Supporters of Marcus Endicott’s Patreon can access weekly or monthly video consultations on this topic.
China has embarked on what is arguably the most ambitious state-coordinated deployment of digital humans in the world, weaving AI-driven avatars into the fabric of governance, education, healthcare, cultural preservation, and commerce at a pace and scale that no other country has matched. What distinguishes the Chinese approach is not merely the sophistication of the underlying technology but the deliberate integration of digital humans into national industrial policy, beginning with Beijing’s August 2022 Action Plan, the first government-level digital human strategy issued anywhere in the world. The result is an ecosystem in which local governments, state broadcasters, technology giants, and small enterprises alike treat the digital human not as a novelty but as infrastructure.
Across China’s government service halls, AI avatars have become an increasingly routine interface between the state and its citizens. A nationwide rollout documented by People’s Daily in April 2025 confirmed deployments in cities including Beijing’s Changping district, Quanzhou in Fujian province, Tongliao in Inner Mongolia, and Wuxi in Jiangsu. Quanzhou became Fujian’s first city to deploy a DeepSeek-powered digital human for civic consultation, fielding queries on policy, social security, real estate, healthcare, and enterprise services through a WeChat public account and more than one hundred village-level service terminals launched in March 2025. Guangzhou has pursued a parallel track, with its Panyu District developing a 5G-enabled government services avatar that won a 2022 provincial innovation award, and its Human Resources Service Center earning recognition as one of China’s top ten government service AI innovation cases in 2024. The Guangzhou Bar Association has used AI digital humans to produce multilingual legal service videos for the Canton Fair, while the city’s justice bureau offers AI legal consultation through its official website. WeChat, mini-programs, mobile apps, and physical kiosks all serve as delivery channels, though it is worth noting that these legal assistance tools function as consultation and education aids rather than formal legal representation.
The regulatory architecture surrounding these deployments is anything but thin. China was the first country in the world to enact binding regulations specifically governing generative AI, a distinction confirmed by the International Association of Privacy Professionals. The regulatory sequence began with the Algorithm Recommendation Provisions that took effect in March 2022, followed by the Deep Synthesis Provisions in January 2023, which directly address deepfakes and require consent for biometric editing, mandatory content labeling, algorithm registration, real-name user verification, and criminal liability for serious violations. The Generative AI Interim Measures followed in August 2023, and the AI Content Labeling Measures that took effect in September 2025 introduced a dual-label system combining explicit markers with implicit metadata and watermarks, a framework more prescriptive than comparable approaches in the European Union or the United States. By October 2025, China had approved thousands of algorithm filings. On April 3, 2026, the Cyberspace Administration of China released draft Administrative Measures for Digital Virtual Human Information Services, with Article 19 specifically addressing digital humans in government and judicial services by requiring human oversight and preserving citizens’ right to refuse automated interaction. The public comment period for those draft measures closes on May 6, 2026. China’s first virtual digital human infringement case, MoFa (Shanghai) Information Technology versus a Hangzhou network company, was decided in 2023 with damages of 120,000 yuan. To characterize this regulatory environment as limited would be to misread the evidence entirely.
In education, the integration of digital humans remains at an earlier and more uneven stage, though the trajectory is clear. The 2025 World Digital Education Conference, organized by the Ministry of Education and held in Wuhan in May 2025, showcased AI-in-education innovations that included digital human applications. Faculty at Wuhan City Vocational College have publicly discussed AI digital teachers, AI online teaching assistants, and intelligent learning companions as part of the evolving classroom toolkit. Henan province, whose capital is Zhengzhou, published guidance through its education department on generative AI-driven virtual digital human teaching resource construction in September 2024. Companies such as Shiyou Technology, Keling, and Huaqiyun offer commercial digital human teaching solutions, and East China Normal University has developed large language model-driven digital humans for teacher training. Zhejiang province operates a digital education platform with more than 5.5 million resources and over seven thousand synchronized courses that incorporate AI-powered personalized learning pathways, and in April 2025 Zhejiang launched AI-assisted education reform trials. Adaptive learning tools from companies like iFlytek are being deployed in schools across the country. These are genuine and significant developments, though they remain pilot and demonstration-stage projects rather than standardized features of the national curriculum.
