such diametrically opposed attitudes is that the two parties are in different positions.

From the perspective of traditional e-commerce platforms, live streaming traffic is a gold mine, and digital humans are tools for merchants to mine gold mines. E-commerce platforms themselves do not participate in mining, and they only need to sell tools to merchants to make profits. Therefore, the core value of the platform is to continuously provide more and more efficient tools. If a tool is not useful, the next new tool can be redeveloped, and the impact on the e-commerce ecosystem is not significant.

However, Video Account and Kuaishou are essentially content platforms, and users are fish in the pond. Digital people, like other content, are daily feeding bait. Content platforms must ensure that they continue to feed enough bait to attract more and more users, and they also need to feed different users with the bait they like, segment user circles, and then open the "pond" to merchants with different products in order to make a profit.

Therefore, content platforms must be responsible for their content. The content produced by digital people with the help of technical means will not only interfere with users' preferences for content, but also pollute the content environment and continue to affect the health of the ecosystem.

2. What’s wrong with digital human live streaming?

Why is digital human live broadcast not recognized by content platforms? The answer to this question can be found in the documents released by the platform.

In the "Implementation Rules for the Release of Low-Quality Content by Video Account Showcase Experts" for which Video Account is publicly soliciting opinions, the "on-hook recorded broadcast" content was changed to "non-real live broadcast" content, which means that the scope of application of the rules has become wider.

In the past, some live broadcast rooms would use the method of recording and broadcasting on hold to maintain their accounts, increase their broadcast time, or take advantage of the rewards for new live broadcasters on the platform. However, this type of content itself does not meet the interactive characteristics that live broadcasts should have, so "recording and broadcasting on hold" is a form of live broadcast fraud.

In the solicitation of opinions, "non-real live broadcast" is defined as the use of non-real live broadcast content or props instead of real live broadcast during the live broadcast. For example, using plug-ins, AI and other tools to generate virtual images for live broadcast. The specific objects of this definition also include "live broadcast fraud" using AI technology.

Since fraud is involved, it is natural for the platform to control related behaviors. However, the platform is not only protecting its own interests, but also helping users avoid potential risks.

In May last year, Deepfake's AI face-changing and voice-simulation sparked heated discussions across the Internet. Some criminals used AI face-changing to defraud 4.3 million yuan in 10 minutes through a video call. In the music section of Bilibili, UP used AI to simulate Stefanie Sun's voice and released many cover songs. In the end, it even attracted the singer Stefanie Sun herself, who responded to her views on AI technology in an article titled "My AI".

*“AI Stefanie Sun” on Bilibili

The situation of AI being rampant has also emerged in the field of live e-commerce: live broadcast hosts who sell goods have collectively used AI to change their faces into female celebrities. Every three steps you take, you see a Dilraba Dilmurat, and every five steps, you see a Tong Liya. If you don’t look carefully, you can’t tell them apart. These celebrity live broadcast rooms produced by AI have also brought a lot of traffic to businesses.

If the stars had not actively defended their rights and the platform had not intervened in time, the 99-yuan AI face-changing software would have become a standard feature of live broadcasts, and digital people would not have become a new trend. Another situation that the platform does not want to see is that if all live broadcasts are controlled by "AI stars", it means that the content ecosystem composed of live broadcasts will lose its diversified characteristics.

Imagine this: when a user returns home from get off work and opens the live streaming app, on the previous screen, "Dilraba Dilmurat" is still showing off lipstick shades in front of the camera, and on the next screen, there is another "Dilraba Dilmurat" promoting sun-protective clothing. Swipe down two more screens and you'll see Gulinaza and Tong Liya introducing beauty devices and hair dryers, respectively.

Not to mention that users will also get aesthetic fatigue after watching too many live broadcasts of celebrities. The trust that live broadcasts of goods originally built on the perception of reality is easily broken by the "celebrity anchors" filling the screen, and it may even make users question the content of the live broadcast and even the live broadcast products. It seems to earn traffic, but in fact it loses jobs.

The same principle applies to digital humans.

In the "Purchasing Dong Ge" launched by JD.com, we can see that the AI ​​digital human has been shaped to be highly similar to a real person, and has even learned Liu Qiangdong's personal habits of rubbing his fingers and nodding. But imitation is imitation after all. Liu Qiangdong rubbed his fingers and nodded when he spoke, which was a subconscious behavior. The generation of these actions and the timing of their execution are completely illogical, and are also beyond the control of digital humans who rely on algorithms and logical judgment to complete cosplay.

In the eyes of AI, if the digital person rubs his fingers and nods when speaking, it means he "looks like" Liu Qiangdong. So the "Purchasing and Sales Brother Dong" in front of the camera will frequently rub his fingers and nod during the live broadcast, and turning a small action that should not be noticed into a part of the character's characteristics is a failure of AI anthropomorphism.

The clues produced by deliberate imitation are not the fault of the digital people, but the practice of creating a platform for digital people and magnifying these "defects" into selling points to promote tools to businesses is itself questionable.

3. Where are digital humans needed?

In the eyes of e-commerce platforms, digital people are a new tool for merchants to participate in the live streaming e-commerce competition, and they also lower the threshold for ordinary merchants to enter the live streaming e-commerce field.

