Should Generative AI Be Considered a Legal Person?

Host’s Note

Viewpoint 1

Artificial Intelligence Can Also Be Considered a Legal Person

A legal person is the ultimate holder of statutory rights and obligations, and the final bearer of legal benefits and risks. The legal person system in a country relates to the behavioral incentives and constraints of its social subjects, which in turn concerns the country’s economic development, social stability, and even the long-term prosperity of human society. In contemporary China, where artificial intelligence is widely involved in various fields of human social life, rationally and scientifically answering whether artificial intelligence is a legal person and reflecting on and reconstructing our legal person system is not only of great theoretical significance but also of extremely high practical value.

From an etymological perspective, the term “person” in the English-speaking world, which signifies a legal person, originates from the Latin word “persona”, meaning the mask worn by actors on a theatrical stage, extending to the roles played by the actors. Correspondingly, a legal person is a general term for various roles active in the legal world, used to distinguish them from the diverse individuals active in the real world. The kinship between the legal person and the mask on a theatrical stage provides us with a fascinating perspective for examining the legal person. Specifically, we can draw the following three significant conclusions: (1) Whether a specific entity can become a legal person is determined by legal standards based on social needs, just as actors are set by drama according to its plot; (2) Legal persons are separate from biological persons, just as the roles played by actors in a drama differ from their real-life personas; (3) Legal persons are established to protect the entities behind them, just as masks are worn to protect the actors behind them (consider the case of the bully Chen Qiang, who was nearly beaten by the audience for playing the role of Huang Shiren).

From a historical legal perspective, the evolution of the legal person has, to some extent, confirmed the validity of the above three conclusions: (1) Although the legal persons recognized by various countries’ laws tend toward a “great unity”, there are still “small differences”; these “small differences” are the result of rational choices made by various countries’ laws based on their social needs; (2) The legal person has undergone a tremendous transformation from “human as non-human” to “everyone as human” and then to “non-human as non-human” throughout history, which is strong evidence of the separation between legal persons and biological persons; (3) The acquisition of corporate legal personality allows shareholders to be shielded from direct claims by corporate creditors, which vividly demonstrates that legal persons have protective functions for the entities behind them against legal sanctions.

Therefore, whether artificial intelligence is a legal person should be comprehensively assessed based on the institutional functions of legal persons, rather than being outright denied solely because artificial intelligence lacks certain essential characteristics of biological persons. After all, legal persons and biological persons are separate; determining whether artificial intelligence is a legal person based on whether it possesses certain biological characteristics is futile and misguided. If artificial intelligence’s legal person status is denied solely on this basis, then fair distribution of benefits generated by artificial intelligence and sharing of risks associated with artificial intelligence will become mere rhetoric, not to mention incentivizing stakeholders in China’s artificial intelligence industry chain to strive to make our country a “major global center for artificial intelligence innovation”!

Viewpoint 2

Analysis of Artificial Intelligence’s Legal Status in Intellectual Property

With the development of artificial intelligence technology, particularly the widespread application of generative AI systems like ChatGPT and Sora, the role of artificial intelligence systems in assisting or even replacing human intellectual creation has become increasingly significant. Can artificial intelligence systems be recognized as inventors for AI-generated inventions? Can they be regarded as authors of “text-to-image” or “text-to-video” works? These questions are now facing intellectual property authorities and courts worldwide. Regarding the legal status of artificial intelligence in intellectual property, at least the following thoughts emerge based on the current intellectual property systems and relevant practices in various countries:

First, AI-generated works, including inventions, written materials, images, and videos created using artificial intelligence systems, while they may be indistinguishable from purely human intellectual creations (i.e., creations made without AI), still face considerable controversy regarding whether they can be treated as objects of intellectual property rights. Different countries and courts may arrive at different conclusions based on varying case specifics. Generally, the greater the intellectual contribution from humans in AI-generated works, such as more complex prompts or judgments inputted into the AI system, the more likely those creations will be considered patentable or copyrightable. Conversely, if the human contribution is minimal, such as merely inputting a simple prompt leading to an AI-generated text, image, video, or technical solution, such a “creation” may not be recognized as patentable or copyrightable.

