table of contents
I. Introduction
2. Originality
III. Authorship (Subject)
4. Author's original expression/"contribution"
5. Fixation and reproducibility
VI. Conclusions
This article is reprinted from the official account of "Journal of Copyright Theory and Practice", written by Sun Yuanzhao, and the original article was published in the third issue of Copyright Theory and Practice in 2024.
【Executive Summary】
This article explores the "fundamentals" of the copyright protection system and some of its core concepts by tracing back to its origins, in order to get to the root of the problem, and then explore whether AI-generated content (including the "Wenshengtu" infringement case recently decided by the Beijing Internet Court) is possible to obtain copyright protection or is there another way. This paper attempts to conduct comparative research and discussion on international and domestic developments and opinions, and puts forward relevant conclusions.
【Keywords】
Artificial Intelligence, Wensheng Diagram, Authorship, Originality, "Spring Breeze Brings Tenderness" Case
I. Introduction
The so-called "everything is inseparable". [1] There is a very similar core concept in martial arts (Taijiquan): "Although there are many changes, reason is consistent". [2] In fact, regardless of civil and military affairs, this central idea is also applicable to the method of studying legal issues, including copyright[3] and other intellectual property rights, and its main meaning is that the examination of the problem must return to and focus on the fundamentals and explore, and only by tracing back to the source can we get to the root of the problem and steadily deduce how to keep pace with the times. [4]
In 1710, the British Parliament passed the Statute of Anne, which was the first example of copyright protection in modern times. [5] After three centuries of continuous evolution, especially the challenges and tests of countless new technologies in the several waves of industrial revolutions, although there are still many problems in the current copyright law, the basic structure, organization, texture, purpose and orientation have been developed quite mature and complete. The "foundation" of the modern copyright protection system consists of at least four aspects: (1) from the protection of publishers to the protection of authors, (2) from the indefinite protection of public monopoly to the protection of private rights with a limited term, (3) from almost absolute exclusive rights to relative exclusive protection, which must take into account the delicate and dynamic balance between authors' rights and social welfare, and (4) from fragmentation to global consistent standards, which must meet the minimum mandatory standards stipulated in international conventions. [6]
At present, the copyright legal systems of various countries or regions around the world are based on these four aspects, which derives the regulation of how long the author can enjoy personality rights and property rights, whether it is the author's original expression as the premise of whether he can enjoy copyright, and whether or not these rights must be required to make some concessions based on the needs of some special social public welfare to balance the needs of society or economic development. In other words, originality, subject of protection (authorship), and fair use (or "limitation of rights") are the "thresholds" that are difficult to circumvent and must pass as long as copyright issues are involved.
In the 60s of the 20th century, when modern computers and artificial intelligence were still in their infancy, scholars began to discuss the issue of artificial intelligence and copyright. [7] After more than half a century of research and discussion, it has been determined that these three main axes cannot be bypassed in any way. Since fair use is only used as a defense in the event of infringement, this article will focus on the first two and discuss other related issues.
2. Originality
Traditionally, all content automatically generated by machines, whether it is text, font, image, sorting, etc., basically cannot enjoy copyright protection. This is based on a classic case that has influenced the development of the global copyright legal system, namely the "White Pages Telephone Directory Case" decided by the United States Supreme Court in 1991, which clarified for the first time "original authorship" (also known as originality, or translated as "original authorship or qualification") is more in line with the original meaning) is the most basic requirement for obtaining copyright protection, and defines its specific connotation. [8] In that case, the court held that although the defendant had copied the plaintiff's telephone directory in large quantities, the specific contents (name, address, telephone number, etc.) were made up of "facts" to which it had no rights, even in alphabetical order, and in fact it was done "with one click" through computer operation, even without even the slightest amount of creativity. The Federal Supreme Court further stated that:
"In considering whether a fact-based product constitutes an original work, [the court] should focus on how the facts collected are screened, reconciled, and arranged. ...... Facts are never original, so the authors of the compilation against them can only assert originality in terms of how they are presented. ...... The requirements for originality are not strict. ...... Innovation is not required. Originality requires only that the author has engaged in the selection or arrangement alone (i.e., there is no choice or arrangement to reproduce other works) and that a minimum degree of creativity has been demonstrated. ”[9]
At the end of the judgment, the court emphasized that the judgment should not in any case be read as a disparagement of the plaintiff's compilation of his telephone directory, but rather as an indication that copyright was a reward for the author's creativity rather than his efforts.