For example, using digital human live broadcasts eliminates the need to build a real live broadcast room, which can save on set and equipment costs. Modeled digital human live broadcasts do not require a complete live broadcast team including anchors, planners, directors, and photographers, which greatly reduces labor costs. In addition, digital human anchors are stable and do not need to rest. They can live broadcast 24 hours a day, and there is no need to worry about anchors changing jobs.

According to a report by the Liberation Daily, based on the annual cost of 20,000 yuan for mainstream digital human anchor products in the market, the average monthly live broadcast cost is less than 2,000 yuan, and it can also be broadcast 24 hours a day. In Hangzhou, the monthly salary of a real anchor starts at 10,000 yuan. A live broadcast room requires two anchors to take turns, and the labor cost alone is 20,000 yuan per month, and the live broadcast time is only 16 hours a day.

The cost-effectiveness of digital human live broadcast and real-person live broadcast is very different, but is this really the case?

In fact, after the topic of digital human live broadcasting began to heat up in the first half of last year, quite a number of businesses tried to use digital humans for live broadcasting. Although there were no platform rules restricting digital human live broadcasting at the time, many businesses were detected by the platform due to duplicate digital human images, resulting in account bans.

The merchant thought that he could broadcast for a year with 20,000 yuan, but in the end, he was blocked for 2 hours after broadcasting for 5 minutes, and could not broadcast normally at all. Some accounts were permanently banned, and all the money spent on buying digital people was wasted.

In terms of efficiency, although the advantage of digital people being able to live broadcast 24 hours a day is very attractive, the traffic of the live broadcast platform is not evenly distributed over 24 hours. The prime time from 8 pm to 1 am belongs to the big anchors on the platform, and digital people can't grab much traffic when they start broadcasting. In other time periods, digital people also have to "PK" with other real anchors, just like accumulating "Cullinan fragments", picking up the traffic forgotten in the corner.

In terms of the categories of goods that digital humans can bring, there are also obvious limitations. At present, the live broadcast rooms of digital humans on e-commerce platforms are better at bringing low-priced, standard products. The reason is that when consumers are faced with low-priced standard products, they spend less time making consumption decisions. Therefore, through the direct guidance of digital humans, there is a greater probability that users will place orders.

However, when faced with high-priced, non-standard products, consumers take longer to make decisions and need timely interactive feedback to facilitate ordering. However, digital humans are not as natural as real people in terms of interactivity and cannot provide timely interactive feedback. In addition, for categories involving product experience such as furniture and home furnishings, digital humans cannot intuitively demonstrate product features through action demonstrations like real people.

Therefore, it is not difficult to find that the core value of live e-commerce lies in the high efficiency of online communication, the interactivity of real anchors, and the emotional connection formed through subjective communication. In contrast, digital humans only win in efficiency, and there is nothing special in other aspects. And the practice of using digital humans to bring goods is itself a waste of technology.

If we look beyond the live streaming and selling scenes, the advantages of digital humans will become more obvious: they are online 24 hours a day and available at any time; they can express themselves like humans and can be standardized. The most suitable application scenarios for these advantages are obviously not "123, go to the link", but private scenarios that can provide personalized services and emotional value.

For example, the "Resurrection of Relatives with AI" that sparked heated discussions during the Qingming Festival this year can synthesize a video using only the photos and audio of the deceased before his death, providing users with a special comfort. The upgraded digital human created with chatbots is more likely to achieve real-life conversations, bringing the digital life in "Wandering Planet 2" into reality.

On this basis, if we let our imagination run wild and create "expert digital humans" in all walks of life, we can not only preserve the professional capabilities of experts in the digital world and help humans continue to break through various scientific research problems, but also use AI's intelligent learning and interactive capabilities to invest in the field of education and promote the popularization and inclusiveness of education.

To give a specific example: Douyin's AI tool "Doubao" has already implemented AI functions such as chatbots, writing assistants, and English learning assistants. The English learning assistant can not only conduct real-time oral conversations with users, but also promptly point out grammatical problems that users encounter when expressing themselves. In the past, similar oral practice usually required offline or online scenarios, and appointments with real teachers were paid.

Therefore, if "Doubao" is combined with digital humans, everyone can have a 24-hour online oral teacher who can learn at any time. And this kind of new breakthrough in the original application scenario is the best destination for digital humans.

Frankly speaking, the core purpose of the e-commerce platforms’ vigorous promotion of digital live broadcasts is to attract more merchants to the live broadcast e-commerce scene and provide new bargaining chips for platform competition. However, traditional e-commerce companies have overlooked a reality: the foundation of live broadcast sales is built on healthy content expression, not extreme conversion efficiency.

For example, the top anchors can create higher GMV than other anchors, not because of more live broadcasts, but because of the long-term accumulation of live broadcasts and the trust relationship established with fans. It’s not that technology can’t do this, but under the product thinking that efficiency is the priority, these initial investments will be regarded as redundant costs.

In the current era of continuous low-price competition, merchants have an urgent need to reduce content costs, but the premise of using scaled logic to create traffic and complete conversions is to first have a foundation of high-quality content. In the final analysis, digital people are just a new tool, and for merchants, it is always more important to figure out the goal than how to reach it.

Author: Bai Lu

WeChat public account: Jingzhe Research Institute (ID: 1033550)