January 31, 2024
Written by Khadija Shamisa, Dual 2L, LTEC Lab R.A.
LTEC Lab began this year with one of the most relevant issues of the time: the impact of artificial intelligence on the law. Faculty, students, librarians, and members from the Windsor community joined us in Windsor Law’s new building to learn from Professor Peter Yu’s valuable perspective on AI in copyright law.
Professor Chapdelaine, Associate Professor at Windsor Law and Director of LTEC Lab, began the event with a land acknowledgement. As LTEC Lab’s first seminar of the year, she notes that we are “starting off with a blast” by hosting a prominent and world-renowned intellectual property scholar on an issue that is top of mind to law and policy makers.
Thank you to Professor Yu, from LTEC Lab members!
From left to right: Khadija Shamisa, LTEC Lab R.A.; Brandon Beck, LTEC Lab R.A.; Professor Yu; Professor Chapdelaine, Director of LTEC Lab; and Mita Williams, Acting Law Librarian at the Faculty of Law.
Professor Peter Yu is a Regents Professor of Law and Communication and Director of the Center for Law and Intellectual Property at Texas A&M University. Professor Yu takes a transnational approach to his research, a theme that resonates strongly with Windsor Law and LTEC Lab’s approach to research and learning. He is also an esteemed colleague in the international IP community, well-known for organizing IP workshops, and mentoring colleagues and students.
Professor Yu begins his presentation stating that his students and colleagues often ask him: what issues do AI present to copyright law? Two issues usually come up in these discussions. The first is whether copyright protects any works created by AI. The second is whether using copyrighted works to train AI systems amounts to copyright infringement.
He begins answering these questions by explaining how Western legal systems have approached this issue. The United States and Europe have not yet offered protection for AI-generated work. Generally, two objections support this: the constitutional objection and the economic objection.
The Constitutional Objection
In United States’ law, the objection to protection of AI-generated works is that the constitutional language does not allow for it. The Copyright Clause in the U.S. Constitution grants protection to “authors and inventors.” Many U.S. government agencies and scholars argue that, within this definition, AI cannot be an author. Further, the Compendium, which is a manual that the United States Copyright Office uses to determine copyrightability, states that protectable work must be created by a human being. Therefore, the Office will not register any works that operate without creative input from a human author.
In the past few months, copyright scholars began to question the position taken by the agency as well as the courts. For example, we look to an article written by Edward Lee, called Prompting Progress: Authorship in the Age of AI. Lee explained that since photographers fine-tune to create their final product, they deserve a copyright interest in their works. He argued that people who use AI may be fine-tuning in a similar way that photographers do.
To answer the question of whether AI-generated works can be copyrightable, we can look back to 19th century cases about copyrightability in photos taken by camera. In the U.S. case Burrow-Giles Lithographic Company v. Sarony,[1] the issue was whether a photo taken with a camera is protected by copyright. The Court held that the owner is he to whom anything owes origin. Yu states, since “he” could mean “she” or even “it,” this decision doesn’t determine whether AI is eligible for copyright. However, this piece at the end of the decision might help: “the Constitution is broad enough to (…) authorize the copyrights of photographs, so long as they are representatives of the original intellectual conceptions of the author.” Yu argues that the original intellectual conception of the author could become increasingly important to the discussion of copyrightability of AI-generated works.
The Economic Objection
The second objection to the copyrightability of AI-generated works is that there is no economic incentive to grant protection to computers. However, Yu states that there is a good opposing argument to this stance. Since people are increasingly investing into the production AI machine-generated data, then we may need protection for those investments. There is precedent about database protection, e.g., the EU which we can draw from.
The seminar was attended by students in a variety of programs, both in-person and virtually
The Future of Copyright Law
We have focused so much on what the future of copyright law ought to be. However, Yu argues that we also think about what the future of copyright will be.
First, he looks to China, because it is a jurisdiction beginning to offer strong copyright protection for AI-generated works. Two cases reflect its position on AI copyrightability: Shenzhen Tencent Computer System Co. v. Shanghai Yingxun Technology Co. Ltd.[2] and Beijing Film Law Firm v Beijing Baidu Netcom Science & Technology Co Ltd.[3]In Shenzhen, the Court granted copyright protection to a finance report generated by AI. The Court held that the report was copyrightable, not because AI created the report, but because a creative team operated the AI system. This fulfills the “arrangement selection” element of the copyright statute. In Beijing, the Court held that the plaintiff’s use of prompts to generate an image reflect the plaintiff’s selection and arrangement. This is similar reasoning to Shenzhen, but the Court went further. They reasoned that even after the plaintiff used his prompts, he continued to refine the photograph by repeatedly prompting the AI-software, until it finally generated a result suitable to his liking.
A different example in East Asia is the case of South Korea. South Korea is in the process of passing a bill which provides sui generis protection for AI-generated works, though it is still underway. The bill proposes up to five years of protection after public disclosure.