Accordingly, it has been the policy of the U.S. Copyright Office that copyright is not registered for content that is generated randomly or automatically by machines or mechanical programs without any human creative input or intervention. [10] The key question is whether the author of a "work" is essentially a natural person, whether the computer or other device used by it is only used as an aid, or whether the traditional elements for determining authorship (elements of expression or selection, arrangement, etc., such as literature, art, or music) are not in fact natural persons, but are conceived and executed by a machine or system, and therefore cannot cross the minimum threshold of originality. [11]
In the European Union, the Court of Justice of the European Union (CJEU) has issued seven judgments directly related to copyright originality since 2009, and the judicial opinions of the United States have basically converged. [12] The CJEU stated that under the EU's historical directives relating to copyright (in particular the Information Society Directive of 2001), a "work" means first and foremost an object that is original (or original), i.e. can only be derived from the author's own intellectual creation and reflect the author's personality. [13] Second, to qualify as a "work", it must be constituted by the expression of the intellectual creation. Conversely, if a creation is determined only by technical limitations or rules, then the creation is not original. The Court stated that the concept of "work" must have a sufficiently precise and objective meaning to be an identifiable object, i.e. it must be "fixed" or "fixed". The Court ultimately made it clear that the national law of any EU member state would not comply with EU law when regulating the protection conferred on copyright by any requirement other than originality (e.g. having a certain aesthetic or visual effect on the work). [14]
III. Authorship (Subject)
Another element that is linked to the element of originality is the identity or qualification of the author, i.e., the subject to be protected by copyright.
In terms of international norms, such as the Berne Convention for the Protection of Literary and Artistic Works and the World Trade Organization's Agreement on Trade-Related Aspects of Intellectual Property Protection, i.e., the TRIPS Agreement) and other major legal instruments do not define "authorship". However, according to Professor Sam Ricketson's research, during the negotiations to conclude the 1967 Berne Convention (Stockholm Act), "Member States generally agreed on the meaning of the name and therefore did not see the need for a further definition...... Therefore, the only logical interpretation of the Convention as to 'authorship' or 'authorship' refers to the natural person who created the work". [15] Moreover, Professor Ricketson points out that if one examines the text of the Berne Convention, it is not difficult to find that the leitmotiv (or leitmotif) that runs through and underpins most of the provisions refers to the creation of human beings (natural persons). For example, the term of protection of copyright is the lifetime of the author plus 50 years after his death, and it is obvious that he or she refers to a natural person, not a machine (although for various reasons such as tax savings and technological obsolescence, machines may be depreciated and scrapped earlier). [16]
In terms of domestic law or regional law, the U.S. Copyright Act does not explicitly stipulate it, but the practice of the judiciary and the U.S. Copyright Office has established humanism, that is, the identity of the author must be a natural person. [17] The French copyright law (governed by Title 1 of the country's Code de la propriété intellectuelle) begins by establishing the personality attributes and rights of the copyright author in Article 1 (machines obviously have no personality to speak of),[18] and in the case of certain types of works (e.g., collaborative or collaborative works, audiovisual works, etc.), the author must be a natural person. [19] In addition, a 2015 decision of the French Supreme Court strongly implied that the author of copyright was the exclusive property of a natural person. [20]
The German Copyright Act clearly defines the "author" as the creator of the work. [21] Articles 1 and 2 of Title I of the Spanish Intellectual Property Code (i.e. the Copyright Code) also specify that the "author" must be a natural person. [22]
In Australia, the Court of Appeal issued a classic judgment in 2010 in an inter-court manner that explicitly denied copyright protection to AI-generated databases (phone books). [23] As for the Court of Justice of the European Union, in its 2009 decision in the Infopaq case, it established that a copyrighted work must be derived from "the author's own intellectual creation and originality". [24]
