The future of AI digital humans

  2021 is the first year of the metaverse, and the popularity of the metaverse concept has led to the rapid heating of the digital human market.
  Digital people are becoming a trend, pouring into people’s daily life-virtual beauty expert Liu Yexi, who has over a million likes in three days after his debut on Douyin, has become a “top stream” in the domestic virtual idol world overnight. ;In Jiangsu Satellite TV’s New Year’s Eve concert, former singer Teresa Teng “returned” to the stage, singing duet with singer Zhou Shen, interweaving the youthful memories of generations; more than 20 digital people appeared on the same stage at the Winter Olympics, serving as sign language Anchors, weather anchors, Olympic public welfare ambassadors and other roles…
  The popularity of digital people has attracted many participants to enter the game. According to the data of the company, there are more than 280,000 digital people-related enterprises in China, and newly registered enterprises in the past five years. The compound growth rate has reached nearly 60%.
  At the same time, capital is constantly pouring into the digital human race. According to Tianyancha data, in 2021, there will be a total of 27 digital human-related investments, with financing amounts ranging from millions of RMB to tens of millions of dollars. In less than a month from the beginning of 2022, nearly 100 financings have been completed in the digital human field, with a cumulative amount of more than 400 million yuan.
Digital people L1-L5 level

Source: SenseTime Intelligent Industry Research Institute

  The popularity of the digital human market continues, and technology-driven and demand traction are also the keys to help. The core of digital human is “human”, which is essentially to improve the comprehensive experience of digital human through digital technology, so that it can bring real-life feeling and interaction.
  On the one hand, with the development and integration of artificial intelligence, virtual reality, high-precision rendering and other technologies, the degree of anthropomorphism of digital humans is getting higher and higher, from image, expression, posture, action, to voice, semantics, voice, etc. In all aspects, it is gradually approaching the level of real people.
  On the other hand, the in-depth application of artificial intelligence technology in digital human image generation, action driving and language interaction will further improve the automation level of digital human production and promote the digital human market from niche to mass.
  On the demand side, both the flow economy and the demographic dividend are facing growth bottlenecks, placing more demands on productivity efficiency and cost. The highly anthropomorphic digital human will replace the real person and enter all fields of production and life, which will create a huge imagination space and application prospect for the new consumer market in the Z era and the digital transformation of the industry. According to the calculation of the Toubao Research Institute, the overall market size of my country’s digital human will reach 270 billion yuan in 2030.
Three basic characteristics

  The anthropomorphism of the digital human and the degree of automation in the production reflect the overall evolution and development level of the digital human system, and represent the comprehensive application ability and maturity of digital technology. According to the two dimensions of anthropomorphism and automation, we can divide digital humans into five levels, L1-L5.
  Among them, we collectively refer to L4 and L5 digital people as “AI digital people”. “They” not only have a high degree of anthropomorphic presentation, but are closer to the level of real people in terms of image, movement and intelligence. They can understand, understand, have memory, self-learning, and can interact with people naturally. At the same time, a large number of artificial intelligence algorithm technologies are also integrated in the production process to improve the production efficiency of the digital human and reduce the production cost of the digital human. Only digital people who reach the L4 level and above can really shine in the consumer and industrial fields.
  First of all, at the application level, multi-modal interaction is the core strength of AI digital human.
  Having sufficient natural and realistic multimodal interaction capabilities is the key for digital humans to gradually replace human characters in a wider range of application scenarios. The so-called “multi-modal interaction” refers to the combination of deep learning neural network and computer graphics, which fully simulates the natural and real interaction between people and realizes “understand, see, and speak” human-machine interactive effects.
Three basic characteristics of AI digital human