Second, given the fundamental objective of intellectual property systems to encourage the creation and dissemination of human intellectual achievements, most countries’ laws and practices deny the possibility of recognizing artificial intelligence systems as inventors or authors of human-created works, further denying them the status of patent or copyright holders. For example, the latest Patent Examination Guidelines in our country explicitly state that inventors must be individuals and cannot be artificial intelligence systems. The U.S. Copyright Office’s guidelines on copyright registration for works containing AI-generated content also clearly stipulate that only natural persons can be registered as authors of works, explicitly excluding non-humans from authorship under U.S. constitutional and copyright law. Therefore, artificial intelligence systems cannot be recognized as authors of works; if an applicant lists an AI system as the author, the U.S. Copyright Office will directly refuse registration. In cases where AI-generated works are deemed copyrightable or patentable, the individual who inputs the prompts or makes selections and judgments on the AI’s output is typically recognized as the inventor or author. This practice may lead to controversies. For instance, in certain cases where prompts are very simple, the inputter may not have made a substantial creative contribution to the AI-generated “work” or “invention”, which is almost entirely the result of the AI system, making it contradictory to the principle of good faith in civil law to list the prompt inputter as the inventor or author. To address this issue, the U.S. Copyright Office’s guidelines further stipulate that if the expressive elements of an AI-generated output are indeed determined by the AI system, then that output does not qualify as a human work and thus cannot receive copyright protection.

Third, although artificial intelligence systems cannot enjoy related rights as authors or inventors of their “creations”, when their “creations” involve infringement, the owners of those AI systems may bear liability for infringement. Recently, the Guangzhou Internet Court handled a copyright infringement case involving an AI company’s text-to-image generation. In this case, the plaintiff was the exclusive licensee of the Ultraman series, and the defendant was an AI system company. The plaintiff alleged that the images generated by the defendant’s AI system were substantially similar to the Ultraman images for which the plaintiff held exclusive licensing rights, constituting copyright infringement. The court ruled in favor of the plaintiff, finding that the defendant infringed on the plaintiff’s reproduction and adaptation rights regarding the Ultraman work. Prior to this case, the Beijing Internet Court also ruled on a copyright dispute involving an AI system’s text-to-image generation, concluding that inputting prompts into the AI system and generating images constituted artistic works, and that the prompt inputter or user of the AI system could be regarded as the author of the AI-generated images, enjoying copyright. Although the Guangzhou and Beijing Internet Courts may have differing approaches to the legal subject recognition of AI-generated images, the potential for AI “creations” to infringe on others’ intellectual property rights is a reality. The question of who should bear responsibility for infringement—whether it’s the owner of the AI system or the user of the AI system—and under what circumstances the owner of the AI system should be held accountable or can receive protection under the “safe harbor” principle are pressing issues currently faced in practice and require in-depth theoretical research.

Viewpoint 3

Artificial Intelligence Challenges the Relationship Between Persons and Objects in Law

Is artificial intelligence a legal person? The answer may be very obvious; the vast majority of scholars and experts would likely give a negative response. ChatGPT and Sora may create content that is similar to or even surpasses that of humans, and self-driving technology may operate more safely and efficiently than humans, but fundamentally, artificial intelligence remains a tool. Artificial intelligence does not possess personal will like a human, nor does it have the collective will of a legal person; viewing artificial intelligence as a legal person fails to provide reasonable incentives through the legal system’s rights and obligations.