The more special one is the United Kingdom. Article 9(3) of the Copyright, Designs and Patents Act 1988 stipulates that "the author of a computer-generated literary, dramatic, musical or artistic work is an individual who makes arrangements necessary for the creation of the work." In addition, article 178 specifically defines "computer-generated, in relation to a work" as "a work produced by a computer (computer) in the absence of a human author". [25] When the law was still before the House of Commons in 1987, the then Secretary of State for Trade and Industry, Lord Young of Graffham, said that as a drafter, the new definition in section 178 "was the first attempt anywhere in the world to attempt to legislate copyright in response to the advent of artificial intelligence." ”[26]
If we simply define a law just to show that it is "forward-looking", we will often end up with problems that will be difficult to deal with in the future. The situation here is that once the two articles are juxtaposed, and the requirement of originality is added, it can be seen that they are irreconcilable. What is certain is that, under the current statutory law in the United Kingdom, there must be at least one natural person "author" (subject) in any work, and in the case of content automatically generated by a computer system, the author is either the creator or developer of the entire AI system (but has little direct involvement or connection with the various objects generated by the program), or the person who makes the necessary creative arrangements for the production of a particular product. There has been no judicial decision to give a direct response to this. [27]
The UK government has recognized this issue in its 2021 dedicated consultation on AI and copyright. [28] However, in the outcome report presented in June of the following year, while acknowledging that this was a very challenging problem, and on the other hand, on the grounds that AI development was still in the "early stages of development" and could not be properly evaluated, it was decided to hold it on hold until the time came to return to review it in the future. [29] The British government apparently failed to see (and in fact no one could have predicted at the time) that the timing was now: two months before the report, an innovative company called OpenAI in San Francisco, California, had just launched an artificial intelligence system called Dall-E 2 that could work on "Wensheng Diagrams", and five months later the company launched another system called ChatGPT, which quickly "exploded" The global wave of development and deployment of artificial intelligence systems has also pushed various issues, including copyright, to the forefront.
In any event, even if the definition and scope of the UK are not yet determined by its own law, the above comparison process should establish at least one basic principle: why a natural person's creative process is key to determining whether the natural person has authorship of a particular expression (or work), and whether or not a natural person is entitled to copyright and what the scope of the relevant work should depend on the result of the original expression. [30]
Fourth, the author's reading and creative expression/"contribution"
The creation of natural persons is based on their own common sense, experience, emotions, perceptions and imaginations, while artificial intelligence systems (especially the currently used large models designed to imitate human convolutional neural networks) are completely based on the data (or dataset data set) placed from their inputs, and then automatically generated (i.e., without any human intervention) specific content through the operation of software programs. Although both start from a "blank piece of paper" and have to go through various "learning" processes, the essence of which is completely different from the transformation and processing of information. One of the main reasons for this is that AI does not have common sense at all, so although it can be engaged in very in-depth and detailed work in a specific field that has been "learned" or "trained (optimized)", and far beyond the ability of humans, it cannot "touch the bypass" and engage in various horizontal related thinking and comprehension. [31] The key point is that, although the programmer can set the various parameters to be processed before each system operation or operation, the entire content generation is completely completed by the computer itself and automatically according to the statistical calculation program preset by the existing data system and software in the system. [32]
It is precisely for this reason that all operations involving the use of artificial intelligence systems, at least in the so-called "Wensheng diagram" process, belong to natural persons whose expressions are only parameters or prompts entered into the system each time, and have nothing to do with what the system will eventually generate, nor can it be predicted, let alone any degree of control. Comparatively speaking, the creation of literature or art by a natural person usually has certain layout and arrangement from the beginning. As to whether those expressions (reminders) can be protected by copyright, it is still necessary to go back to the fundamentals and examine whether they are original on a case-by-case basis, and as for content that is completely automatically generated by machines or systems, there is obviously no copyright at all.