These stances are different from the stance taken at the U.S. Copyright Office. Three cases show the U.S. position: Sarony, the case of artist Jason Allen, and the case of the comic book Zarya of the Dawn. Together, they show that the U.S. Copyright Office generally protects compilations of copyright works or data (i.e. their selection or arrangement), but generally denies protection for works created by AI.
Some U.S. policy makers argue the U.S. is not obliged to follow the decisions of China and South Korea. However, this may not be entirely true, due to the influence of the Agreement on Trade-Related Aspects of Intellectual Property Rights (TRIPS). Yu notes that since TRIPS member states are enacting legislation to protect AI-generated works, the United States may choose to follow suit. The question is, will they? Yu states that we cannot answer this question by considering only the constitutional and economic argument. We must also consider the impact of what other countries are doing.
Further, there is a tendency for us to make a strict distinction between AI-generated works and human-generated creation. While the White Paper from the World Intellectual Property Organization Copyright Treaty (WCT) that this is a good distinction to make, Yu argues otherwise. He states that we are more accurately in a hybrid situation, where certain percentages of works are made by humans, and certain percentages are made by machine. Further, if human creative input continues to drop, and machine creative input continues to increase, at what point do we need to change our copyright system? This is a big question for every country trying to decide how they will address this. Yu notes that since this question is based in policy, different countries will likely reach different conclusions.
Ingestion and Reverse Legal Transplanting
Ingestion refers to the use of copyrighted works to train AI systems. There are many pending lawsuits in the United States about whether ingestion constitutes an infringement of copyright, and whether this will be covered by fair use. At this point, the creative output from AI machines could compete with the underlying works that have been used by the system. Further, for those jurisdictions which do not have fair use and fair dealing, they may resort to other exceptions to copyright infringement such as “the text and data mining exceptions.”
By contrast to the U.S. approach, Japan’s copyright statute grants exceptions for copying if made outside of the personal enjoyment of the thoughts or sentiments in the protected work. Although this rule is broad, the statute provides three examples of such an exception. They are to test and develop technology into practical use, to use in data analysis, and to process computer data. Japanese law therefore permits ingestion. This is a contrast to the United States’ approach, which requires a multi-factor analysis of fair use before reaching a conclusion.
Yu discusses the case of Japan to show that the text and data mining exception came long before its adoption in the U.K. and E.U. This describes a type of legal transplanting which is not common. The theory of legal transplantation, which describes a process by which one country “transplants” the law of another, is usually observed form Western countries (Europe and U.S.) to other countries. However, with AI, we see legal transplanting from Asian countries like Japan, and possibly recent developments in Singapore, to Western, developed countries. This is what we consider a “reverse legal transplant.”
Professor Yu’s Three Observations
The second half of Professor Yu’s presentation was directed at three observations that will be helpful to analyzing the question of AI in copyright law.
1. Will we see divergence or convergence between differing AI developments in other countries?
Whether there is a divergence or convergence is a typical question asked in comparative law studies. Divergence means that countries adopt similar laws about certain topic, where divergence means that they differ on a topic. Yu explains that so far, there is convergence on two issues. First, no country has ever granted copyrightability to AI as an author. Second, some countries have allowed, or are in the process of allowing, the use of copyrighted works for training purposes. This has been done through specific exceptions to copyright infringement. However, since the issues are novel, we may see divergence as well. Consequently, we could also have “crossvergence,” a term coined by Professor Yu. Crossvergence describes a hybrid situation where convergence and divergence occur at the same time, but on different topics.
2. Can we regulate AI through international treaty?
Professor Yu’s second observation is that copyright law regarding AI may be regulated through international treaty. However, Yu explains that treaties are not an obvious route because of the lengthy process involved before they take effect.
Further, the “Brussels Effect” could have impact. The Brussels Effect is a phenomenon by which non-E.U. countries tend to abide by E.U. regulations, despite not having any legal obligation to do so. Since the power of E.U. market is strong, other countries find it impractical to not adopt E.U. standards. So, if the E.U. prescribes AI policy, it is possible that other countries will follow it.
It is also interesting to consider the global race in AI. Many countries are thinking about AI for national security purposes, so it is strategic for them to promote AI. It is possible that we may see “AI nationalism,” which is the AI version of the vaccine nationalism we experienced during the pandemic.
Further, we might seek international soft law, as opposed to treaty formation, to form international norms on AI. Although soft law is not legally binding, it can hold some persuasive force. Yu states that the soft law approach might be better than no international norm at all.
3. Potential Copyright Law Changes
Yu moves on to give the audience five issues to think about regarding potential changes in copyright law.
1) First, there may be changes in the idea-expression dichotomy. In prior copyright law, there used to be a distinction between expressions and ideas. Mark A. Lemley writes an article in in the Stanford Law Review called How Generative AI Turns Copyright Law Upside Down. He explains that copyright law was originally formed to protect the expression of an idea rather than the idea itself. Now, an expression is generated with AI using prompts. This means that ideas have become more important in guiding that expression. This changes what we think about copyrightability of AI-generated works.
