It was only a few months ago, in late December, that China released draft provisions on “human-like interactive AI” services for public comment, so it was something of a surprise when a new, similar-sounding draft on “digital virtual human” services was released on April 3. Despite their related titles, however, the documents address very different, albeit overlapping, technologies and risks.
Scope
The earlier “human-like AI” draft addresses AI services that simulate human personalities, thinking, and communication to interact emotionally with human users through text, images, audio, video, or other media.
The new “digital human” draft is concerned with visual representations of humans, operated either by a human or a machine, that exist only in the virtual world and have characteristics such as speech, personality, OR interactive capacity.
By these definitions, a text or voice-only chatbot could be a human-like AI, but not a digital virtual human. By contrast, a virtual news anchor, pop star, or livestreaming avatar can be a digital-human even if it isn’t automated at all, but merely controlled by a human puppeteer via motion capture sensors and filter overlays. Virtual digital humans also do not need to be interactive, and might just appear in online content.
The definitions are not mutually exclusive. Article 18 of the new digital virtual humans draft acknowledges situations where digital virtual humans (likenesses) are used to provide human-like AI services (interactive machine relationships); ie, an interactive AI that includes a visual human avatar
In sum, the human-like AI draft is about machines (AI) that act like humans, and the psychological impact of users’ relationships with them. The digital human draft is about regulating visual portrayals that look like humans as a new medium for conveying information online.
The finalized rules for Human-like AI were released shortly after this article was published, and are available here.
Regulatory focus:
The earlier human-like AI draft is an ambitious first effort to construct an overarching framework for a new type of human-machine relationship that blurs the line between tool and companion, but also brings novel risks to users. It focuses on service providers’ liability for the AI’s design and activity, emphasizing requirements to protect user mental-health and prevent excessive dependence.
The digital-human draft is more typical of other Chinese regulations addressing new means of sharing information online. China often releases technology-specific regulations (such as on instant messaging, livestreaming, online fiction, and even user comment sections) that incorporate and apply broader rules to the specific new context. They can also assign regulatory responsibilities among the relevant state agencies and introduce new enforcement mechanisms such as administrative penalties.
The digital-human draft thus focuses on digital virtual humans primarily as digital works and means of conveying information, rather than the unique challenges of their ‘humanity’. It incorporates rules for protecting personal information, personality rights (no clones of people without their consent), intellectual property, and data security in the creation of use of such avatars. It adapts content moderation requirements and other regulatory obligations that apply to broader media, such as algorithm filing requirements. Where violations are not covered by other laws, Article 24, authorizes a number of penalties including fines of up to 200,000 RMB, and shutting down the service.
Rather than go through each obligation, I have loosely annotated a translation below, linking to some prior sources of the protections and innovations.
First, however, it is worth considering a few of the more unique provisions:
- Government digital persons:
Article 19 addresses the use of digital virtual persons to provide government services or participate in public management or judicial activities. Digital avatar civil servants and even judges have already been put into service, although their functions are usually limited to providing public information rather than decision-making.
The new draft requires that
- Citizens can opt out of the digital virtual human service.
- There are mechanisms for manual oversight, and
- their use should be guided by the principles of necessity and reasonableness (as well as legality).
These requirements feel like a chastening reminder that not every situation needs automation, and that automation is a supplement, rather than a substitute, for government responsiveness to the public.
- Identity Concealment
Article 11(5) addresses the use of digital virtual persons to evade legally required identity verification systems using tech such as facial or voice recognition. Providing and verifying real-name identity information is a requirement for receiving many online services in China, but this provision adds a more specific focus on the unique use of digital humans to evade biometrics.
This article, and Article 11(7)’s recitation of existing prohibitions on improperly registering and trading accounts, are also reminiscent of the recent draft Cybercrime Law’s efforts to reinforce the real-name system.
- Protections for Human Operators
Article 11(6) contains a brief acknowledgment of the rights of the human drivers of some digital humans. It bars infringement of the drivers’ rights and interests, such as in their personal information or the right to freely choose employment.
This brief provision may conceal some big issues. As with traditional performers, the real person operating a digital human may become inextricably associated with the character they perform. And in an employment relationship, the blurred identity of actor and character, operator and avatar, can create difficult questions.
For example, when does actors’ subsequent work infringe the employer’s rights in a proprietary character? Is Harrison Ford infringing on the Indiana Jones IP if he wears a fedora (or carries a whip, etc.) to film an unrelated cola ad? Are there any limits on what the employer can portray the character, with Mr. Ford’s likeness, doing?
These issues aren’t new to digital humans, but could become more prominent. A digital human may be created by capturing a performer’s face, voice, motions, or other biometric information, making it even more visually associated with the character. At some point, disputes over control of a digital human may become disputes over the performer’s personal information, reputation, and freedom to choose future work.
