ChatGPT has recently exploded in popularity.

As an artificial intelligence chatbot developed by OpenAI, this program has been highly praised by many professionals, who consider it a revolutionary product.

Compared to previous chatbots, ChatGPT resembles an ideal AI consultant. You can make various practical requests to ChatGPT using natural, conversational language. For example, if you ask it to write a video script, it will list key points based on the topic; if you ask how to systematically learn and understand a subject, it will outline a relatively clear knowledge framework; even if you have no programming background, you can use ChatGPT to write a program that runs immediately.

Using ChatGPT to write a tic-tac-toe game

1 You Can Even Make Money by Writing Mini Programs with ChatGPT

While chatting with a friend, I learned that he had already started making a profit from code generated by ChatGPT. While marveling at how easily he earned money, this also prompted my thoughts: Ordinary mini-program code is protected as a work under copyright law, but can code generated by ChatGPT be protected as a copyrightable work? Moreover, does the copyright in code generated by ChatGPT belong to my friend? If not, does using this mini-program for profit infringe upon the rights of others? The copyright issues arising from this series of ChatGPT outputs are essentially the long-standing problem of protecting AI-generated content.

2 Can AI-Generated Content Constitute a Work?

Whether AI-generated content can constitute a work under copyright law is highly controversial in both academia and existing judicial practice, with conclusions even directly opposing each other.

(A) Legal Provisions on Work Qualification

At the level of statutory law,

Article 3 of the Copyright Law : A work as referred to in this Law means an intellectual achievement that is original and can be expressed in a certain form in the fields of literature, art, and science. According to this definition, to constitute a work under copyright law, at least three elements must be satisfied:

An intellectual achievement in the fields of literature, art, and science; Possessing originality; Capable of being expressed in a certain form.

However, this article does not explicitly stipulate that creation by a natural person is a constitutive element of a work. Nevertheless, some local judicial documents, such as The Trial Guidelines for Copyright Infringement Cases of the Beijing High People’s Court, explicitly require that the creation must be based on the creative output of a natural person to be recognized as a work .

For instance, Guideline 2.1 [Review of Whether a Work is Constituted] of the aforementioned guidelines clearly provides: In examining whether the object for which the plaintiff claims copyright constitutes a work, the following factors shall generally be considered:
(1) Whether it belongs to a creation by a natural person within the fields of literature, art, and science;
(2) Whether it possesses originality;
(3) Whether it has a certain form of expression;
(4) Whether it is reproducible.

(B) Academic Views on Whether AI-Generated Content Can Constitute a Work

In academia, there are many different views on whether AI-generated content can constitute a work.

1. AI-Generated Content Does Not Constitute a Work

Professor Wang Qian , starting from the process of generating AI content, argues that so far AI-generated content is the result of applying algorithms, rules, and templates, and does not constitute creation; it cannot reflect the creator’s unique personality and therefore cannot be recognized as a work, thus there is no need to discuss authorship.[3]

Professor Qiu Rungen insists that creation by a natural person is a prerequisite for constituting a work.[4]

Professor Liu Yinliang , while adhering to this view, questions the ideological value and emotional significance of AI-generated content as works, arguing that it does not help achieve the basic goals of the copyright system and should not be protected as works.[5]

Professor Tao Qian , based on the subjective authorial standard of originality, believes that recognizing AI-generated content as works would violate traditional copyright jurisprudence, and therefore such content does not constitute works.[6]

2. AI-Generated Content Can Constitute a Work

The “instrument theory” represented by Professor Li Yang argues that AI is merely a tool assisting human creation; the real subject engaging in creation in the process of generating AI content remains human. As long as the requirements for a work are met, it constitutes a work.

