Artificial intelligence (AI) poses significant challenges for the legal system, especially in the field of copyright law. As AI becomes more sophisticated, there are questions about who owns the output generated by AI systems, the companies that create the systems, or the users who provide prompts and inputs. Can a balance between AI innovation and protection of the rights of creators be achieved? This article will explore the basics and traditional application of copyright law, the impact of artificial intelligence on traditional copyright protection and its application in recent litigation including a dispute over AI-generated artwork.
What is Copyright?
Copyright is a form of intellectual property that grants exclusive rights to authors of original works of authorship as soon as they create the work in a fixed and tangible form of expression or medium. The scope of copyright law covers a wide range of works, such as paintings, photographs, illustrations, musical compositions, sound recordings, computer programs, books, poems, blog posts, movies, architectural works, plays, derivative works and many others.
The idea of copyright originated in England. It was embraced by the founders in Art. I, Sec. 8, Cl. 8 of the U.S. Constitution as follows:
“promote the progress of science and useful arts, by securing for limited times to authors and inventors the exclusive right to their respective writings and discoveries.” (emphasis added)
The first federal copyright law was passed in 1790 and amended and updated thereafter. Today, U.S. Copyright Law can be found in Title 17 of the United States Code (“Copyright Act”). The U.S. Copyright Office will “register an original work of authorship, provided that the work was created by a human being.” This position is based upon the interpretation of the U.S. Constitution, the Copyright Act and case law.
What is the Purpose of Copyright Law?
The purpose of copyright law is to encourage creativity and innovation by granting the creators exclusive rights to control how their works are used and to benefit from them financially. It protects the original works of authors, artists, composers, and other creators from unauthorized copying, distribution, adaptation, or performance. Any unauthorized use of any these original works is copyright infringement. There are exceptions to this basic rule, primarily the fair use doctrine, which we will discuss below.
Ownership and Licensing
The creator owns the copyright, unless of course the creator created the work during the course of employment or enters into a contract to make the work. If created during the course of employment or under contract the work is a “work made for hire” and the employer or the contracting party owns the copyright. The owner of a copyright can sell the rights to the copyright by issuing a license or giving it to someone else, who then becomes the owner of the copyright.
When is the Work Protected?
Under current law, copyright protection begins when an author’s creation is fixed in a tangible medium of expression, such as by being written on paper or recorded. However, a copyright notice should be affixed to the medium of expression by simply adding the “©” symbol or the word “Copyright,” together with the copyright holder’s name and the year of first publication—for example, “© 2023 Maryland State Bar Association.” This designation should appear on or near the title page in printed works, and on an early screen in electronic works. For works of art, the artist should always affix their signature to the artwork. It is also highly recommended that a writer or artist register their artwork with the U.S. Copyright Office.
How Long is a Created Work Protected?
Under the Copyright Act, the duration of copyright protection depends on several factors. Generally, for works created in 1978 or thereafter, the copyright term commences upon creation and continues for the life of the author plus an additional 70 years. For employer owned copyrights, the copyright term lasts for 95 years from first publication or 120 years from creation, whichever expires first. Many works created before 1923 are now in the public domain.
Fair Use Doctrine
Fair use is the right to use a copyrighted work in certain circumstances, without permission or license from the copyright owner. Fair use allows the user to build upon prior copyrighted works in a manner that does not unfairly deprive prior copyright owners of the right to control and benefit from their works. Examples of fair use include using the copyright work for teaching, research, by a non-profit educational institution, or restricted access to students. Examples of activity specifically excluded from fair use include entertainment, any commercial activity, open access to anyone or by payment of a fee. Fair use is actually an affirmative defense to a claim of copyright infringement, meaning that the alleged infringer has the burden of proving their use was a fair use. It is codified in Section 107 of the Copyright Act, which provides that fair use of a work “for purposes such as criticism, comment, news reporting, teaching (including multiple copies for classroom use, scholarship, or research) is not an infringement of copyright.”
A derivative work is a work that is based on or derived from one or more existing creative works. The derivative work becomes a second, separate work independent in form from the first work. The Copyright Act, 17 U.S.C. § 101 defines a “derivative work” as “a work based upon one or more preexisting works, such as a translation, musical arrangement, dramatization, fictionalization, motion picture version, sound recording, art reproduction, abridgment, condensation, or any other form in which a work may be recast, transformed, or adapted.” A derivative work may also incorporate elements from different sources, such as a mashup, remix, or collage. A derivative work is subject to the copyright of the original work, unless it falls under an exception or limitation, such as fair use or fair dealing. To create a derivative work legally, one must obtain the permission of the original copyright holder. It is important to note that the copyright of a derivative work is separate from the copyright to the original work. The copyright owner of the original work also owns the copyright rights to derivative works. Therefore, the owner of the original work may bring a copyright infringement lawsuit against someone who creates a derivative work without permission. A copyright for derivative work only covers the additions or changes to the original work, not the original itself. Therefore, if the copyright owner gives someone a license to create a derivative work, the copyright owner of the original work retains the copyright to the original work. In other words, only the derivative rights are being licensed.