Protection of Minors.
Many of the requirements of Article 10 incorporate existing protections for minors online:
- No inducements to spending (Law on the Protection of Minors 74)
- No inducements to religion (Internet Religious Information Services regs art. 14)
- Prohibit content that might impact minors’ physical or mental health. (Measures for Categorizing Online Information that Might Impact Minors’ Physical and Psychological Health)
- Prevent Addiction (Regulations on the Protection of Minors Online, 42)
More interesting, however, are requirements that minors must not be provided with emotionally resonant digital virtual human services, such as virtual family or companions. This seems to address the emotional dependence risks that are more typical of the Human-like AI provisions than this draft regulation. The draft of those provisions, however, stopped short of banning the provision of AI-driven companions to minors, only requiring additional guardian controls. (Article 12) The closest proposal is a prohibition of AI-simulated relations for elderly users, not minors (Article 13), in a paternalistic effort to prevent fraud.
One way to make sense of this distinction is that the concern here might not be emotional attachment at all, but a different danger unique to human-operated digital persons: adults using virtual avatars as disguises to befriend or groom minors.
Labeling Requirements:
Article 13 creates an obligation to prominently label digital virtual human services with the text “digital human” written in the area where the avatar is displayed for the entire duration of the service. This is a new labeling requirement, beyond the more detailed labeling requirements for AI-generated content, which still apply separately.
Even though virtual digital human services are not necessarily AI-generated or controlled, most will fall within the scope of China’s deep synthesis regulations, that covers tools for generating or editing faces and voices, and also include labeling requirements. (articles 16,17). The AI labeling rules require both ‘explicit’ labels, which can be perceived by humans, and “implicit” metadata labeling. As in the AI-labeling rules, the service providers, users, and information publication platforms all bear some responsibility for this new labeling requirement.
6, Additional annotations
| Chapter II: Protection of Rights and Interests | NOTES |
| Article 7: Where any organization or individual uses natural persons’ sensitive personal information for modeling, image generation, scene construction, or other activities, they shall obey the laws and administrative regulations and meet the following requirements: | Personal information protection requirements |
| (1) Obtain the natural persons’ independent consent, and truthfully, accurately, and completely notify them, in a conspicuous fashion using clear and easily understood language, of the purposes of the handling, its necessity, and its impact on personal rights and interests, as well as other matters provided for by laws and administrative regulations; and where the personal information of minors less than14 years-old is used, the independent consent of their parents or other guardians shall be obtained. Except as otherwise specified by laws and administrative regulations. | 7(1) recites the requirements for obtaining independent informed consent to handle sensitive personal information, including giving conspicuous and understandable notice of the purpose, necessity, and impact on rights from handling the information. (PIPL 17, 28-31)
|
| (2) After a natural person withdraws their consent, employ methods such as deletion of relevant personal information to eliminate the impact, and personal information must not be retained in any form or used for other purposes, except as otherwise provided by laws and administrative regulations. Except where parties have otherwise agreed, the digital virtual human shall also be cancelled. | 7(2) requires digital virtual human service providers to delete sensitive personal information after consent is revoked in accordance with PIPL article 47. It goes slightly further, however, by saying that the digital virtual human itself may be cancelled, unless the parties otherwise agreed. |
| (3) Respect the lawful rights and interests of others; the rights and interests of others in their image, reputation, honor, privacy, and personal information must not be infringed. Where the personal information of the deceased is used to carry out related activities, the deceased’s relatives may lawfully exercise corresponding rights in the deceased’s personal information for their own lawful and legitimate interests, except where the deceased made other arrangements before their death. | 7(3) addresses the rights of family members in deceased relative’s personal information in accordance with PIPL article 49, which allows them to seek access, reproduction, correction, or deletion of the information, unless the deceased made agreements to the contrary before death. Curiously 7(3) goes beyond the PIPL to address surviving family members’ rights to exercise broader personality rights beyond personal information rights, a concept now found in Civil Code Article 994. |
| Article 8: Any organization and individual providing or using digital virtual human services must not malign, deface, or infringe on the personality rights of others in any form, and digital virtual human services that can identify the identity of a specified natural person must not be provided without the consent of that specified natural person. Including but not limited to: | Article 8 is largely an adaptation of existing civil code provisions preventing unauthorized and harmful uses of individuals names and appearances. (Civil Code 1017, 1019)
One potentially new wrinkle is that the article incorporates and idea similar to the definition of personal information- information from which a natural persons’ identity may be determined. It says that where a digital virtual human service is similar enough in name or likeness to a real human that it identifies that person, their consent must be acquired before the service is provided to anyone. This language isn’t typical of personality rights discussions, but past rules on deepfakes have required consent before manipulating persons faces and voices. (Deep Synthesis Regulations Art. 14) |
| (1) Using the pen name, stage name, online alias, translated name, brand, name, or shortened name of other persons that have a definite degree of social recognizability; | |
| (2) Using likeness or voice that is highly similar to specified natural persons. | |
| Article 9: Any organizations or individuals providing or using digital virtual human services shall respect intellectual property rights and commercial ethics. The intellectual property rights lawfully enjoyed by others must not be infringed upon in using others’ text, art, photography, music, audio-video, or other works or pieces to produce digital virtual humans, as well as during the use of digital virtual humans by any type of entity. | Article 9 simply states that intellectual property rights must be respected in the creation and use of digital virtual humans. This follows earlier more general provisions on generative AI products, but provides little guidance on what constitutes an IP violation, ie: What is the scope of fair use? (Generative AI Regs, article 4(3), 7(2)).