On this basis, Professor Li Yang further reinforces this argument from the perspective of Kantian philosophy, specifically the “unity of subject and object epistemology” and the notion that “man is an end.” Professor Li Yang believes that no matter what stage AI develops to, it can only be an object and tool used by humans and can never be equated with humans in status. Whether AI-generated content constitutes a work should be judged according to the requirements for a work, and rights should be attributed to the AI designer or user under current copyright theory.[7]

The “will representation theory” represented by Professor Xiong Qi , drawing on the institutional arrangement for works made for hire (legal persons), argues that “the results generated by AI can be regarded as creative acts representing the will of the designer or trainer, and therefore, based on the rationale that creation represents the will of the owner, copyright should belong to the owner of the AI.”[8]

The “creator theory” argues that “anthropocentrism” should be abandoned and an objective reader-oriented criterion of originality should be adopted. As long as the AI-generated content is independently completed by the robot and meets the requirements for a work, it should be protected by copyright.

Since AI cannot exercise rights or assume responsibilities like a legal person or natural person, the rights should be exercised by its creator, i.e., the AI’s creator or owner.[9]

The “fruits theory” represented by scholar Huang Yuye , from a civil law perspective, argues that AI-generated content is the fruits (fructus) produced by AI, a new thing independent of the original thing. This view is similar to the “instrument theory.”[10]

(C) Practical Views on Whether AI-Generated Content Can Constitute a Work

At the judicial level, there are also inconsistent precedents regarding the protection of AI-generated content.

1. Feilin v. Baidu Case

The Beijing Internet Court held:

The graphics in the disputed article were automatically generated by software based on data; the differences in shape resulted from data variations, not from creation. Therefore, the graphics lacked originality and did not constitute works.

Second, regarding the automatically generated analysis report by the Wolters Kluwer database (Weike Xianxing Ku): Looking at the generation process, the report was automatically generated by inputting specific keywords and using the “visualization” function. Its content involved a judicial analysis of the film and entertainment industry, meeting the formal requirements for written works. The content reflected selection, judgment, and analysis of relevant data, and thus had a certain degree of originality. However, originality alone is not a sufficient condition for a written work. Under the Copyright Law, creation by a natural person remains a necessary condition for a work under copyright law, and this should not be overridden.

The analysis report was formed by the Wolters Kluwer database using input keywords combined with algorithms, rules, and templates. In a sense, it could be said that the database “created” the analysis report. However, because the report was not created by a natural person, even if it had originality, it was still not a work under copyright law. Therefore, even if the analysis report “created” by the database had originality, it was still not a work under copyright law, and the database could not be recognized as an author entitled to the rights under copyright law. From this, it is clear that the Beijing Internet Court considered creation by a natural person remains a necessary condition for a work under copyright law, and therefore AI-generated content does not have originality.

2. Tencent v. Yingxun Case

The Shenzhen Nanshan District Court held: First, the disputed article was generated by the plaintiff’s creative team using the Dreamwriter software. Its external form met the formal requirements for written works, and its content reflected the selection, analysis, and judgment of relevant stock market information and data from that morning. The article had a reasonable structure and clear logic, exhibiting a certain degree of originality. Second, the generation process of the disputed article mainly involved four stages: data service, triggering and collaboration, intelligent verification, and intelligent distribution. In these stages, the input of data types and processing of data formats, setting of trigger conditions, selection of article framework templates and corpus, and training of intelligent verification algorithm models were all arranged and selected by the relevant members of the creative team. Therefore, the form of expression of the article was determined by the individualized arrangement and selection of the plaintiff’s creative team. The form of expression was not unique, and it had a certain degree of originality.

The court held that, based on an analysis of the external form and generation process of the disputed article, the specific form of expression originated from the creator’s individualized selection and arrangement, and the creative process “generated” technically by the Dreamwriter software satisfied the protection conditions of copyright law for written works. Therefore, the court held that the disputed article constituted a written work protected under China’s Copyright Law. It is evident that the Shenzhen Nanshan District Court deviated from the view that “creation by a natural person is a necessary condition for a work under copyright law.”

3. Case Commentary

However, unlike academic controversy, both the Feilin v. Baidu case and the Tencent v. Yingxun case recognized that AI-generated content has a certain degree of originality. Nevertheless, from the above analysis, there is considerable divergence in the judicial reasoning and outcomes of the two cases.