Take, for example, the case of image infringement generated by artificial intelligence ("Spring Breeze Sends Tenderness"), which has recently aroused great interest and discussion in China, and was decided by the Beijing Internet Court. [33] In its judgment, the court detailed the parameters (or prompts, see figure below) that the plaintiff had entered into the Stable Diffusion AI system.
It can be seen from this that the part where natural persons participated or intervened ended there, and this fully reflected the content of the plaintiff's statement. There is no composition, perspective, or sketch related to the final picture, which is an indispensable basic framework or connotation of art creation. Such as "ultra-high-quality and high-detail original image data processing format color photos, exteriors, Japanese idols, highly detailed symmetrical and charming faces, angular and symmetrical faces, perfect skin, skin pores, fantastic dark eyes ......", etc., whether viewed individually or as a whole, are nothing more than a variety of ordinary, commonly used adjectives and purely functional or quite subjective expressions (according to the terminology of the Copyright Law, they are still in the stage of "thought"), and do not have the minimum originality, so they do not constitute "expression", and there is no copyright at all.
On the other hand, if we take 10,000 steps back and assume that the plaintiff enjoys copyright, then what is the scope of its rights? In fact, it should be very clear that the plaintiff gave those parameter instructions or prompts. Unless the defendant also used the same or substantially similar parameter instructions (prompts), there is still no infringement to be claimed. In any case, the plaintiff could not claim to be the "author" of a piece of content that was entirely automatically generated by a machine system (in this case, multiple frames of images).
There is an easy misunderstanding here, which is the so-called "contribution" problem. Since there is no firm basis for determining which part is the author of whom, i.e., the parts that are completely automatically generated by the machine or system are not separated from the beginning, it is natural to consider the extent to which the natural person who gives the parameter command or sets up the AI system has "contributed" to the final generated content. It should be noted that in the field of copyright, the only issue of "contribution" is that the co-author must split the profits, and that is only an approximation. The so-called "contribution" itself is highly misleading, and in fact it is simply impossible or even impossible to quantify, especially in terms of a few percent. For example, as the saying goes, "finishing touch", for the person who only lightly clicks on the longan with a pen at the end, how big is his (her's) "contribution" and how high is the proportion?
However, the basic argument and logic of the court in the case of "Spring Breeze Sends Tenderness" is that "the plaintiff designed the picture elements such as the characters and their presentation through prompts, and set the layout and composition of the picture through parameters, reflecting the plaintiff's choice and arrangement." On the other hand, after the plaintiff obtained the first picture by entering prompts and setting relevant parameters, it continued to add prompts, modify parameters, and constantly adjust and correct them, and finally obtained the pictures involved in the case, and this adjustment and correction process also reflected the plaintiff's aesthetic choice and personality judgment. ...... Therefore, the images involved in the case are not "mechanical intellectual achievements". In the absence of evidence to the contrary, it can be determined that the pictures involved in the case were completed by the plaintiff independently, reflecting the plaintiff's personalized expression. [34] According to this argument, suppose that a professor gives some principled advice to the graduate student under his supervision, and then sees that the student is not satisfied with the first draft of the thesis, so he proposes some revisions. There are many industries and markets such as the publishing business, which have serious consequences.
5. Fixation and reproducibility
Another issue of relevance is fixation versus reproducibility. The Museum of Modern Art (MOMA) in New York City held a special exhibition on AI-generated images in 2023. They first asked: If an artificial intelligence system were to go through all of the museum's collections, what would they be able to "dream" and create themselves? So the museum invited Turkish-American multimedia artist Refik Anadol to design a visual representation that never stops, never repeats, and can constantly change itself with the changes in the surrounding objective environment, called "Refik Anadol: Unsupervised" (Refik Anadol: Unsupervised). All the paintings are displayed on a large screen wall (24×24 feet, about 7.32 ×7.32 meters) and ultra-high-definition pixel projections, and the stereoscopic image scene is constantly changing according to the surrounding environmental factors (such as sunny days, rain, light, movement of people, etc., see the picture below). [35]
Image Credit: The New York Times
Therefore, this inevitably creates a lot of problems. Even if the debate over whether copyright or infringement is obtained is set aside, the mere question of who should be "attribution" to these paintings (and how to attribution) can be highly controversial and even conflicting: the person who gave the "hint" or the "prompt engineer" he hired, the AI itself, the person who developed the AI (the software designer or/and his employer), or the author who provided the objects of the machine's "deep learning"? Second, even if the content generated by AI can be considered to be empowered, how should a presentation that has never been fixed and is constantly changing, and what should be the scope of the coverage? Is it two-dimensional or three-dimensional, or both? Does it meet the requirements of copyright law for "fixation"?