|
| Article 10: It is prohibited to entice minors to become addicted to digital virtual human services. Minors must not be provided with virtual relatives, virtual companions, and other virtual close relations, digital virtual human services that induce excessive spending or religious belief, as well as digital virtual human services that contain information that might impact the physical and psychological health of minors such that which might lead or induce minors to imitate unsafe behavior, carry out acts that violations social mores, produce extreme emotions, or cultivate bad habits. | Minors No virtual close relations
No inducements to spending (Law on the Protection of Minors 74)
No inducements to religion (Internet Religious Information Services regs art. 14)
|
| Chapter III: Regulation of Services | |
| Article 11: Any organization or individual providing or using digital virtual human services shall obey the laws and administrative regulations, respect social mores, ethics, and morality, and must not engage in any of the following activities: | Providers and users prohibited conduct:
|
| (1) Generating or disseminating content that endangers the nation’s security, honor, and interests; incites subversion of state sovereignty or the overturning of the socialist system; incites separatism or undermines national unity; advocates terrorism, extremism, or historical nihilism; goes against the Core Socialist Values, public order, and good custom; carries out illegal religious activities; promotes ethnic hatred and discrimination; stirs up opposition between groups, disseminates obscenity, pornography, gambling, violence, or the instigation of crimes; spreads rumors; or insults or defames others, harming the lawful rights and interests of other.; | Illegal content categories; Regulation of Online Information Content E-cosystem art. 6)
|
| (2) Including content that harms the nation’s image in the design, dress or symbols, personality or preferences of digital virtual humans; | 2) character design that harms nation’s image. Similar to rules for online account names. Also similar to removed |
| (3) Distorting or maligning the images of persons such as heroes and martyrs, twisting or altering the deeds and spirit of heroes and martyrs, using the images of persons such as heroes and martyrs for commercial purposes; | Heroes martyrs, leaders Law on Protection of Heroes and Martyrs 22
|
| (4) Engaging in illegal activities such as false advertising for goods or services, maliciously inducing purchases, or telecommunications fraud; | Fraud and advertising violations Measures on the Administration of Internet Advertising Law on Countering Telecommunications Network Fraud
|
| (5) Using virtual digital humans to evade facial recognition, voice recognition, or other identity authentication mechanisms when laws or administrative regulations require that real identity information is to be provided; | Evading online protection systems |
| (6) Infringing on the personal information rights, right to choose one’s profession, or other lawful rights and interests of the real-person driver of a human-driven virtual human; | Protection for human driver. |
| (7) Improperly registering or trading internet accounts; | Evasion 2 |
| (8) other conduct in violation of laws and administrative regulations. | |
| Article 12: Any organization or individual providing or using digital virtual human services shall employ measures to consciously defend against and reject the generation and transmission of content with a negative impact on the online ecology, such as sexual innuendo or suggestive flirtation; horror or cruelty, or incitement of group discrimination. | Negative information prevention |
| Article 13: From the start of digital virtual human services, the digital virtual human service providers, service users, and the providers of online information content transmission services shall continuously display a conspicuous alert containing the text “digital human” in the area where the digital virtual human is presented for the entire process, and are to comply with state provisions related to the labeling of AI generated or synthesized content. | Labeling: Continuous throughout service Compare: Measures for Labeling of AI-Generated Synthetic Content |
| Article 14: Digital virtual human service providers and service users shall follow laws and administrative regulations to carry out data handling activities for specified purposes and within a specified scope, use data from lawful sources and implement responsibility for data security protections, employ suitable technical measures and other necessary measures to ensure the security of data storage and transmission, to prevent the leakage or improper use of data. | Providers and users: DATA Handling – follow laws – lawful sources – protections
Data Security Law, eg: articles 27, 32 |
| Article 15: Digital virtual human service providers and service users shall establish mechanisms such as for monitoring digital virtual human services’ security risks, early warning and emergency response, and prevention of addiction, and establish and improve systems for managing content orientation; allot technical capacities and personnel corresponding to the scale of operations, employ methods that combine technical measures such as AI and big data with manual review; strengthen the identification, monitoring, and early warnings for digital virtual human services, and record and store log information. | Providers and users: Risk Management
Must have appropriate means and staff to review,
|
| Where it is discovered that digital virtual human services are being used to engage in unlawful activities, measures such as dynamic identity verification, warnings, limited functions, or terminating services shall be promptly employed; and where major risks are discovered, the digital virtual human services shall be immediately suspended or terminated, deregistering the digital persons and eliminating the impact. | Violations using DVH Consequences of suspending services.