The main divergences are: first, different standards for judging originality; second, different scopes of the creative process for AI-generated content. In Feilin v. Baidu, the originality standard adopted was “independent creation” plus a “reader-oriented standard of creativity”; whereas in Tencent v. Yingxun, it was “independent creation” plus an “author-oriented standard of creativity.”
In Feilin v. Baidu , the court considered that an expression only needs to satisfy independent creation and that, from the reader’s perspective, it reflects selection, judgment, and analysis of relevant data to possess originality, regardless of whether such selection and judgment are made by a machine or a natural person. However, curiously, while the court in Feilin v. Baidu adopted an objective reader perspective on originality without insisting on the source of creativity, it ultimately held that the expression did not constitute a work because it was not created by a natural person. This inconsistent stance reflects the court’s contradictory attitude toward AI-generated content: wanting to protect AI-generated content while unwilling to break out of the established framework of traditional copyright law. Recognizing that AI-generated content has a certain degree of creativity is the theoretical basis for its protection, while holding that it lacks the element of natural person creation and therefore does not constitute a work is an adherence to traditional copyright law.

In Tencent v. Yingxun , the court held that only expressions that satisfy independent creation by the author and reflect the individual personality, selection, and judgment of a natural person possess originality. This is a typical example of “instrumentalism.” The Tencent v. Yingxun court, while adhering to a subjective author-oriented originality standard, expanded the scope of the creative process for AI-generated content to encompass the overall process of designing and using the AI. This strengthened the element of human involvement in the expression, emphasizing the individualized selection and judgment of humans, and ultimately found that the AI-generated content constituted a work.

3 Ownership of AI-Generated Content

There are no specific provisions in copyright-related legal documents concerning the ownership of AI-generated content. Therefore, this issue is also debated in academia and practice.

(A) Academic Views on Ownership of AI-Generated Content

In academia, most scholars believe that AI-generated content has substantial economic value and can bring positive benefits to the copyright market, thus it should be protected. On this basis, academic debate focuses on what institutional design should be adopted for AI-generated content and to whom the relevant rights should be attributed to better promote the enrichment of the cultural content market.

There are three main views on the ownership of AI-generated content:

1. “Work-Copyright Theory”

This view holds that AI-generated content should be considered as works with originality and human personality elements. In cases of human-machine collaborative creation, “machine authors” and human authors both make substantive contributions to the work, and copyright attribution may be handled by analogy with works made for hire or by agreement of the creators.[11]

2. “Investment-Neighboring Rights Theory”

This view argues that AI-generated works arise from the “non-creative” investment of investors, and the interests of investors should be the focus of legal protection. The scope of broad neighboring rights includes content that expresses ideas but does not constitute a work. Therefore, for AI-generated works, a new type of neighboring right should be designed for protection.

On this basis, Professor Tao Qian argues that, absent an agreement between the parties, the neighboring right should be attributed to the AI user. This is because granting copyright protection for the program itself to the AI designer (owner) is already sufficient to provide incentive, and there is no need to allow them to claim rights over the content generated by the program. The AI designer (owner) merely makes it possible for the program to generate content but does not objectively bring the result into existence. Attributing the neighboring right to the AI user is more conducive to the dissemination of the results and to achieving the legislative goals of copyright law.[12]

3. “Fruits-Ownership Theory”

This view, from the perspective of civil law theory, regards AI-generated content as a new thing arising from the AI as a thing. The ownership of this intelligent creation may combine the Roman law principle of “origin-based ownership” and the Germanic law principle of “production-based ownership,” recognizing the AI developer, user, or owner as the copyright holder.

(B) Practical Views on Ownership of AI-Generated Content

1. Feilin v. Baidu Case

The Beijing Internet Court held:

The creator is neither the software developer nor the software user, but AI-generated content needs protection and can be protected as a type of interest, attributed to the software user.

Regarding the software developer: The software developer (owner) did not input keywords to search according to its own needs, and the analysis report did not convey an original expression of the software developer’s (owner’s) thoughts or feelings. Therefore, the analysis report should not be deemed as created by the software developer (owner).