However, artificial intelligence challenges the traditional legal definitions of the relationship between persons and objects. Traditional law presumes that humans are creative subjects with free will, while objects are entities to be controlled by humans. Based on this presumption, law establishes different legal systems around persons and objects. For example, contracts and torts center around human beings, with typical legal designs including elements like “autonomy” and “fault”, which are characteristic of humans; property rights and ownership focus on objects, and product liability also centers on objects, often neglecting human subjective states or characteristics, displaying a “in rem” nature that can be directed at unspecified individuals. Intellectual property can be viewed as a system linking persons and objects, emphasizing the personal characteristics of intellectual property while also treating works and patents as objects subject to the principle of exhaustion of rights.

The changes brought about by artificial intelligence are that AI is becoming increasingly like a person, while humans, in many cases, are becoming more like machines. For instance, in smart investment and quantitative trading, AI algorithms replace humans in expressing intentions; in self-driving, human drivers transform into passengers or safety supervisors with less oversight obligations. The most typical example is in copyright within intellectual property. Many AI-generated works appear closer to human works, possessing high value and historical creativity, while some human works may resemble valueless machine outputs, at most exhibiting psychological creativity. If we adopt a purely human-centered or romantic author perspective, the conclusion would be that the law can protect low-value human works but cannot protect high-value machine works.

In light of the blurring boundaries and changing relationships between persons and objects, traditional legal systems need to evolve and upgrade. For instance, regarding legal liability for artificial intelligence, the law can categorize it based on specific characteristics. For AI systems that constitute part of a product, such as self-driving systems, the law can regulate them as product components and impose product liability. For independent AI service systems, such as generative AI, the law can treat them as services provided by humans or human organizations, regulating the behavior of different human subjects involved with AI. For AI-generated works, the law can avoid a binary application of copyright and offer more granular protection. For example, the law can grant initial authorship rights to AI companies, providing reputational incentives while offering database-like protection and safeguards against unfair competition for AI works.

In traditional legal studies, the relationship between persons and objects is also a core hidden issue. Discussions regarding the division of property debts and modes of property rights transfer in private law, as well as debates over whether to regulate persons or objects in public law, all involve discussions of the relationship between persons and objects. In recent years, there has also been a rise in the study of modular theory in foreign legal academia, aiming to deepen the understanding of departmental law and legal theory. In this sense, whether artificial intelligence is a person or an object is not merely a trending hot topic but a genuinely profound legal proposition.

Viewpoint 4

Artificial Intelligence Does Not Have Civil Legal Personality

Does artificial intelligence possess civil legal personality? Should legislation grant it such a status in the future? This has been a major point of contention in recent years within the theoretical community. The main purpose of establishing legal personality for artificial intelligence is to address the issue of liability in cases involving artificial intelligence, which can be achieved through the interpretation or analogical application of existing laws within the scope of the nature of artificial intelligence as an object. Creating legal personality for artificial intelligence would not only disrupt the existing civil law system but also lead to difficulties for victims in claiming damages and result in unreasonable risk distribution. I hold a negative view, with specific reasons as follows.

First, current law does not recognize the legal personality of artificial intelligence.

Some scholars view artificial intelligence as the agent of its owner or user, while others see it as a substitute liable party in cases of tort; some suggest treating artificial intelligence as a quasi-person, similar to a legal entity; and more radical foreign legislative drafts directly grant it a new type of electronic personality. However, this is not the legal reality in our country. There are no provisions in China’s Civil Code that establish the civil legal personality of artificial intelligence, nor have there been any rulings based on the “agency theory”, “substitution theory”, or “legal entity theory” in practice.

Second, the “object interpretation theory” is sufficient to resolve existing tort issues involving artificial intelligence.

Based on the types of artificial intelligence models, they can be divided into discriminative AI products and generative AI services. The former predicts or classifies different outcomes based on input data through an “input-output” relationship; the latter constructs models by learning the distribution and generation rules of input data and generates new data samples accordingly. Both are fundamentally tools in the sense of being objects.