Another issue that is often raised in the debate about whether AI-generated content should be empowered or not is "replicability", which has left many scholars struggling. [36] The reason why this element was put forward is that when early copyright laws were formulated in the 18th and 19th centuries, such as the United States, there was a supporting copyright registration system and required the applicant to submit samples for deposit, so it was necessary to make all copyright registration applications "fixed" to a specific medium so that the specific original expression of the parties could be determined and could be reproduced. [37] As in this special exhibition at the Metropolitan Museum of Art, the content of the exhibition is neither fixed nor reproducible, so it is practically impossible to define the content and scope of the rights of his works, even if empowerment is to be considered.
This example also shows to some extent that a human-centred copyright system designed and developed to stimulate human creativity and preserve cultural assets is incompatible with a systematic tool designed to mimic human thought and expression as closely as possible. Now, on the contrary, they want to turn the results of this system tool upside down and apply it back to the laws and regulations designed for mankind, but because it is difficult to set up and abandon the existing foundation of the copyright legal system, it seems that it is "cut to fit for purpose".
VI. Conclusion[38]
There is no internationally accepted definition of "artificial intelligence", but there is one common denominator that can be summarized from the different existing definitions: an AI system is an information processing technology that integrates models and algorithms. Professor Edsger W. Dijkstra, winner of the second Turing Award, said in his keynote speech at the 1984 Association for Computing Machinery Conference (ACM) Central and South Regional Conference: "The question of whether a computer can think is as relevant as the question of whether a submarine can swim." [39] This raises the intriguing question of whether a submarine can 'swim'. This question is actually a paradox, after all, there is no comparison between a machine with diving functions and a person who swims, and if we insist on applying the human concept of "swimming" to a lifeless machine, it will inevitably produce quite absurd results: for example, if a submarine can really "swim", what about the person sitting in the submarine? Is it also "swimming"? If it is an unmanned submarine, is the person who remotely operates the unmanned submarine on land also "swimming"? Therefore, is it appropriate to use human swimming competition rules to regulate submarines?
In the same way, at least at this stage, if people always unconsciously look at artificial intelligence with an "anthropomorphic" projection, and cannot look at this machine tool that has been developed for more than 70 years and has recently been introduced again with the logical algorithm set in the software program, it is easy to mislead themselves. After all, the biggest problem with artificial intelligence is that it has no common sense, so it never guarantees that the results of its generation and output are correct, and it is impossible to know for sure whether the factors of infringement and the risk of "nonsense" have been ruled out, so it cannot be trusted. [40] Humans have been trying to solve this problem for decades, but so far it has been unsuccessful, and it has become the "Achilles' heel" (i.e., the fatal shortcoming) of the continued development of artificial intelligence. The "deep learning" of machines is basically statistical work, which can only be quickly calculated and deduced based on the large-scale data collected, which is not at all the same concept as the original expression of human beings, and I am afraid that it cannot be analogized, otherwise it will be the same as the paradox of "whether a submarine can 'swim'".
Although the discussion and application of artificial intelligence in the society has been surging recently, and related fields have been favored by large and small investors, and have obtained unprecedented resource investment for a while, but because many of these processes are not open and transparent, it has greatly deepened the doubts and even panic of the outside world, so a wave of copyright infringement and other types of litigation has also arisen at the same time. [41] As long as there is a judgment of infringement in it, it is almost certain that more follow-up lawsuits will be filed. This will obviously greatly increase the cost and market access threshold for the subsequent development of the artificial intelligence industry, and will also have a great negative impact on related innovation and research and development. Given that these lawsuits are often protracted, unless the parties reach a pre-litigation settlement, the entire field is likely to be subject to many uncertainties for much time to come.
