[Feature: AI and Intellectual Property Rights] Tsukasa Aso: AI-Generated Works and Intellectual Property Law

"Keio University is a private university in Japan, with its headquarters located in Shibuya-ku, Tokyo. Its predecessor was a school for English studies founded by Yukichi Fukuzawa in 1858, and it was renamed to its current name in 1867. Keio University is one of Japan's leading prestigious universities, boasting high educational standards in a wide range of fields including economics, law, medical sciences, science and technology, and humanities. Furthermore, as a comprehensive university, Keio University has numerous faculties, graduate schools, professional graduate schools, the School of Medicine, and the Faculty of Nursing and Medical Care."

This is the result of typing "Keio University" into the currently popular ChatGPT and asking for an explanation (results from early April 2023). Of course, there are inaccurate descriptions (for example, the headquarters is in Minato-ku, it was a school for Dutch studies rather than English studies in 1858, and it became Keio University in 1868), but the response feels as though it was written by a human.

Recent AI, including not only such linguistic AI but also image-generation AI, has become capable of creating things close to human creation. Consequently, eventually (or perhaps already), things generated by AI (generated works) will become completely indistinguishable from things created by humans. So, will such AI-generated works be subject to protection under intellectual property law? If they are not subject to protection, should we seek to protect them through legislation?

In this article, I would like to introduce the discussions taking place in Japan regarding the relationship between AI-generated works and intellectual property law *1.

Since I am not an AI expert, I will explain based on generally accepted content. Even when we say "AI," various technical levels are envisioned, and there seems to be a distinction between Artificial General Intelligence (also called strong AI) and Narrow AI (also called weak AI). Artificial General Intelligence can perform various types of thinking and examination and can respond to situations encountered for the first time—essentially an AI like Doraemon that can do anything. On the other hand, Narrow AI is an AI that excels only in thinking and examination regarding specific content, such as an AI whose sole purpose is to win at chess. Since Artificial General Intelligence has not been realized, Narrow AI is the premise.

Furthermore, under current circumstances, generation is not performed by AI without any instructions from a human (even in the previous ChatGPT example, I typed "Keio University" and clicked), so it is also a premise that a minimum level of human involvement, such as pressing a button, is necessary. Here, assuming a case of "just typing Keio University," I would like to consider the protection of AI-generated works created through simple instructions *2.

1. Copyrighted Works

Article 2, Paragraph 1, Item 1 of the Copyright Act defines a work as "a production in which thoughts or sentiments are expressed in a creative way and which falls within the literary, scientific, artistic or musical domain."

To be a copyrighted work, it must first be an expression of "thoughts or sentiments." It is considered that the person who expresses these thoughts or sentiments is a natural person—that is, an ordinary human. For example, a photograph taken by a monkey does not manifest human thoughts or sentiments, so it does not meet this requirement. Therefore, AI-generated works created through simple instructions, such as just pressing a button, are not expressions of thoughts or sentiments made by a human and do not satisfy the requirement for thoughts or sentiments.

Since they do not satisfy the thoughts or sentiments requirement, AI-generated works do not become copyrighted works, but I will mention the creativity requirement as there is ongoing discussion. Creativity has traditionally been said to be the manifestation of the author's personality or individuality. This means that human personality or individuality appears in the creative process. Therefore, according to this traditional concept of creativity, since AI is not a person, AI-generated works also fail to satisfy the creativity requirement.

On the other hand, recently, there is an influential view that creativity is the "range of choices in expression." This is a position that considers whether creativity exists based on the degree of options for expression conceivable when expressing certain content—that is, the breadth of the range of choices. Among these, there is a theory called the "range of choices in expression from a competition law perspective," which examines the room for choice in expression for someone other than the expresser who intends to make a similar expression. In that case, creativity is judged by whether a range of choices remains for others based on the resulting AI-generated work; usually, even with an AI-generated work, a range of choices remains for others, so it would satisfy the creativity requirement.

Furthermore, the final requirement of "falling within the literary, scientific, artistic or musical domain" is considered to be satisfied even by AI-generated works.

Looking at it this way, it is understood that regardless of other requirements, the thoughts or sentiments requirement cannot be met, and AI-generated works do not become copyrighted works.

2. Subject

The issue of protection for AI-generated works under the Copyright Act does not end there.

