For legal consultation about Intellectual Property・IT・Artificial Intelligence・Venture Businesses—STORIA Law Office

(2) Unique features of Article 30-4 of Japan’s Copyright Act

As explained above, Article 30-4 of Japan’s Copyright Act, which allows a broad scope for the use of copyrighted works if it is for the purpose of “data analysis”, is said to be quite unique from a global perspective.

Although other countries also have provisions allowing the free use of copyrighted works if it is for “data analysis”, in Great Britain, for example, only the “reproduction for the analysis of texts and data for non-commercial research” is subject to the limitations of rights [United Kingdom], Copyright, Designs and Patents Act 1988, Article 29A, as revised in 2014). ¹In addition to the example of the UK, for the situations in the United States, India, and China, please refer to Appendix 2 (“Study Group Report on New IP Systems Issues”) from page 55 of “(Survey on Issues Related to New IP Systems
such as Design and Brands Protection Overseas)”, (PDF)

On the other hand, since Japan’s Copyright Act does not have such a limitation on purpose, Article 30-4 of the Copyright Act applies to cases where a commercial enterprise engages in data analysis for commercial purposes.

This point is the quintessential feature of Article 30-4 of Japan’s Copyright Act.

I would like to write about an issue related to this about which I am often consulted.

(3) To what kinds of acts does the Japanese Copyright Act apply?

When we talk about the merits of Japan’s Copyright Act, the discussion inevitably ends up raising the issue of how Japan’s Copyright Act applies to a certain act or conduct. This is really an issue of which governing law applies to a certain act. For matters concerning copyright laws, in general, the law of the “place where the copyrighted work in question is used” (the place or country of the act of use) should apply.

Therefore, when you engage in the act of use of copyrighted works to develop AI software, if the place of the act of use involved for such use occurs in Japan, then the Japanese Copyright Act will apply. Conversely, if the same actions take place in the U.S., then the Copyright Law of the U.S. will apply.

So, to what location does the “place of the act of use” refer to when the act of use crosses national borders, such as when a creator located in Japan performs an act of use involving a copyrighted work (such as downloading data, machine learning, and the like), which is necessary for the development of AI software, on a server located in the United States?

If both “the place where the main user (the creator) is located” and “the location of the server where the act of use takes place” are in Japan, then Japan is the “place of the act of use”, and Japan’s Copyright Act will apply.

On the other hand, if either “the place where the main user (the creator)is located” or “the location of the server where the act of use (such as reproduction) takes place” is located in a foreign country (for example, when a creator located in Japan selects a foreign region to work), the conclusion as to whether Article 30-4 of Japan’s Copyright Act applies is unclear, leaving a risk that the Japanese Copyright Act will not apply.

(4) Even when the copyright holder of copyrighted work to be used or the user of the copyrighted work is a foreign company (or an individual who is a foreign national), is the act of use of such copyrighted work lawful under Article 30-4 of Japan’s Copyright Act?

The issue of governing law is only a question as to whether the “place ofthe act of use” took place in Japan or not. It is irrelevant whether the copyright holder of such copyrighted work is a Japanese company or a foreign company or whether the user of the copyrighted work is a Japanese company or a foreign company.

Even when the copyright holder of the copyrighted work is a foreign company(or a foreign national) or when the user of the copyrighted work is a foreign company, Article 30-4 of Japan’s Copyright Act will apply and be lawful as long as the place of the act of use of such copyrighted work is located in Japan.²Tatsuhiro Ueno, “Data Mining and Copyright: Japan as ‘Paradise for Machine Learning’” (Journal of the Japanese Society for Artificial Intelligence, Vol. 36, No. 6, Page 748)

(5)Is it lawful to provide AI software globally in SaaS format, which is generated lawfully in Japan and deployed on a server located abroad, or, to license such AI software to a company located overseas?

It is lawful.

In this respect, if “such AI software is a derivative work of the data (the copyrighted work) used for training”, such AI software, in principle, cannot be used without the consent of the copyright holder of such data(the copyrighted work) (Copyright Act, Article 28).

However, inside AI software that is generated based on certain data, such data is usually parameterized in the training process, and the data itself is not retained in a form capable of being recognized. As such, such AI software is not a derivate work of such right from the start.