Regarding the liability for torts involving discriminative AI products, product liability rules can be applied. From the perspective of protecting victims, strict liability does not consider fault elements, excluding the obligation for victims to prove the perpetrator’s fault. At the same time, product liability reduces the burden of proof for causation on victims. From the perspective of market position, manufacturers of AI products (the perpetrators) are in a favorable position to bear the economic burden of product liability through market pricing and other means. From the perspective of industry development, strict liability can impose higher quality requirements on manufacturers, incentivizing their investment and the advancement of the industry.

Regarding the liability for torts involving generative AI products, the liability recognition rules for network service providers can be analogously applied. Generative AI service providers are obliged to handle personal information legally and have a duty to protect the security of personal information; otherwise, they should be deemed at fault. Service providers have a heightened duty of care to screen for illegal or sensitive content involving public interest, but they cannot be required to exercise general duty of care over all generated content; additionally, the safe harbor rule can be analogously applied to generative AI service providers.

Third, the “subject legislation theory” does not meet the necessary requirements for legal subjects.

As civil legal subjects capable of bearing liability, the two most important requirements are independent property and the ability to express intent. On one hand, even if the establishment of legal personality for artificial intelligence is supported, when it comes to damage compensation, its property sources would still be the traditional civil subjects involved in the design, manufacture, sale, and use of artificial intelligence. On the other hand, current artificial intelligence remains in the “weak AI” era, lacking independent consciousness, meaning there is no ability for intention expression. Therefore, the legislative theory of establishing a subject does not conform to the inherent meaning of legal subjects.

In summary, artificial intelligence currently does not possess civil legal personality; its status cannot be fabricated through interpretative means nor established through legislative means.

Viewpoint 5

There Is Room for Artificial Intelligence to Be Included as a Legal Person

With the development of artificial intelligence, the question of whether it can be considered a legal person has become a hot topic in the legal community. The core of the discussion revolves around examining whether artificial intelligence can possess human-like intent and other factors to determine if it can be classified as a legal person. However, this perspective overlooks that “legal person” has never been a fixed concept.

First, “legal person” is a constructed concept. American legal scholar Bodenheimer once said, “Justice has a Protean face, ever-changing, capable of presenting different shapes and forms.” This statement applies equally to “legal person.” A “legal person” is a construct defined by legal values rather than a biological assessment. Biological persons and legal persons are different concepts. A wolf child is a biological person but may not be a legal person. Similarly, a host who has undergone gender reassignment surgery is biologically a man but legally a woman. The “legal person” concept must consider not only biological characteristics but also social characteristics. Following this logic, when artificial intelligence possesses social characteristics similar to those of humans, such as forming social relationships, it may indeed be evaluated as a “legal person”.

Second, the scope of “legal persons” varies across different eras. It is an undeniable fact that in slave societies, although slaves are biological persons, they are not legal persons. Historically, in many countries, the penalties for killing slaves and harming livestock were roughly equivalent. The recognition of Black people as legal persons in the U.S., granting them voting rights, is only a few decades old. Before the French Revolution, women were not fully recognized as legal persons. Even today, in many tribal societies (including parts of India), many types of people are effectively excluded from being recognized as legal persons due to factors like skin color, race, surname, belief, and gender. Clearly, throughout history, slaves, Black people, women, and specific ethnic groups have been excluded from the category of “legal persons.” As society evolves, the definition of “legal persons” has gradually expanded, increasingly accommodating various constructed entities. The responsibilities of legal persons in civil law and corporate crime in criminal law are results of this expansion.

Finally, while artificial intelligence is not a biological person, it can become a “legal person”. This, of course, depends on two conditions: first, the pace of artificial intelligence development; based on the current speed of generative AI development, it is estimated that in about ten years, artificial intelligence will possess capabilities akin to human intention and emotional capacity. Even if it does not reach the level of a scientist, it will surpass that of a high school student. Second, as artificial intelligence develops, legal concepts will gradually need to be adjusted; the concept of “legal person” was created due to the emergence of companies and other organizations. In the future, the scope of “legal persons” will continue to expand, potentially including certain artificial intelligences.