Healthcare presents some of the most compelling use cases. On February 26, 2025, Fudan University’s affiliated Zhongshan Hospital, working with the Shanghai Scientific Intelligence Research Institute, released the beta version of CardioMind, described as China’s first cardiovascular specialty AI medical large model. Led by Academician Ge Junbo, a member of the Chinese Academy of Sciences and director of Zhongshan Hospital’s cardiology department, CardioMind integrates multimodal data from electrocardiograms, ultrasound, and laboratory tests with top specialists’ clinical experience to support the full care process from history-taking to assisted diagnosis. The model was trained on hundreds of thousands of electronic medical records. In Shandong province, hospitals have adopted AI medical tools with growing enthusiasm. Shandong University’s Qilu Hospital deployed a DeepSeek-based model and created a round-the-clock intelligent health assistant agent for its internet hospital, along with an AI virtual patient for clinical training. Shandong Provincial Third Hospital integrated iFlytek’s Spark model for interactive consultation, and Shandong No. 1 Medical University’s First Affiliated Hospital deployed DeepSeek for pathology diagnostics. The most prominent digital human doctor deployments to date, however, have concentrated in Beijing’s Haidian district community health centers, in Shanghai where official policy mandates the use of digital human or digital twin physician assistants, and in Shenzhen.
Elder care has emerged as one of the most urgently needed and extensively documented domains for AI companion deployment. China’s population aged sixty and above reached 310 million by the end of 2024, representing twenty-two percent of the total, with projections exceeding 400 million by 2035 and a nationwide shortage of 5.5 million elder care workers. The policy response has been direct: in December 2024, the CPC Central Committee and State Council issued guidelines calling for the acceleration of humanoid robots and AI in elderly care, and in July 2025, China released its first national White Paper on Smart Elderly Care in Chongqing. Specific deployments have proliferated. The companion robot Peipei, deployed in a Chongqing welfare institute, achieves ninety-two percent emotion recognition accuracy, produces thirty-two facial expressions, speaks in Chongqing dialect, and can replicate family members’ voices, with plans to expand to one million users within three years. Xinhua reported on the Yang Yang robot at Pacific Care Home in Chengdu and the Xia Lan humanoid robot at a Shenzhen nursing home, both in March 2025. In Beijing, the home companion robot Xiao Li provides medication reminders and fall detection, and more than seven hundred households have been equipped with care robots. Baidu’s Five Blessings AI Elder Assistance program has been deployed in three Beijing Dashilan communities using Xiaodu smart screens, while Tencent has spent three years testing an AI fall detection system at a Shenzhen nursing home with over ninety percent accuracy. UBTECH’s Xiaobao robot was developed in partnership with China’s National Committee on Aging. The market for smart elderly care is estimated to exceed fifty billion yuan in 2025, growing at more than thirty percent annually.
Cultural heritage preservation has become one of the most publicly visible arenas for digital human technology in China. The opera master Chang Xiangyu, legendary founder of the Chang school of Yu Opera who lived from 1923 to 2004, has been digitally resurrected through multiple independent projects. In October 2024, APUS and Henan Broadcasting used diffusion transformer video generation technology to revive Chang Xiangyu from black-and-white film footage for a cultural documentary in which the digital figure interacts with contemporary opera performers across generations. Separately, a student innovation team at Henan Finance University spent three years creating a comprehensive Chang Xiangyu AI digital human using five AI models for image generation, body construction, voice restoration, and audio-video fusion, drawing on more than five hundred hours of audio-video archives and incorporating 5G holographic technology. The Song Dynasty poet-warrior Xin Qiji, one of the greatest ci lyricists, who lived from 1140 to 1207, has likewise been recreated through AI. An AI short film produced by Jichuan AI Studio around October 2025 depicted Xin Qiji’s encounter with Yue Fei, using multiple AI platforms for character design and dynamic generation to achieve near-CG-quality results. These cultural resurrection projects exist within a broader ecosystem of digital humans on China’s most-watched stage. At the 2024 Spring Festival Gala, an AI virtual Li Bai appeared at the Xi’an venue, reciting classical poetry before tens of thousands using optical motion capture. The 2025 Spring Gala featured AIGC technology animating ancient paintings of opera characters alongside digital humans and extended reality virtual production. The 2026 Spring Gala introduced the first digital human concept performance using volumetric video capture for hyper-realistic 3D rendering. Museums have joined the movement as well, with the National Museum’s AI guide and the Capital Museum’s digital docent both confirmed by Xinhua in August 2025, and AI digital humans serving as virtual narrators in anti-corruption education bases and red history exhibits. Shanghai’s Jing’an District documented an AI digital human revival of revolutionary figure Xiang Jingyu for ideological education in January 2026, and MoFa Technology (魔珐科技) provides pre-built templates for Party building, governance, and red education content.