In any case, the first thing that must be clarified about the various problems and controversies raised by AI is who should be held responsible once a problem occurs? At present, the situation is usually that when it comes to empowerment, there is only one competition among the others, but whenever there is a problem, when it comes to obligations and responsibilities, the parties involved begin to blame each other or blame each other. There is a Roman proverb: "there is no right without remedy" (ubi jus ibi remedium), which may be used as a basic criterion for dealing with related issues of rights and responsibilities.
Even in the discussion of whether there is a need to empower AI-generated objects, it has never been denied that only natural persons can have "authorship" or "authorship" (the reverse meaning is that all objects automatically generated by machines cannot be copyrighted because they are not original), and it is not only possible to make incompatible and absolutely exclusive policy choices between the two extreme options of "blanket denial" and "ex officio empowerment". To a certain extent and scope of human control, should AI-generated objects be protected by certain rights and interests, as was the case with the protection of photographic works in the past? Even if copyright protection can be granted beyond or bypass the threshold of authorship, to whom should the rights belong? What is the object of protection? What should be the scope? In the special exhibition at the Museum of Modern Art in New York, the presentation projected on the screen is never repeated, and it changes with the surrounding environment all the time, how should rights be defined?
All in all, even if some kind of empowerment of AI-generated content is to be considered, copyright is clearly not appropriate and is not a good policy option. After all, the main purpose of copyright law is to encourage human beings to engage in the creation of culture, art and science and to protect cultural assets. Since the code of AI is not public, it does not need to be exchanged for protection, and machines do not need to be incentivized to view their products from an anthropomorphic perspective. Therefore, it may be possible to explore whether to create a special and separate right (sui generis right) to provide some limited protection, but there are still too many unknowns and difficulties that need to be further clarified and overcome, and must not be rushed, or more benefits can be obtained through the "self-regulation" of the manufacturer (for example, having users sign an agreement, similar to the current practice of obtaining consent before accessing any website or using a database). This may be in the light of the huge debate between the EU and the US in the 90s over whether databases should be given some kind of enabling protection (although there is a consensus that data itself is presumed not to be empowered). [42] The EU's choice to protect databases with special empowerment led to a number of problems, which were later severely limited by the Court of Justice of the European Union in four judgments,[43] and ultimately two empirical studies conducted by the European Union were also ineffective. [44] One of the most criticized criticisms is that this kind of empowerment is like "covering up the ears and stealing the bell", and the real purpose behind it is to protect specific investment interests, which has long deviated from the basic principles of exchanging protection for disclosure and public knowledge. After all, investing is supposed to be at your own risk. The enactment of special laws to protect the interests of specific investments is tantamount to the formation of a larger monopoly by means of legislation (public power) to promote the interests of specific enterprises or businesses in disguise.
In light of this, perhaps as New York Times columnist Ezra Klein put it, "We're so obsessed with thinking about what this technology can do that we lose sight of the more important question: how will it be used, and who will decide what it's going to do?...... There is a more banal but perhaps more pressing question: who will these machines serve?"[45] He argues that AI is not really serving consumers, but rather doing what it takes and what investors do: "We talk too much about the technology of AI and basically ignore the business models that drive it. Add to this the fact that the eye-catching display of AI serves only a business model that attracts a hype cycle of huge investments and acquisition offers. Klein also quoted Margaret Mitchell, chief ethics scientist at the artificial intelligence company Hugging Face, who was also a key leader in AI research at Google, as commenting sharply that AI is not made to predict facts, they are actually made to make things that look like facts. [46] The most dangerous situation is that people are manipulated and persuaded by intelligent tools, when in fact a small group is trying to control the majority, after all, the people behind the machine tools are programmers and commercial interests.
Source: Journal of Copyright Theory and Practice
Editor: Sharon