Regarding the software user: The software user merely submitted keywords for searching, and the analysis report automatically generated by applying the “visualization” function also did not convey an original expression of the software user’s thoughts or feelings. Therefore, the analysis report should not be deemed as created by the software user.

Thus, the Beijing Internet Court held that concerning the attribution of authorship for AI-generated content, neither the software developer (owner) nor the user, as non-creators, can be named as authors. Instead, in the interest of protecting the public’s right to know, maintaining social honesty and credibility, and facilitating cultural dissemination, the analysis report should include an identifier of the generating software, indicating that it was automatically generated by the software.

At the same time, the Beijing Internet Court held that while AI-generated content does not constitute a work under copyright law, this does not mean it enters the public domain and can be used freely. The court considered that granting the software user certain lawful interests would facilitate further dissemination of the invested results (i.e., the analysis report).

For the software developer (owner), its interests can be obtained through software usage fees, etc., and its development investment has already been compensated. If the relevant interests in the analysis report were granted to the software developer (owner), the developer might not actively apply the results, which would be detrimental to cultural dissemination and the development of science.

However, the court did not clearly define what these relevant interests are, nor are there statutory provisions, leaving this matter for further discussion.

2. Tencent v. Yingxun Case

The Shenzhen Nanshan District Court held: Ownership follows “agreement.”

In its judgment, the court wrote: “Given the actual circumstances of this case and the fact that the software copyright owner has agreed with the plaintiff that the copyright in works created using the licensed software belongs to the plaintiff, there is no need to examine the matter further, and it is disregarded.” Therefore, it is evident that the court considers that the ownership of copyright in works created using software can be clarified by agreement. However, the court did not address how ownership of AI-generated content should be determined in the absence of an agreement.

4 Protection Pathways for AI-Generated Content

In the Feilin v. Baidu case, the Beijing Internet Court held that while the generated report does not constitute a work, it should be protected as a type of interest belonging to the AI user. The court did not clearly define what these relevant interests are, nor are there statutory provisions, leaving the matter for further discussion. In the Tencent v. Yingxun case, the court held that AI-generated content constitutes a work, and therefore the protection pathway should be the same as for ordinary works.

In academia, due to differing views on the ownership of AI-generated content, the protection pathways also vary. The main views are as follows:

1. The “Work-Copyright Theory” School

Among them, Professor Wu Handong argues that a dual subject structure of “author-copyright holder” can be constructed, distinguishing between the qualification of the creative subject and the right holder. That is, the author is not necessarily the first copyright holder. Based on the reality of “algorithmic creation,” future law could create a fictional “machine author,” while the copyright-related rights would be exercised by the right holder.[11]

Professor Yang Lihua argues that while protection of AI-generated content should be strengthened, it should also be limited to maintain the balance of interests under copyright law. She proposes amending legislation to clearly define the scope of the object of rights for AI-generated content; shortening the term of protection for AI-generated works; requiring mandatory attribution for AI-generated works to protect the market for natural person creations; and establishing a statutory licensing model to overcome the dilemma of machine reading in fair use.[14]

Scholar Sun Wen argues that AI-generated content should be attributed to the AI designer (owner), but such an ownership structure may hinder the use of AI by users and control over third-party use. In this case, the copyright in the generated content should be transferred from the designer to the user by means of legislative transfer of rights, along with certain adjustments to copyright ownership rules.[15]

2. The “Investment-Neighboring Rights Theory” School

Among them, Professor Tao Qian argues that a new neighboring right—the right of data processors—should be established, attributed to AI users, to protect AI-generated content that does not constitute a work. Second, this neighboring right should be limited; for example, the term of protection for data products should be shorter than that for copyright economic rights and general neighboring rights. Finally, regarding the exercise of rights, because the AI-generated result itself originates from the program’s deep learning of published works and information, other entities should not be restricted from using technology to extract and utilize AI-generated results again.[16]

3. The “Fruits-Ownership Theory” School

This school argues that protection and transfer should follow the property law theory in civil law.