Source: SenseTime Intelligent Industry Research Institute

  The AI ​​digital human with multi-modal interaction ability can not only present multimedia information that cannot be displayed by traditional voice dialogue, but also complete multiple interactive tasks such as identity recognition, gesture recognition, and emotion recognition by combining visual AI technology, making the interaction process more convenient. Rich and efficient. At the same time, the visual realistic image also endows the AI ​​digital human with a unique emotional temperature, which helps to establish a humanized emotional bond.
  Secondly, at the value level, autonomous learning is the creativity of AI digital people.
  Behind every AI digital human is a “strongest brain”, which can be based on natural language processing, knowledge graph and other technologies, combined with knowledge bases and massive data training in different fields, to carry out in-depth learning and self-iteration to make oneself more and more The more “smart” you are, the more professional you are, so that you can quickly adapt to the ever-changing market changes and segmented scene demands, constantly break the existing application boundaries, and continue to create new value and new experiences.
  Third, at the production level, AIGC is the productivity of AI digital humans.
  High production costs and long production cycles hinder the large-scale development of the digital human industry. In the traditional process, each digital person relies on manual “carving”. Of these, 3D modeling alone takes months. Creating a high-precision, high-fidelity 3D digital human image often requires millions of capital investment.
  AI reshapes the production process and assists the automated generation of digital humans. It is the basis of AI digital human productivity. It can accelerate the production of digital humans and reduce the production threshold and cost input. For example, the AI ​​digital human image of SoftBank COO Yasuyuki Imai created by SenseTime for the SoftBank Conference is based on facial scans of a small number of photos, combined with AI algorithms to quickly generate high-precision 3D models of digital human beings, and the traditional 3-6 month production cycle shortened to just 15 days.
Three application directions

  According to the purpose of use and the underlying logic, the development of AI digital human can be roughly divided into three application directions.
  Direction 1: Mainly for the purpose of creating IP influence or building a fan economy, including virtual idols, virtual KOLs, virtual actors, virtual anchors, etc.
  Based on “IP incubation + content operation”, “them” is given unique personality and personality traits, in order to attract the attention of different audience groups, thus forming a certain scale of traffic base and emotional links. Then through various means such as live broadcast, cross-border brand endorsement, IP authorization of peripheral derivatives, entertainment and performance, etc., to achieve closed-loop value or commercialization.
  Compared with real IPs, digital human IPs are more malleable. The creative freedom including image, character design and background story brings more imagination space for the business innovation of digital human IP and reshapes the fan economy.

  For example, users or fans can be invited to participate in the creation and incubation process of digital human IP, and a strong emotional connection between IP and users can be established through “co-creation”, making IP more realistic and vital. Especially in the path of brand self-built digital human IP, the digital human IP that fits the brand tonality and consumers’ psychological expectations is more conducive to the effective transmission of brand concepts and rapid breakthrough of the circle, thereby obtaining more benefits. At the same time, the digital human IP is also more controllable, and will not be affected by uncertain factors such as the collapse of human design, negative news, schedule or contract issues, and high commercial security and stability.
  Direction 2: Mainly for the purpose of replacing real-person services and achieving cost reduction and efficiency improvement, including virtual customer service, virtual front desk, virtual tour guide, virtual host, etc.
  ”They” can provide uninterrupted service support “7 x 24 hours”, especially for standardized and highly repetitive human services, which can realize digital replacement, and combine business process automation to help enterprises further improve production efficiency and reduce labor service costs , providing a new path for the digital transformation of enterprises.
  Compared with human service, AI digital human has higher flexibility and is not affected by subjective, time, environment or external uncertain factors. The hidden cost of certainty.
AI digital human application direction

Source: SenseTime Intelligent Industry Research Institute

  At the same time, the phenomenon of increasing marginal benefits of AI digital humans is remarkable. On the one hand, although a certain amount of investment is required for the production of digital human beings in the early stage, the marginal cost of copying and using digital assets is very low, and the variable cost of a single digital human being is also lower than that of a real human being; on the other hand, AI digital human has strong self-learning ability , combined with knowledge graph technology and data training, can continuously optimize service accuracy and expand business breadth, thereby improving the input-output efficiency of enterprise digital human assets.
  Direction 3: With the gradual maturity of artificial intelligence and related technologies, through self-learning and cognitive generalization, AI digital people will comprehensively break through the application boundaries and upgrade to become super assistants in the digital world.
  Different from the first two application directions, the third major application direction of AI digital human not only follows the “replacement” logic of the real world, but also aims to meet the user’s connection and interaction needs with the digital world, and realize direct operation of the digital world. These digital humans will become our AI agents in the digital world. Through direct interaction with “them”, it can “adaptively adapt” to the personalized and diverse needs of users, provide all-weather, all-round humanized companionship and intelligent services, and become a super entrance for people to the world of virtual and real integration.
  No matter which application direction, for the digital human industry, it will be a potential market with a scale of 100 billion. Then, how to organize resources and ecology, promote the application and development of digital technology in the digital human industry, and improve the production efficiency of digital human and the intelligence level of digital human will be an important proposition.