Large Language Models
Large language models like ChatGPT 3.5 and 4.0, and others, are computer generated networks that can generate natural language texts based on a given input or prompt. They are trained on massive amounts of text data (“dataset”) from various sources, such as books, websites, social media, etc. The models learn the statistical patterns and relationships between words, sentences, and topics from the data, and use them to produce coherent and relevant (sometimes not relevant and sometimes hallucinated, see “NY Lawyers and Law Firm Sanctioned for Citing Fake Cases Derived from AI” and “Using ChatGPT for Legal Research. Not so fast!“) responses. The models can also adapt to different styles, tones, and domains of language by using parameters that control the output.
Recent Litigation Involving Artificial Intelligence and Copyright Law
Silverman v. Open AI
Sarah Silverman, author and copyrighted owner of a book called The Bedwetter, filed class action lawsuits against OpenAI and Meta, on behalf of other authors, alleging copyright infringement under the Copyright Act and the Digital Millennium Copyright Act (17 U.S.C. s. 1202). The Digital Millennium Copyright Act (DMCA) generally addresses the rights of creators’ and their copyright protections on the internet.
OpenAI offers and sells products that are large language models, as do other tech companies. See The Next Generation of AI Tools. As discussed, a large language model is software trained on a dataset that will give a response based upon a prompt. It relies upon the dataset upon which it is trained to produce new content. According to Silverman’s lawsuit, the dataset OpenAI uses to train its large language model comes from copyrighted works, “without consent, without credit, and without compensation.” She alleges, the large language models offered and sold by OpenAI (ChatGPT 3.5 and 4.0) can produce writing in the style of any author in the dataset based upon patterns and connections it learned from that author’s work, without a license. Ms. Silverman further alleges that OpenAI’s large language model pulls copyrighted text directly out of her and other authors’ works and re-produces it in its response. Additionally, Silverman’s lawsuit claims the large language models produce “derivative works” of the original creators work without permission and in violation of their rights under the Copyright Act. Silverman’s allegations against Meta are substantially similar in kind but cites Meta’s own large language model, LLaMA, that Meta uses to train chatbots.
OpenAI and Meta have both asserted the affirmative defense of fair use. As this case develops, it will have a profound effect on the use of artificial intelligence and the protections of copyright law, particularly the fair use and derivative works doctrine.
John Grisham, George R. R. Martin, et. al. v. OpenAI
On September 19, 2023, John Grisham (A Time to Kill and The Firm) and George Martin (A Game of Thrones) filed a class action lawsuit alleging OpenAI committed copyright infringement by using their copyrighted works to feed and train the OpenAI large language models, without consent of the copyright owners. The lawsuit is substantially similar to Silverman’s and is currently working its way through the court system.
U.S. Copyright Office and Thaler v. Perlmutter, Director of the U.S. Copyright Office
Recently, Judge Beryl A. Howell of the United States District Court for the District of Columbia, ruled that artificial intelligence created works cannot be copyrighted because of the lack of human authorship. In this case, Thaler used artificial intelligence to create a piece of visual art much like a human paints an original work of art. Thaler tried to register the work of art with the U.S. Copyright Office. The Copyright Office denied the application on the basis that the work “lack[ed] the human authorship necessary to support a copyright claim” in line with current case law that copyright law only extends to works created by human beings. Judge Howell cited current copyright law in the order stating that “human authorship is a bedrock requirement of copyright.” Thus, he reasoned, the originator of the art must be a human being in order to claim copyright protection.
U.S. Copyright Office and Jason Allen
In a related case, the U.S. Copyright Office denied a claim by artist Jason M. Allen, who used an AI system called Midjourney to create a digital artwork. Midjourney is a tool that helps artists enhance their sketches, photos, or drawings with various filters, effects, and transformations. It also gives feedback and suggestions to improve the artworks’ quality and originality. Allen’s artwork, which he made by entering text prompts into Midjourney, won the 2022 Colorado State Fair contest in the digital art category. The U.S. Copyright Review Board (Board), ruled that his artwork did not qualify for copyright protection because it was not the result of human authorship. The Board said that Allen’s only human contribution was the text prompts, which were not enough to show his creative expression. The Office followed its previous decisions and court rulings that require human involvement for copyright registration and protection. The Office also said that it would accept artworks made with AI if they showed sufficient human authorship, but not when the human input was minimal.
This is not the end of litigation concerning copyright protection for works of art generated in whole or in part by generative AI. Judge Howell and the Board simply upheld the status quo by affirming current case law and upholding the Copyright Office’s administrative position that a work generated exclusively by a generative AI tool is not protected under the Copyright Act.
The intersection of artificial intelligence and copyright law has raised important questions and challenges for the future of creativity and innovation. For now, the Thaler decision upheld current case law and the Copyright Office’s position regarding protection of artwork on this very important topic. More questions arise regarding the use of copyrighted works to train large language models. Writers and artists frequently admit that they draw explicitly or implicitly from past works that have inspired them to create, with or without permission of the original creator. Generative AI relies on training data from copyrighted works without the consent of the copyright holder to produce outputs. Can these be reconcilable? Is the fair use doctrine applicable to the output generated by these large language models or is the output a derivative work? These are complex questions surrounding copyright protections and ownership of AI-generated content. The legal landscape is still evolving and uncertain, yet exciting and interesting!