5 Tort Liability for AI-Generated Content?

Although AI research has become an academic hotspot, there are relatively few academic research results on the allocation of tort liability for AI-generated content. Currently, the main views are as follows:

(A) Employer Liability Theory

Employer liability, also known as “vicarious liability,” refers to the situation where an employer is liable for acts of an employee within the scope of employment that infringe on the rights of others. The employer liability theory posits that the autonomous operation of AI to perform specific tasks means that the owner or user is the beneficiary of its operational results. That is, if an intelligent robot actually “acts as agent or representative” of a legal subject in conducting activities or making decisions, the person deploying the intelligent robot can be held vicariously liable, analogous to an employer’s liability for its employees. For example, Bezár et al. point out that the basic principle should be that whoever “employs” is responsible. Under employer liability, the liable party bears no-fault liability.[17] Additionally, there is a special form of employer liability—labor dispatch liability. The European Parliament’s “Civil Law Rules on Robotics” (February 2017) is based on such a labor dispatch relationship. The rationale is that the user unit directly benefits from the results of the worker’s labor, and thus should bear stricter tort liability than the employer.

(B) Agency-Principal Relationship Theory

Regarding liability for infringement by AI-generated content, one theory treats it as an agency-principal relationship, viewing the AI program as an agent completing tasks for the principal. The principal controls the AI’s production methods and content. If the AI produces work the principal dislikes, the principal can modify the AI’s existing instructions to reflect the principal’s wishes. Ultimately, the principal is the one who benefits from the work created by the AI. Therefore, when the agent’s work infringes copyright, the principal should be liable to the injured party for that infringement.

(C) Legal Subject Liability Theory

1. Analogy to Natural Persons with Limited Personality – Guardian Liability

The guardian theory holds that AI possesses a certain degree of autonomy but cannot think and act fully independently like humans. It is analogous to a person without or with limited capacity for civil conduct in human society. If a user employs such AI, the user has a duty to supervise and manage the AI’s autonomous behavior and should bear guardian liability for the AI. Lai Yuqiang proposes that AI is created by humans to serve them; its status cannot exceed that of humans, nor even reach the same level of importance. At most, it can have limited capacity for civil conduct and limited civil capacity. This capacity can be analogized to that of a minor—a “human-like” entity. Therefore, the owner should bear the responsibility of a guardian for the AI.[18]

2. Analogy to Legal Persons with Limited Personality – Entity in Formation and Electronic Person

Scholars Liu Xiaochun and Da Yachong propose that intelligent robots should have limited legal rights, i.e., they can become legal entities in formation.[19] Becker argues for establishing a new subject between humans and tools, binding all participants involved in the intelligent robot (developers, programmers, producers, users) into a new legal entity—an “electronic agent” or “electronic person.” He believes the concept of legal personality has been quite successful in dealing with corporate matters, at least ensuring that losses caused by the cooperative actions of all parties to the corporation are not borne solely by one individual. Therefore, intelligent robots could be granted a specific legal “personality,” giving autonomous robots a certain degree of legal autonomy. This view is consistent with the theory of subordinate legal subject status.

3. Directly Granting AI Legal Subject Status

Many scholars, starting from legal subject theory, argue that there are no theoretical or ethical difficulties in directly granting AI legal subject qualification. Scholar Zhou Xiang believes that there are many precedents for “social personification,” sufficiently demonstrating that there is no fundamental obstacle in legal reasoning or methodology to “personifying” intelligent robots today. The content of rights subjects is gradually constructed and is a constantly evolving concept. The essential element of intelligence in intelligent robots necessarily requires granting them legal subject status, thereby having a positive effect on social development. Thus, granting intelligent robots legal subject status is in line with the requirements of the times. For example, Xu Zhongyuan proposes recognizing the civil subject personality of legal persons in civil law.[20] Scholar Zhang Tong argues that the distinction between legal subjects and objects should not be limited to whether they have life or independent will. Instead, the focus should be on whether it is necessary to grant them subject status so they can enjoy capacity for rights. Just as legal persons have artificial personality, granting artificial personality to AI products is not impossible. Where the best protection of the interests of other legal subjects (natural persons and man-made legal entities) is not restricted, AI systems should be recognized as legal subjects. Therefore, in the field of intellectual property law, some scholars, based on AI’s legal subject status, propose granting copyright to AI creations.[21]