To put it simply, “AI software generated based on certain data” is a copyrighted work (copyrighted program work) completely separate from such data. Since such data is not used when disclosing or providing such AI software, it becomes unnecessary to deal with the rights related to such data.

Accordingly, the acts of providing AI software globally in SaaS format, which is generated lawfully in Japan and deployed on a server located abroad, and, licensing such AI software to a company located overseas are lawful.

(6) Limitations of Article 30-4 of Japan’s Copyright Act

Understandably, Article 30-4 of Japan’s Copyright Act also does not permit the unqualified exploitation of another person’s copyrighted work for the purpose of “data analysis”. A copyrighted work may not be exploited “if it would unreasonably prejudice the interests of the copyright owner in light of the nature or the purpose of the work or the circumstances of its exploitation” (Japan’s Copyright Act, Article 30-4, proviso).

The problem is which cases fall under the category of cases that “would unreasonably prejudice the interests of the copyright owner in light of the nature or the purpose of the work or the circumstances of its exploitation”.

More specifically, the issue under discussion is “whether the act of generating AI software capable of expressing the style of a particular author by learning only the image of a specific author” falls under the category of cases that “unreasonably prejudice the interests of the copyright owner…of the work”.

In this regard, Yasuyuki Echi states in AI Products, Machine Learning and the Copyright Act(Patent,Vol.73, No. 8, p. 142) referring to the meaning of the phrase “unreasonably prejudice the interests of the copyright owner” in the proviso in Article 30-4, that, as with Article 35-1 which uses the same phrase, “the meaningis determined by whether exploitation would interfere with potential future sales channels for the copyrighted work”. Based on this, he then concluded that the act of importing all Disney films (“the work” referred to in the proviso) into a computer and having the computer use machine learning to develop AI in order to create a new film in the Disney film style could plausibly be importing copyrighted works (the existing Disney films) as data for the purpose of creating a film that could possibly compete with “the work” (the Disney films) in the future. Therefore, it seems that the potential future sales channels for the copyrighted works might be impeded, thereby “unreasonably prejudicing the interests of the copyright owner”.

However, I personally have doubts about this view.

Of course, there is no doubt that the acts of generating and using expressions that are identical or similar to those already existing in the Disney films by using a trained model learned with such a method Echi described constitute copyright infringement. (I will discuss this point inIssue 3).

However, the question is what kind of actions are permitted to generate AI software, that does little else but possibly cause copyright infringement, at a stage when such infringement has not yet occurred.

If the applicability of the proviso in Article 30-4 is determined by the standard of whether (or not) it would “interfere with potential future sales channels for the copyrighted work”, the proviso would be applicable simply because “there would be even a slight possibility that the AI software would generate copyrighted works that are identical or similar to existing copyrighted works.”

However, since no one can prove that “there is no possibility that AI software will generate a copyrighted work that is identical or similar to an already existing copyrighted work”, if this interpretation is adopted,every act of generating AI software would constitute a copyright infringement. This would greatly undermine the purpose of Article 30-4 (see the Report of the Council for Cultural Affairs Copyright Committee (April 2017), p. 41), which is a provision allowing the exploitation of copyrighted works in the case of “non-enjoyment use” where there is no original act of use (such as viewing) of existing works.

In addition, with reference to the example of the Disney films, I think it is unlikely that just because some people use the Disney films for data analysis, there would be a negative impact on the “potential sales channels that use Disney movies for viewing”.

Therefore, in my opinion, the act of generating AI software capable of expressing a particular author’s style by learning only the images of a specific author does not fall under the category of “unreasonably prejudicing the interests of the copyright owner” under the proviso of Article 30-4.

However, it must be noted that what is lawful here is “the act of generating AI software” (Issue 1). The issue of what liability the AI software provider would have if, as a result of providing such AI software, a user who uses such AI software generates a copyrighted work that is the same or similar to an existing copyrighted work, thereby causing a copyright infringement (Issue 3(1)).

In other words, the question of Issue 1 is “whether the act of generating and providing a tool with a high risk of infringement itself is unlawful, regardless of whether or not an act of copyright infringement by the user has actually occurred”. The question of Issue 3(1) is “whether, in the case where an act of copyright infringement by the user has actually occurred, the act of providing tools used in such copyright infringement is unlawful”.