Article 2, Paragraph 1, Item 2 of the Copyright Act defines an author as "a person who creates a work." Unless AI is granted legal personality, which is a different matter, AI is not recognized as having personality at this point, so AI cannot be an author. If that is the case, the question of who becomes the author arises.

In this case, there are two people who seem related to the creation. One is the person who gives the instruction to create to the AI—that is, the AI user (the AI user who typed Keio University: the author). And the other is the creator of that AI system, such as the person who created the program or the trained model (the creator of ChatGPT).

Here, for a work to be a creation by a human, creative intent and creative contribution are required. Regarding the AI user, they only gave a simple instruction (typing Keio University) and lack creative intent or creative contribution. Consequently, an AI user who only gave a simple instruction does not become the author of the AI-generated work. And the creator of the AI system would basically reach the same conclusion. This is because, usually, the person who created the program or the trained model has no creative intent or creative contribution toward the AI-generated work.

3. The Problem of False Attribution

Consequently, AI-generated works created through simple instructions do not become copyrighted works as mentioned earlier, and neither an author nor a copyright holder exists.

What is pointed out there is a problem called the "problem of false attribution" (Sennsho). This is the concern that if AI-generated works are currently not protected under the Copyright Act, people might appear who falsely claim that AI-generated works, which cannot receive copyright protection, are human works. If such a situation is left unaddressed, there is a risk of harming the safety of transactions such as licensing—for example, if it is later discovered that a work was actually an AI-generated work that is not a copyrighted work.

4. Rights Infringement

As described above, it is currently difficult for AI-generated works created through simple instructions to receive protection as copyrighted works under the Copyright Act.

And the issue regarding AI-generated works is not only whether they are protected as copyrighted works. Whether it constitutes copyright infringement when an AI-generated work is created using another person's work is also an issue. A typical example is whether copyright infringement occurs when a single painting in the training data and a painting output by the AI are very similar.

For copyright infringement to be recognized, it is necessary to have relied on another person's work and to be using a work that is identical or similar to that person's work. "Reliance" means accessing another person's work and basing the new work on it, and "similarity" means maintaining the identity of the essential characteristics of the form of expression in the other person's work.

What is particularly problematic in relation to AI-generated works is reliance, and there are various discussions regarding in what cases reliance is recognized. For example, there is a position that recognizes reliance if the work used to train the model remains as data, or even if it does not remain, if there was access to said work. This is a position that recognizes reliance simply because the AI learned that picture. There is also a position that affirms reliance if the original work contributed to the formation of a set of parameters and the generated work was produced based on that set of parameters, and denies reliance if the original work did not contribute to the formation of the set of parameters.

Furthermore, the question of who the infringer is when an AI-generated work is created is also an issue. Between the AI user and the creator of the AI system, who becomes the person who infringed the copyright? If both are the same, there is no doubt that person is the infringer. Then, what about cases where the creator of the AI system (the creator of ChatGPT) and the AI user (the author) are different, as in the example at the beginning? In this case, the AI user would likely be the infringer. However, since negligence is required for damages, if they did not know that the AI had learned another person's work, it is possible that the AI user is found not to be negligent. Whether the creator of the AI system also becomes an infringer at this time can only be said to be on a case-by-case basis. Since the subject of reproduction is judged by considering various factors such as the object and method of reproduction, and the content and degree of involvement in the reproduction, whether the creator of the AI system performed the reproduction is judged after considering these factors (even if they are not found to be the infringer, they may bear liability for damages based on joint tort as an aider).

In any case, since there are no judicial precedents yet, it is unclear how this will be judged in the future.

1. Inventions

Article 2, Paragraph 1 of the Patent Act defines an invention as "the highly advanced creation of technical ideas utilizing the laws of nature." Similar to the interpretation of the definition of a copyrighted work, expressions like "ideas" and "creation" appear in the definition of an invention; if one takes the position that human involvement is necessary for an invention, AI-generated works would not be recognized as inventions. On the other hand, since an "idea" in the Patent Act is a technical idea and an objective existence, it can be considered different from the "thought" linked to human thought in the Copyright Act. Furthermore, if an invention is a technical solution for a certain purpose, there is an interpretation that an invention does not necessarily need to be a human creation. From this position, an AI-generated work is evaluated as an invention.

2. Subject

The main clause of Article 29, Paragraph 1 of the Patent Act stipulates that "A person who has made an invention that is industrially applicable... may obtain a patent for that invention," and this "person who has made an invention" is envisioned to be a human. Based on this interpretation, an AI, which is not a person, cannot be an inventor.