Given the current limitations of artificial intelligence development, we are still debating whether it can become a “legal person”. I believe that in 3-5 years, the question will shift to “when” artificial intelligence will become a “legal person”. Ten years from now, we may be discussing “in what way” artificial intelligence should become a “legal person”.

The law can only serve technology. In the struggle between systems and technology, technology is always the victor. Systems may strive to dominate, but it is technology that designs the systems.

Viewpoint 6

The Legal Status of Artificial Intelligence Should Be Based on Responses to Industrial Needs

Whether artificial intelligence can be regarded as a person in law involves three main aspects: first, whether artificial intelligence can be a subject of rights and enjoy statutory rights or interests; second, whether artificial intelligence can bear corresponding legal responsibilities as a subject of liability; and third, whether the operational processes of artificial intelligence can be viewed as actions of a legal “person” that produce corresponding consequences. Taking the currently hotly debated copyright law as an example, the corresponding questions for these three aspects are: first, can artificial intelligence become the subject of rights related to its generated works? Second, when artificial intelligence generates content that infringes upon prior copyrights, can it be held liable? Third, can the process of generating content by artificial intelligence be regarded as “creation” that forms a work?

According to current legal principles and regulations, there is indeed no space to recognize artificial intelligence as a legal person. However, to directly exclude the possibility of artificial intelligence obtaining legal personality based solely on its nature as a machine rather than a person is overly simplistic. In reality, various interests have always influenced the recognition and arrangement of “persons” or “subjects” within the legal system. The “legal person” is the most important constructed subject in the legal system, independent of the will and legal status of individual natural persons, playing a foundational role in social and economic life, and is a typical legal creation that responds to the needs of production modes and interest relationships. Furthermore, legal protections for the interests of the deceased, fetuses, and even animals are not merely mechanical considerations limited to “natural persons”. Specific arrangements, such as in copyright law, while legal persons cannot directly engage in natural intellectual creation, works can be viewed as created on behalf of the legal person, allowing the legal person to be regarded as the author; after an author’s death, their rights, such as authorship and publication rights, continue to be protected, which differs from the protection of the deceased’s reputation in civil law, as these rights do not fully transform into protections for the interests of close relatives, but rather exist as relatively independent interests, including the public’s right to know. Therefore, the recognition of “persons” in law is not simply based on natural attributes but often responds to social needs and specific interest arrangements.

Of course, if artificial intelligence were to be established as a legal person, it would require sufficient justification for its necessity. In the three levels mentioned above, the third level involves relatively specific and localized interest balancing; whether to view the process of creating content by artificial intelligence as “creation” to obtain copyright protection fundamentally requires considering whether it is necessary to provide production incentives for artificial intelligence-generated content through such a construct. Alternatively, one could opt not to break through the law to establish such a construct and instead employ special rights protection models (such as creating neighboring rights) or protections under unfair competition law, or simply not provide any protection. This situation involves value judgments within specific legal frameworks. The first and second levels—whether artificial intelligence can become subjects of rights and responsibilities—entail more foundational and global decisions, requiring answers to whether artificial intelligence, as a means of organization and participation in production, has generated a need for independent legal status. If a one-to-one correspondence can be established between artificial intelligence and its providing legal entities or natural persons, then legal rights and responsibilities can be entirely borne by the latter, negating the need to independently establish artificial intelligence. However, if a separation between the enterprise and the market arises in a Coasean sense, and the process of artificial intelligence’s organization and participation in production becomes overly complex, involving multiple participants, it may necessitate recognizing a separate “artificial intelligence” entity in law to reduce transaction costs and promote efficient production and distribution. At present, however, there is no evident necessity for this acknowledgment given the current state of artificial intelligence technology and industry development.