2) The second issue is about copyright infringement. In U.S. Law schools, the case of Ty v. GMA Accessories, Inc.[4] is an important part of the copyright law curriculum. In this 1997 decision, the court held that the defendant’s access to the infringed work can be assumed in an infringement suit if similarities between two works are strong. However, now, it is not possible to know exactly what access the defendant has if they used AI, neither is it possible to know what work has been used to train the AI model. The question becomes more difficult because the data used in machine training is not transparent. In response, the European Union is pushing to require disclosure of the data used in training AI models.
3) The third issue is about Fair Use. This is very important in the U.S., as well as countries like Singapore. Under the fair use exception, the test consists of four factors: (1) the purpose and character of use, (2) the nature of the copyrighted work, (3) the amount and substantiality of the portion used, and (4) the effect of the use upon the potential market of the copyrighted work. Usually, the first and fourth prong weigh heavily in the legal analysis, and the second factor is not so important, with a special exception for computer software. In the context of AI, the second factor might become much more important. Consequently, we may see a resurgence of the importance the second factor.
4) The fourth issue is a change to the Intermediate Copying Doctrine, which allows for copying of protected work for the purpose of reverse-engineering. This allows an individual to create a different product that competes with the copyrighted one. The key to this doctrine is that the final product will not include the protected work. Changes to this doctrine will become more important in the future.
5) The Fifth issue is Professor Yu’s fear that our scope is too narrow. The scope of AI issues is much broader than copyright issues; they are also relevant to areas like national security and trade. Therefore, we cannot just look at caselaw in copyright to determine what to do with AI. Further, Yu proposes that we are asking copyright law to do too much. AI presents to us many ethical issues that we need to resolve. We don’t want to treat copyright law as the single arena to address our technical concerns because this would put too much strain on copyright law.
Mita Williams’ Commentary
Mita Williams provided a commentary to Peter Yu’s presentation. Mita Williams is the Acting Law Librarian at the Windsor Law Library, who is also widely published. She speaks and writes about libraries, civic technology, and game design. Recently, she has been directing much of her attention to the emergence of large language models as a new form of cultural artifact.
Mita Williams provides her commentary and questions for Professor Yu
Williams emphasizes how Canadians know the impact of trade agreements on our intellectual property landscape. These agreements create significant changes to the Canadian public domain and the materials within them.
Williams notes that trade is caused by an imbalance of resources across states. For example, when one country has technology but no resources, and another country has resources but no technology, they will enter trade agreements to receive mutual benefit. She suggests that a new agreement in AI may “level” that imbalance. She wonders whether this will bring difficulties to any new agreements. She asks Professor Yu: how much will the global state of copyright be based in trade protectionism?
Professor Yu considers this is a very good question. He states that you can understand the pro-AI position taken by Asian countries as a strategic decision because it can benefit their economies and raise their world positioning. However, you can also see this as a response to the past 30 years of western maximum IP protection policies. Further, many governments worldwide are not providing enough public funding to develop AI networks. If the government supplies funding this will obviously be easier, but since they’re asking individuals to develop AI networks, the government needs to provide the support and protection.
With regards to an international regime of copyright, it will be difficult to enforce international AI regulation for a couple reasons. First, until powerful countries can agree on common policy, they will not be able to persuade other countries to join the agreement. For example, in TRIPS, the U.S., E.U. and Japan were able to form a position and convince other countries to join it. However, in AI, we don’t even have an agreement between the U.S. and E.U. Plus, China has become more powerful since then. Although major powers can take small steps, it still proves difficult to find consensus on this topic.
Second, Yu tells us that he is currently co-authoring an article called “The Race to the Middle.” He believes many countries will race not towards the top, not towards the bottom, but towards the middle in term of AI regulation. If we can place AI model training within non-expressive uses category (i.e., treat such use as an exception to copyright infringement), we may be able to have a consensus between countries. However, Yu states, we must wait to see how it will actually play out.
Windsor Law LTEC Lab extends a big thank-you to Professor Yu for sharing his insights with us and for Mita Williams for providing an insightful commentary. We also wish to give a big thanks to Ms. Sue Arnold, Ms. Dalia Mazhar, and LTEC Lab R.A.s Brandon Beck and Khadija Shamisa for their efforts in making this event a success.
Citations
[1] Burrow-Giles Lithographic Co. v Sarony, 111 U.S. 53 (1884).
[2] Shenzhen Tencent Computer System Co. Ltd. v. Shanghai Yingxun Technology Co. Ltd. (2019), Guangdong 0305 Civil First Trial No. 14010 (Nanshan District Court of Shenzhen China) [Shenzen].
[3] Beijing Film Law Firm v Beijing Baidu Netcom Science & Technology Co Ltd. (2019), Jing 0491 Min Chu No. 239 (Beijing Internet Court China) [Beijing].
[4] Ty, Inc. v. GMA Accessories, Inc., 959 F. Supp. 936 (N.D. Ill. 1997).