The commercial applications of digital humans in China have scaled with remarkable speed, particularly in livestream e-commerce. Baidu’s Huiboxing platform, the company’s full-stack AI digital human livestream solution, has created more than 320,000 avatars and over 120,000 voice profiles. It offers AI-powered script generation using Wenxin and DeepSeek models, voice cloning from a fifteen-minute recording, avatar cloning from a five-minute video with ninety-eight percent lip-sync accuracy, and autonomous round-the-clock livestreaming with AI-driven question-and-answer capability at a ninety-five percent usability rate. IDC ranked it first in the e-commerce livestream digital human category. During the 2025 Double Eleven shopping festival, eighty-three percent of streaming merchants used Huiboxing, and gross merchandise value rose ninety-one percent year over year. The platform has already expanded to Brazil and is planning entry into the United States and Southeast Asia. Huawei Cloud’s Flexus MetaStudio product offers a turnkey solution at 999 yuan, roughly 140 US dollars, designed for small and medium enterprises that lack technical resources, providing script writing integrated with DeepSeek models, voice cloning from a one-minute audio sample, and 1080p video production with ninety-five percent lip-sync accuracy. By mid-2025, virtual digital humans comprised nearly twenty-three percent of the overall digital human application market. One operator achieved 1.9 billion yuan in annual sales using a single digital human account, and the digital human of livestream commerce celebrity Luo Yonghao generated more gross merchandise value in twenty-six minutes than his real self managed in one hour.
The rural dimension of this commercial expansion deserves particular attention. In Guangdong, the Zhoutian E-Commerce Town deployed AI digital humans for round-the-clock agricultural product livestreaming and was named a national county-level livestream e-commerce center typical case by the provincial commerce department. In Guizhou’s Bijie, Baidu created a digital human named Pan Fang to sell deer antler products, tripling online sales. In Xinjiang’s Shufuxian, the local government partnered with China Telecom to hold AI digital human e-commerce training for more than 150 rural participants. In Beijing’s Tongzhou district, a Ministry of Agriculture program trains farmers in digital human livestreaming as a new agricultural tool. Shanghai’s Chongming district won recognition for digital human livestreaming of agricultural products. These deployments represent a deliberate policy effort to extend the commercial benefits of digital human technology beyond the coastal technology hubs and into the agricultural economy.
The intellectual property landscape around digital humans is evolving in parallel with the technology. In June 2023, the China Digital Human IP Certification and Protection Platform launched at the 13th China International Trademark Brand Festival in Dongguan, initiated by a subsidiary of the National Intellectual Property Administration’s IP Press, offering one-stop services for certification, copyright protection, digital identity management, and licensing. In December 2023, a separate digital human IP database launched under the guidance of Communication University of China’s Digital Human Research Institute, providing IP cards, influence assessment, and brand cooperation services. These are industry platforms for evidence preservation and certification rather than formal administrative registration systems comparable to trademark or patent registration, but they represent a purposeful institutional effort to create rights infrastructure for a new category of digital asset. Patent activity in related technologies is extensive globally, with significant filings in AI lip synchronization, emotional modeling, and related domains across both Chinese and Western jurisdictions.
