The use of generative Artificial Intelligence (gAI) is a rapidly increasing due to its widespread availability. These gAI programs use large training sets to ‘learn’ so that they can produce desired content including text, speech, visual art, and video using a prompt supplied by a user.
This article will address how copyright is implicated by this increase in the use of gAI. More specifically, it will address three main questions: what are the copyright implications for the training sets used by gAI?; what are the copyright implications for user supplied prompts?; and what are the copyright implications for content produced by gAI? In order to minimize the length of this article, it will cover the use of gAI to create visual art. However, many of the topics discussed herein extend beyond visual art.
A gAI program uses creative works by you or others as training sets to “create” desired content. This “use” is often done without permission or even knowledge of the authors of the creative works. As we know, the definition of copyright infringement is the copying, distributing, performing, publicly displaying, or creation of a derivative work without the permission of the author. The question is: who is liable for this infringement?
Liability for copyright infringement relating to training sets is the subject of several pending court cases in the US. In one example, art was taken without permission from Deviant Art, an online gallery and community that was later purchased by Wix. After the purchase, Wix buried an opt OUT provision on their website. The artists were not aware of and were not asked permission to use their art. In another example, artists were opted IN without their consent on Art Station, a professional portfolio site. Mid journey, who used this data set admitted to using hundreds of millions of images without permission. Art Station artists protested once they learned of the use. There, artists posted an image on their front page of “AI” with a red circle and line through it. AI images created following the protest had visible artifacts of the no AI image. In a third example, a web crawler, Lannion 5B, was used by Stable Diffusion. There, no opt in or opt out was possible. These are some examples of large amounts of copyrightable works that were used without permission by gAI products. At the least, it is expected that this “use” by gAI will fall under copying and distributing without permission.
As a user of gAI, you might wonder if your prompts or the content created using your prompts is subject to copyright protection. Several criteria are used to determine whether or not something is copyrightable in the US. Some of the most relevant requirements here are: was the work created by a human author?; does the work constitute copyrightable subject matter?; is the work sufficiently original?; and was the work independently created? We will take each of these in turn.
Human authorship: Copyright Law is limited to original intellectual conceptions of the author. The office will refuse to register a claim if it determines that a human being did not create the work. See, Burrow-Giles Lithographic Co v. Sarony, 111 US, 53, 58 (1884). A few examples of these refusals for non human authorship include the Monkey Selfie, a painting by an elephant, and most recently, visual art in a comic created using Mid Journey (Zara of the Dawn).
Copyrightable subject matter: Works of authorship include the following categories: 1) literary works, 2) musical works including any accompanying words, 3) dramatic works, including any accompanying music, 4) pantomimes and choreographic works, 5) pictorial, graphic, and sculptural works, 6) motion pictures and other audio, visual works, 7) sound recordings and 8) architectural works. Items that are NOT copyrightable include ideas, procedures, processes, systems, methods of operation, concepts, principles, or discoveries.
Originality: To qualify for copyright protection, a work must be original to the author, which means that the work must be independently created by the author, and it must possess at least some minimal degree of creativity. See, Feist Publications Inc. v. Rural Telephone Service Company, Inc., 499 US 340, 345 (1991).
Independent Creation: To qualify for copyright protection, the work must be independently created by the author. The term “independent creation” means that the author created the work without copying it from other works. See, Feist 499 US at 345. The copyright law protects those components of a work that are original to the author, but originality does not require novelty. Id. at 345, 348. A work may satisfy the independent creation requirement even though it closely resembles other works, so long as the similarity is fortuitous, and not the result of copying. Id. at 345.
The use of gAI to produce copyrightable content runs afoul of at the least, the human authorship, originality, and independent creation elements required for something to be copyrightable. In addition, if the content is deemed a form of compilation, the authorship involved in selecting, coordinating, and arranging pre-existing material or data must be objectively revealed in the deposit copies to be copyrightable. See, Feist, 499 US at 348. Additionally, originality requires that the author make the selection or arrangement independently, and that it displays some minimal level of creativity. For instance, a compilation of statistics is not copyrightable if the author’s selection, coordination, or arrangement of data is not evident in the claim. Here, the content created by gAI is an output - a piece of content, whether that be a piece of visual art, a simulated photograph, text, or the like. There is no way for the user of gAI to control or represent exactly how gAI uses the training set to come to the final content and thus not only did the user’s prompt not specifically select, coordinate, or arrange the preexisting material, but it is not objectively revealed in the deposit copy, thus barring copyright protection for that product.
User prompts likely fail at least the element for copyrightable subject matter, as they are likely merely ideas or concepts. Additionally, user prompts have other barriers to copyright protection. For example, under copyright, the merger of idea and expression is when there may be only one way, or only a limited number of ways to express a particular idea, procedure, process, system, method of operation, concept, principle, or discovery. This would be the case where the user prompt is so specific that arguably there is only one or very few ways it could produce a certain type of content. In one known example, the courts stated that the office cannot register a claim in the mere idea for a story that is based on a common theme, such as “a quarrel between a Jewish father and an Irish father, the marriage of their children, the birth of grandchildren, and a reconciliation.” See, Nichols v. Universal Pictures Corp., 45 F.2d. 119, 122 (2nd Cir 1930). And, in another, the office cannot register a claim based solely on standard programming techniques that are commonly used to achieve a specific result in a computer program. See, e.g., Sega Enterprises Ltd. v. Accolade, Inc., 977 F.2d 1510, 1524 (9th Cir 1992).
Copyright is implicated in each stage of the gAI process, from the use of training sets to user supplied prompts, to the resulting content. As consumers and as attorneys, this is an exciting area to watch and something to be cognizant of before you use gAI and before you advise your clients to use gAI in their business. It is important to advise clients to carefully read the Terms of Service for any gAI product prior to use and to consider creating policies for their employees around permitted uses of generative AI in the workplace. Additionally, there are issues such as bias in the training sets, confidentiality of user supplied inputs, and the like, that must also be considered. At this writing, there is also great concern for what are referred to as hallucinations by gAI, where the result is not accurate but nevertheless the gAI product delivers it as a confident response not justified by its training data. It is indeed an exciting time to be an IP attorney. I am curious to see how the use of gAI is embraced and regulated so that we can avoid at least some of the downside to its widespread adoption.
Comments