3. Document Description Requirements

Furthermore, as a description requirement for the application at the time of patent filing, the name and other details of the inventor are required (Article 36, Paragraph 1, Item 2 of the Patent Act), and here too, the inventor is envisioned to be a human.

Finally, regarding trademarks protected under the Trademark Act (word marks like "SONY" or figurative marks like a bitten apple) and indications of goods or services (trademarks, trade names, etc.) that become issues under the Unfair Competition Prevention Act, these do not presuppose human creation, so even AI-generated works created through simple instructions are subject to protection.

As described above, in laws that protect creation such as the Copyright Act and Patent Act, the protection of AI-generated works created through simple instructions is in a difficult situation. Therefore, legislative proposals for their protection are being made.

1. Copyright Act

Regarding content such as pictures and music, there is a position that aims for protection under the Copyright Act because there is no outward difference between human creations and AI-generated works. From such a position, a proposal is made to revise the thoughts or sentiments requirement in the definition of a work. And in this position, it is proposed that the author shall not exist, and regarding the copyright holder, it should be based on the provision in Article 29 of the Copyright Act that "The copyright to a cinematographic work... belongs to the maker of the cinematographic work," and the provision in Article 2, Paragraph 1, Item 10 of the Copyright Act defining a maker of a cinematographic work as "the person who takes the initiative and responsibility for the making of a cinematographic work." That is, it is a proposal that if a person has "initiative and responsibility" for the generation of an AI-generated work and publishes it under their own name, that person shall hold the copyright. Since there is no author, moral rights of the author do not arise, but it is argued that there is room for substitution of moral rights of the author through property rights, etc.

On the other hand, if the protective value of an AI-generated work lies in the investment required for its production, it is also pointed out that protection through neighboring rights rather than copyright is preferable.

Furthermore, as a response to the problem of false attribution where an AI-generated work is falsely claimed to be a human work, the application of criminal penalties is also being considered. Article 121 of the Copyright Act stipulates that "A person who distributes... a reproduction of a work on which the true name or a well-known pseudonym of a person who is not the author is indicated as the name of the author shall be punished by imprisonment for not more than one year or a fine of not more than 1,000,000 yen, or both." (The imprisonment part will become "imprisonment with work" once the law, which is currently not yet in force, is enforced). Since AI-generated works created through simple instructions are not copyrighted works, the wording "reproduction of a work" needs revision, and forms of provision and presentation other than distribution need to be added, but it is a proposal to deter false attribution through the application of criminal penalties.

2. Patent Act

Even if one interprets that an invention does not require human creation, since AI is not a person, the description of the inventor in the application would need revision. As a revision plan, there are proposals to make the inventor "the person involved in the creation by AI" or "the person who possessed the invention with the intent to own it."

3. Others

In addition, positions have been shown that if an AI-generated work acquires distinctiveness that allows it to be distinguished from other people's goods through use, that value should be recognized and protected by a system similar to the Trademark Act or Unfair Competition Prevention Act, and that legislation within the Unfair Competition Prevention Act focusing on publicity value (the power to attract customers) is also conceivable.

4. Arguments for Legislative Caution (Unnecessary)

On the other hand, since no particular necessity for the protection of AI-generated works created through simple instructions is currently observed, the argument for legislative caution (unnecessary), which holds that protection for such AI-generated works is currently unnecessary, is also influential. As for the position of the Japanese government, while it intends to conduct continuous review regarding the protection of AI-generated works, there is currently no momentum to create new legislation.

So, how should we think about this? If asked whether we should revise the law to protect AI-generated works created through simple instructions, I am currently skeptical.

Even if legislation for protection of AI-generated works created through simple instructions were to be considered, as the argument for legislative caution (unnecessary) states, providing new protection means that acts that were previously free will be prohibited, so the necessity of legislation becomes an issue. According to the incentive theory, which states that protection under intellectual property law is necessary because development costs cannot be recovered without protecting intellectual property, leading to a decrease in new creations, the premise is a situation where valuable information will not be created if it is not protected. However, at present, specific cases where protection under intellectual property law is necessary for AI-generated works have not been clarified (it might be a situation where it can be said that AI-generated works are being actively created even if they are not protected by intellectual property law).

Therefore, it is difficult to affirm the necessity of granting some kind of new protection for AI-generated works at this point. It would be appropriate to continue examining whether such protection is necessary while monitoring the future situation.