Western technology companies operate in the same technological domain but through a fundamentally different strategic logic. Epic Games’ MetaHuman framework, which exited early access with Unreal Engine 5.6 in June 2025, provides free tools for creating photorealistic digital humans, including a cloud-based creator, facial performance capture that now works with standard webcams, and scan-to-character conversion. Its licensing expanded in 2025 to allow characters in competing engines and digital content creation tools. MetaHuman produces visual characters of extraordinary fidelity but does not itself include AI speech, intelligence, or interactivity, which require separate integration. NVIDIA’s Avatar Cloud Engine provides that complementary layer through its suite of AI technologies for interactive digital avatars, including real-time lip synchronization, speech recognition, and conversational AI, with partnerships spanning gaming studios and enterprise applications in healthcare, financial services, and retail. Synthesia, the leading AI avatar video platform, offers more than 240 stock avatars in over 160 languages, serving more than one million users and fifty thousand teams, though its output is pre-recorded video content rather than the real-time interactive avatars that characterize Chinese platforms like Baidu Huiboxing and Tencent Cloud Avatar Studio. North America still holds the largest share of global digital human revenue, at roughly forty-three percent in 2025, and Western technology, particularly NVIDIA GPUs and Epic’s rendering frameworks, underpins much of China’s own ecosystem. The Western approach is market-driven, modular, and platform-based, selling tools and APIs to developers rather than orchestrating consumer-facing avatar deployments through state coordination. This represents a strategic choice rather than an absence of capability, though it does mean that no Western country has produced anything resembling the integrated, policy-driven, society-wide deployment that China is pursuing.
The ethical terrain is contested on both sides of the divide, though the specific flashpoints differ. In the West, therapy bots have become a source of acute concern. Lawsuits filed against Character.AI in December 2024 alleged that the platform caused self-harm and isolation in minors, and by June 2025, roughly two dozen digital rights organizations had filed an FTC complaint urging investigation of Character.AI and Meta for what they characterized as unlicensed practice of medicine. A Brown University study in October 2025 found that AI chatbots systematically violate core mental health ethics standards. Several US states, including Illinois, Nevada, and Utah, have banned AI from providing therapy, with California, Pennsylvania, and New Jersey drafting similar legislation. The crisis around sexualized avatars and nonconsensual deepfakes has been even more far-reaching. The Grok platform operated by xAI generated nonconsensual sexualized deepfakes, including of minors, triggering formal investigations by the European Union and the United Kingdom’s Ofcom, bans in Malaysia, Indonesia, and the Philippines, and a California attorney general investigation. Sixty-one privacy regulators across four continents jointly warned AI companies about the practice. The TAKE IT DOWN Act, signed on May 19, 2025, became the first US federal law criminalizing the distribution of nonconsensual intimate deepfakes. UNICEF research found that twenty-eight percent of AI companion apps feature female-presenting characters designed for romantic or sexual interaction with widespread absence of age verification. Intellectual property enforcement has likewise intensified. Disney has pursued cease-and-desist actions against ByteDance over its Seedance 2.0 tool in February 2026, filed a major copyright infringement lawsuit with Universal against Midjourney in June 2025, demanded that Character.AI remove unauthorized chatbots in September 2025, and accused Google of infringing its copyrights on a massive scale through AI-generated character images. Soul Machines, the New Zealand-based digital human company that had raised more than 135 million dollars in venture capital, filed for receivership on February 5, 2026, owing 19.5 million dollars to creditors, illustrating the commercial fragility of Western digital human ventures even at significant scale.
The contrast between the Chinese and Western approaches to digital humans is not one of capability versus incapacity but of coordinated national strategy versus distributed market innovation. China’s digital human market was valued at 418 million dollars in 2024 and is projected to reach 4.1 billion by 2034 at a compound annual growth rate of more than twenty-seven percent. The European research landscape includes meaningful work on avatar accessibility, particularly through EU Horizon 2020 projects developing sign language translation avatars, but these remain at the research and pilot stage rather than constituting government-wide deployments. China’s trajectory is shaped by explicit policy mandates, centralized regulatory frameworks, massive demographic pressures in elder care, a cultural disposition toward state-led technology adoption, and the sheer scale of its consumer digital economy. Whether the result will be a model that other nations seek to emulate or a cautionary demonstration of the risks inherent in pervasive AI-mediated interaction is a question that only the coming years will answer. What is already beyond dispute is that China has moved from experimentation to deployment at a speed that has no parallel elsewhere in the world.
[Apr 2026]