November 28, 2023 - ChatGPT and other artificial intelligence ("AI") programs, like the image-generating Midjourney, have accelerated conversations around generative AI, raising important public policy questions on the protectability of AI-generated works.
The full impact of generative AI has not yet revealed itself, and all creative fields are essentially confronted with a future that is hard to predict amidst the fast-paced changes taking place. However, the U.S. Copyright Office has taken some initial steps to issue guidance, even launching a formal AI initiative. While it has fallen short of issuing any rules, the Office has left a breadcrumb trail that forms the center of this debate: What might satisfy the Copyright Office's requisite human authorship requirement to afford AI-assisted creative content copyright protection?
Where the debate stands, and why the issue matters for all creative fields
Many creatives consider AI to be "plagiarism machines" particularly regarding the use of their copyrighted work as references by the AI apps without their permission. Generative AI formulates its output by drawing on patterns learned from pre-existing works. Some might argue this is no different than an artist being inspired by prior works, but the plaintiffs in recent cases argue that generative AI, by design, directly uses existing works to feed and train its algorithms. The recently resolved strikes by the Writers Guild of America ("WGA") and the Screen Actors Guild ("SAG-AFTRA"), have highlighted the issue of how to govern the use of generative AI in creative fields.
In the WGA's negotiations with the Alliance of Motion Picture and Television Producers ("AMPTP"), the WGA demanded that AI be prohibited from writing or rewriting material for production or being used for source material, and that material created under the Minimum Basic Agreement not be used to train AI.
However, the AMPTP, and other funders of creative works did not want to close the door on a fast-growing technology already proving useful. For example, Disney has been experimenting with AI technology since June 2020 to create characters for its Star Wars franchise. The resolution between the parties included many of these demands, including the requirement that AI-generated material will not be considered "source material"; and a compromise that the WGA reserved the right to assert claims that a writer's material was used to train an AI model. See "Why AI Is the Most Important Issue in the Writers' Strike," The Writing Cooperative, May 25, 2023.
The debate at issue in the strikes may become moot as new rules make their way through the Copyright Office or even federal regulation. However, Congress may be concerned with over-regulating the development and application of AI technology, as the U.S. is always in competition with other nations to develop new technologies and markets.
Copyright office guidance and rulings offer indications of the office's direction
No copyright protection for works lacking requisite human authorship
The Copyright Office has been deciding AI-related applications and issuing guidance for years. In 2022, Stephen Thaler attempted to copyright a two-dimensional AI-authored image titled "A Recent Entrance to Paradise," created by an image-generating AI tool he developed called the "Creativity Machine." The Copyright Office rejected the application, stating that the work lacked "the human authorship necessary to support a copyright claim."
In response, Thaler sued the Copyright Office, arguing that he is the owner of the work and that it was a work made for hire by the AI tool. On March 16, 2023, the Copyright Office issued "Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence," a Statement of Policy, "to clarify its practices for examining and registering works that contain material generated by the use of artificial intelligence technology."
Headlines that followed largely stated that AI-created works would not be protectable. While this is true for works created exclusively by AI, what has been largely ignored is how a work co-authored by a human and a generative AI can garner protection under the right circumstances.
In the March Statement of Policy, the Copyright Office sought to offer guidance on two questions contested by Thaler and others utilizing generative AI in the creation of expressive works, addressing whether (1) a generative AI can qualify as an author, and (2) whether a work produced by a generative AI can be copyrighted as a work made for hire by the person that input the prompts.
Unfortunately for AI proponents, the answer to both questions was a simple "No," as: "In the Office's view, it is well-established that copyright can protect only material that is the product of human creativity." As such, where an AI only receives a prompt, the resulting expressive work is not copyrightable, because the human user does not have sufficient creative control over the ultimate output.
How much human involvement is enough to secure a copyright
The March Statement of Policy also addressed whether a work that consists "of both human-authored and AI-generated material" can be registered and what information an applicant must provide to the Copyright Office to register such works. It states that in cases where AI-generated work is subsequently edited or manipulated by a human in a "sufficiently creative" manner, the resulting work constitutes an original work that can be copyrighted. So, although the underlying AI-generated material is not protected, the manipulated work can be.
Additionally, the Statement left open the possibility of copyright protection for creative works where the user-prompts result in an expressive work that is more predictive. The implications are apparent for certain creative endeavors: In the future, will there remain a need for an entire writers' room? Or, can the initial drafts be created by generative AI and then edited by one or more humans?
Additional clues as to how a human can collaborate with an AI on a copyrighted protected work are found in the story of Maria Kashtanova and her graphic novel "Zaraya of the Dawn." The case was the primary subject of a Feb. 21, 2023, memo issued by the Copyright Office, which explained how significant human editing of an AI-generated work can satisfy the human authorship requirement to receive copyright protection.
Kashtanova's copyright application garnered sensationalist, misleading, and imprecise headlines like "Copyright Office Denies Claim to Copyright in Generative AI Images;" however, although the ruling denied copyright protection to the individual AI-generated images, the graphic novel, in its entirety, was deemed copyrightable:
Based on the representation that the selection and arrangement of the images in the Work was done entirely by Ms. Kashtanova, the Office concludes that it is the product of human authorship. Copyright therefore protects Ms. Kashtanova's authorship of the overall selection, coordination, and arrangement of the text and visual elements that make up the Work.
For one last example, an April 20, 2023, letter from the Copyright Office to The Mechanical Licensing Collective offered further guidance regarding mechanical music licenses. The letter, which largely parallels the guidance issued in the March Statement of Policy, stated that if a musical work is produced solely by generative AI with the use of prompts, it cannot receive a copyright, as it fails to have the requisite human authorship. However, the Copyright Office stated that where a human selects or arranges a musical work in a "sufficiently creative way," the resulting work will be an original work that satisfies human authorship.
The future for creatives and copyright
The apocalyptic rhetoric regarding how generative AI will be the death knell for all human-created expressive works, including journalism, media, art, and entertainment, may be hyperbolic in the short term, but there is no doubt that generative AI will displace a certain degree of human creativity and involvement. It is not far-fetched to envision a future where the first draft of an expressive work produced by AI is then edited by humans. Or, conversely, to have human-authored works then edited or manipulated by AI remain protectable works.
Even with a clearer picture of what constitutes the requisite amount of human authorship to obtain copyright protection for AI-assisted expressive works, the issue remains of whether and how content creators should be compensated if their works are used to train AI technologies, as seen in class-action suits filed by Sarah Silverman and others against OpenAI and Meta, as well as an open letter to AI companies from over 8,000 authors demanding compensation for any use of their works to train AI models without permission.
In a June 12, 2023, letter to Stephanie Weiner, Acting Chief Counsel of the National Telecommunications and Information Administration, the Copyright Office stated that it is preparing to take on the issue of whether the use of existing works to train AI models constitutes a fair use. The Office plans to release a notice of inquiry seeking public input later this year. Some other countries (such as Japan and Israel) have already taken steps to codify a fair use exception for model training to create safe havens for innovation.
The battles waged today will have profound implications on creativity, ownership, and control. The business, technology, legal, and creative communities must stay current with the latest trends in order to address such issues in their policies, procedures, and contracts.