|
| Article 16: Digital virtual human service providers shall sign service agreements with technical support providers and service users, clarifying and ensuring content safety as well as content on the rights and obligations, such as data collection, use, and storage. | User and tech agreements on safety obligations |
| Article 17: Service providers that provide services for the transmission of online information content shall establish and improve systems for managing the orientation of content, allocate technical capacities and personnel corresponding to the scope of operations, perform content safety management responsibilities, optimize content review and account management responsibilities, strengthen management of user-published information, promptly address illegal and negative information, and store log information. | Content transmission- must have content orientation management , necessary staff etc.
|
| Article 18: Where digital virtual human services are used to provide human-like interactive artificial intelligence services, and users select to cancel specified service functions or leave the service, they must not be tricked or excessively induced to continue the services. | Alignment with human-like provisions -no obstacles to exit |
| Encourage the adoption of necessary measures to conduct active intervention and professional assistance for user inclinations towards suicide, self-harm, or other threats to health and life. | Human-like dependence protections |
| Article 19: When digital virtual human services are used in areas such as government services, public management, and judicial activities, the principles of legality, reasonableness, propriety, and necessity shall be complied with, mechanisms for manual oversight review shall be set up; and users shall have the right to choose whether to accept or reject the digital virtual human services. | In official settings, users can reject DVH |
| Chapter IV: Oversight Inspections and Legal Responsibility | |
| Article 20: Digital virtual human service providers and the providers of services for the transmission of online information content shall establish mechanisms for digital virtual human users’ appeals and public complaints and reports, set up convenient portals for appeals, complaints, and reports, and promptly address them and give feedback on the outcome. | Providers- Report and complaint portals |
| Article 21: Digital virtual human service providers that have public opinion properties or the capacity for social mobilization shall perform formalities for filing algorithms and the modification or cancellation of filings in accordance with the Provisions on the Management of Algorithmic Recommendations in Internet Information Services. | Public opinion properties, File algorithms
|
| Digital virtual human technical support providers are to perform filings, modifications, and cancellations of filings with reference to the preceding paragraph. | Tech support- also filings |
| Article 22: Digital virtual human service providers providing internet information service with a public opinion character or having the ability to mobilize the public shall carry out security assessments in accordance with relevant state provisions. | Security assessments for public mobilization |
| Article 23: Cybersecurity and informatization departments and relevant departments are to carry out oversight inspections of digital virtual human services on the basis of their duties. Digital virtual human service providers, technical support providers, service users, and the providers of services that provide services for the transmission of online information content shall cooperate in accordance with law and provide the necessary support and assistance. | Oversight inspections by CAC. Cooperation by involved parties. |
| Article 24: Where these provisions are violated and laws and administrative regulations have other provisions, punishment is to be given in accordance with those provisions and civil liability is to be borne in accordance with law; where laws and administrative organs are silent, the relevant departments are to give warnings, circulate criticism, or order that corrections be made in a set period of time on the basis of their duties; and where corrections are refused, or the circumstances are serious, they are to order that the provision of services stops, and may give a concurrent fine of between 10,000 and 100,000 RMB; and where it involves harmful consequences endangering citizens’ life or health, a concurrent fine of between 10,000 and 200,000 is to be given. | General Punishment authority in the absence of other law. |
Jeremy Daum is a Senior Fellow of the Yale Law School Paul Tsai China Center, based in Beijing, with over a decade of experience working in China on collaborative legal reform projects. His principal research focus is criminal procedure law, with a particular emphasis on protections of vulnerable populations such as juveniles and the mentally ill in the criminal justice system, and is also an authority on China’s ‘Social Credit System’. Jeremy has spoken about these issues at universities throughout China and in the U.S.; and has co-authored a book on U.S. Capital Punishment Jurisprudence for Chinese readers. He is also the founder and contributing editor of the collaborative translation and commentary site Chinalawtranslate.com, dedicated to improving mutual understanding between legal professionals in China and abroad.
He translates, writes, edits, does web-design, graphic design, billing, tech support, and social media outreach for China Law Translate.
- Jeremy Daum
- Jeremy Daum
- Jeremy Daum
- Jeremy Daum
English
中文(简体)
Be First to Comment