Over the last several years, new technologies have rapidly developed in the realm of generative AIs, computer models that can produce images or text outputs through user prompts. These models operate using neural networks, which take works with tagged associations—for example, a picture of a dog tagged with “dog”—and produce similar works according to user prompts (“Generative AI”, 2023). As generative AI like DallE and ChatGPT rapidly gain prominence in our society, new challenges for courts arise in the realm of copyright. U.S. law must decide such questions including who owns the copyright for AI-generated work and whether they should have copyright at all. It must be asked whether the provision of that copyright would be beneficial or detrimental for the U.S. economy. From a legal perspective, AI-generated works should not enjoy copyright protection. Additionally, it would be more beneficial for artists if AI-generated works did not enjoy copyright protection.
It is well established in US copyright law that human authorship is a requirement for copyright protection. One case in which this was heavily debated was the case of Naruto the black macaque monkey, in which the photographer David Slater set up a camera that a monkey later took a selfie with (Guadamuz, 2018). The issue in dispute was whether the photo’s copyright was held by Slater, held by the monkey Naruto, or whether the photo was public domain. Publication of the photo by Wikipedia, which labeled the photo as public domain, led Slater to bring a case to the U.S. Copyright Office (Perkal, 2018). The Copyright Office concluded that “because copyright law is limited to ‘original intellectual conceptions of the author’”, it would not be possible for Slater to register the work under his copyright (Perkal, 2018). After all, the created work was not something he facilitated. Simultaneously, because U.S. courts only grant standing (the ability to sue someone) in court to human actors, it would be impossible for Naruto the monkey to claim copyright over the photo (Guadamuz, 2018; Perkal, 2018).
In the same vein, AI-generated works should not enjoy copyright protection because of their lack of human authorship. The vast majority of AI-generated work operates through the use of text prompts, in which a user gives the model several short phrases from which it generates either text or an image. Unlike in other artforms, such as writing or drawing with pencils, the user has no direct control over what the work that the model produces looks like. They can make alterations to the input prompts to generate different images, but nowhere in the process does the user have any control over the direct placement of pixels in an image, or words on a page. This is a key point for the nature of copyright, as discussed in the case Thaler v Perlmutter, one of the first cases to evaluate whether AI-generated artworks were copyrightable. It evaluated the case of Stephen Thaler who sued the U.S. copyright office after it refused to provide copyright for the work “A Recent Entrance to Paradise,” created using AI image generation software (Thaler v Perlmutter 2023, 2).
The court in Thaler evaluated several historical trends, including the invention of photographs, to determine whether a new technology such as AI changed how copyright would function (Thaler v Perlmutter 2023, 8). Past courts eventually decided that works like photographs or film did deserve copyright protection because the authorial intent of the photographer directly correlated to the outcome of the text. Specifically, the photographer arranged the subject to “evok[e] the desired expression” to craft the “overall image” (Thaler v Perlmutter 2023, 8). In contrast to those photographers, users of generative AI offer no direct control over the output of their work. They may define the inputs of a work, but they have no direct control over what is produced on the screen by the generative model. Specifically, the same input can have many different outputs, making it unlike any other form of art in which the same inputs produce the same result, i.e. taking a photo of the same location at the same time of day with the same camera settings will produce the same photo. Thaler concludes that AI-generated works do not, using the example of a garden grown over time with the help of human action. Specifically, there can be no copyright found in the arrangement of a “cultivated garden,” because despite a human hand guiding the creation of that garden, the end result of that garden was not directly controlled by human authorship (Thaler v Perlmutter 2023, 12). Even if a human created “the plan” for the garden, it was other forces (such as nature) that guided the actual result of the garden, so copyright is not afforded to a human (Thaler v Perlmutter 2023, 12). This logic can be applied to AI-generated works as well. A user of generative AI may be the cultivator of the work by submitting inputs to a machine to produce an image, but they do not directly control the end product. Instead, the force, or blackbox of the AI model, shapes the final product. Because it cannot be said that the users in question have control over their works, the work is not their original intellectual conception and is not copyrightable.