From the above overview, it is clear that there is still a diversity of opinions on the legal protection of AI-generated content. However, regardless of the legal controversies, AI has irresistibly entered our work and life. If you don’t believe it, just look:

---

References

[1] See Beijing Feilin Law Firm v. Beijing Baidu Netcom Science & Technology Co., Ltd. (Dispute over Infringement of Authorship Rights, Right to Integrity of Works, and Right of Information Network Dissemination), Beijing Internet Court Civil Judgment No. 239 of 2018 (Initial).

[2] See Shenzhen Tencent Computer System Co., Ltd. v. Shanghai Yingxun Technology Co., Ltd. (Dispute over Copyright Infringement and Unfair Competition), Shenzhen Nanshan District People’s Court Civil Judgment No. 14010 of 2019 (Initial).

[3] See Wang Qian, “On the Characterization of AI-Generated Content under Copyright Law,” Legal Science (Journal of Northwest University of Political Science and Law), Issue 5, 2017.

[4] See Qiu Rungen & Cao Yuqing, “On Copyright Protection of AI ‘Created’ Works,” Journal of Nanchang University (Humanities and Social Sciences Edition), Issue 2, 2019.

[5] See Liu Yinliang, “The Status of AI-Generated Works under Copyright Law,” Politics and Law, Issue 3, 2020.

[6] See Tao Qian, “Protection of AI-Generated Results under Copyright Law – Establishing a Data Processor’s Right as a Neighboring Right,” Jurisprudence, Issue 4, 2018.

[7] See Li Yang & Li Xiaoyu, “Exploring Copyright Issues of AI-Generated Content from the Perspective of Kant’s Philosophy,” Jurisprudence Journal, Issue 9, 2018.

[8] See Xiong Qi, “Copyright Determination of AI-Generated Content,” Intellectual Property, Issue 3, 2017.

[9] See Wu Handong, “Institutional Arrangements and Legal Regulation in the Age of Artificial Intelligence,” Legal Science (Journal of Northwest University of Political Science and Law), Issue 5, 2017.

[10] See Huang Yuye & Sima Hang, “Ownership of Rights in AI-Generated Works from the Perspective of Fruits,” Journal of Henan Normal University (Philosophy and Social Sciences Edition), Issue 4, 2018.

[11] See Wu Handong, “The Copyright Law Question of AI-Generated Works,” Peking University Law Journal, Issue 3, 2020.

[12] See supra note [6].

[13] See supra note [11].

[15] See Yang Lihua, “Exploring Copyright Issues of AI-Generated Content,” Modern Law Science, Issue 4, 2021.

[15] See Sun Wen, “Research on Copyright Protection Pathways for AI-Generated Content,” https://mp.weixin.qq.com/s/WDfJml3sUxjlmOHrFAPDMw , last accessed February 14, 2023.

[16] See supra note [6].

[17] See supra note [9].

[18] See Lai Yuqiang & Yuan Xiuyang, “Exploring the Dilemma and Solutions for Liability of AI Infringements,” Journal of Chongqing Radio and Television University, Issue 1, 2019.

[19] See Liu Xiaochun & Da Yachong, “Examination of the Legal Personality of Intelligent Robots,” Frontiers, Issue 3, 2018.

[20] See Cheng Sumei & Gao Shiyu, “Should Intelligent Robots Have Legal Subject Status?” Journal of Xi’an Jiaotong University (Social Sciences Edition), Issue 1, 2020.

[21] See Zhang Shangen, “Theoretical Basis and Technical Discrimination of the Subordinate Legal Subject Theory for Artificial Intelligence,” Seeking, Issue 6, 2021.