Despite this distinction, however, there is some ambiguity in the Copyright Act of 1976 as to whether AI-generated work should be considered copyrightable by the person using generative AI. After all, one might argue that generative AI is an artistic tool like any other. One could also argue that creating a work does not always require direct intention towards every aspect of it, only some intention towards its creation. For example, a photographer who takes a photo but suddenly realizes that a person was hiding in the background of it does not suddenly lose their copyright of that photo. This forces the U.S. legal system to confront not just precedent but also the original purpose of U.S. copyright law. Copyright was created as a mechanism in the U.S. Constitution to “promote the progress of science and useful arts” by guaranteeing that authors had control over the reproduction and distribution of their works (U.S. Const., art. I, §8). Specifically, copyright ensured that other entities would not be able to reproduce an artist’s work before that author was rightfully compensated for their time and labor by the market, which allowed artists to benefit economically from their efforts.
Affording copyright protection to AI-generated work would go against the desired aims of U.S. copyright law. If the goal of copyright law is to afford more economic possibilities for artists to support themselves on their work, the copyrightability of AI-generated artworks would hurt this. Before the existence of generative AI, companies would hire illustrators to create works for promotions and releases (Zhou, 2023). With the advent of generative AI, however, companies are able to use AI models instead of hiring illustrators. When looking for promotional material and image backgrounds, companies like Tencent resorted to using AI-image creation instead of hiring illustrators, resulting in a loss of 70% of illustrator jobs in China over the last year (Zhou, 2023). Instead of allowing more artists the ability to subsist off their work, AI-generated art denies it, ensuring that fewer and fewer artists can make a living creating art. Even if a small portion of artists are able to complete more work using generative AI, the overall job loss is a net-negative for illustrators. Therefore, the proliferation of AI-generated works would go against the purposes of the U.S. copyright act by removing the ability of more artists to maintain a living off of their work.
Instead of affording copyright protection to AI-generated art, U.S. courts should deny it. By denying copyright protection to AI art, the U.S. would ensure that only works with human authors would be protected. If work created by AI was freely distributable in the public domain, companies would have a strong incentive not to employ AI art when creating products. Instead, they would be more inclined to hire traditional human artists, preventing any job loss from occurring. The U.S. is in a time of legal ambiguity for copyright, and it is unknown whether precedent will hold or if future courts will decide differently than the Thaler court. To ensure the best quality of life for artists, and support the protection of their livelihoods, the U.S. should deny copyright to AI-created works.
The views expressed in this publication are the author’s own and do not necessarily reflect the position of The Rice Journal of Public Policy, its staff, or its Editorial Board.
References
“Generative AI,” Nvidia, https://www.nvidia.com/en-us/glossary/data-science/generative-ai/, Accessed 5 Nov. 2023.
Guadamuz, Andres. “Can the monkey selfie case teach us anything about copyright law?” Wipo Magazine, Feb. 2018, https://www.wipo.int/wipo_magazine/en/2018/01/article_0007.html, Accessed 5 Nov.
2023.
Perkal, Paulina Julia. “Monkey business finally settled: the ‘monkey selfie’ disputes” Kluwer
Copyright Blog, 5 Feb. 2018, https://copyrightblog.kluweriplaw.com/2018/02/05/monkey-business-finally-settled-monkey-selfie-disputes/, Accessed 5 Nov. 2023.
United States District Court for the District of Columbia. Thaler v Perlmutter, 22-1564.
Zhou, Viola. “AI is already taking video game illustrators’ jobs in China,” Rest of World, 11 Apr. 2023, https://restofworld.org/2023/ai-image-china-video-game-layoffs/, Accessed 5 Nov. 2023.
Photo by Electric Frontier Foundation