July 29, 2024 • Posted by Design Bay Area
Interview with business and intellectual property attorney Owen Seitel, co-founder and Partner at Crown, LLP, San Francisco-based boutique law firm specializing in intellectual property and entertainment matters.
Generative AI has quickly and significantly altered the way we live, work, and create. The proliferation of AI-generated images, music, and text have challenged our fundamental understanding of ownership, fairness, and the very nature of creativity itself. While the artificial intelligence industry continues to rapidly evolve, the U.S. Copyright Office is working on developing a legal framework that incorporates modern AI developments and provides clarity to creatives, as well as companies and courts. Knowing that the rules are yet to be written for much of the legal precedent in this domain, Owen sheds light on the considerations creatives should keep in mind to protect themselves and their work against future unknowns.
Can AI art be copyrighted? Does generative AI violate copyright laws?
Copyright for AI-generated works is evolving with many gray areas yet to be filled in, but one thing is clear – the U.S. Copyright Office requires that works have human authorship to be eligible for copyright protection and registration. Applications for works created by AI without significant human involvement have been rejected repeatedly by the Copyright Office. However, when AI is used as a tool, with a human making the creative decisions, the resulting work is more likely eligible for copyright protection.
- Is a human-generated prompt enough? No.
- How much human involvement is necessary for copyright protection? That will be sorted out by the courts, unless Congress passes legislation in this regard. In the meantime, the Copyright Office has said that AI-generated content that is more than de minimis must be excluded from an application for copyright registration because it is not protectable.
The New York Times and other media and content companies are suing AI providers, claiming copyright infringement due to the unauthorized use of their content by AI crawlers and that the resulting works are derivative of the original material, thus infringing. A generative AI created work could conceivably violate the copyright of a third party work just as any other infringement. As of now, however, there is no certainty and the outcomes of these suits, and the inevitable appeals of these actions, will determine whether AI created content itself or the use of AI to generate content violates copyright law.
If a person uses AI-generated material as a starting point and adds their own creative elements—how much human input would be required before the work qualifies for a copyright?
We know that work created entirely by AI is not eligible for copyright protection and registration because there is no human “author.” But if, say, AI is used as a starting point, how much human intervention is necessary to make the resulting work eligible for copyright protection and registration? Unfortunately, at this time, that question remains unanswered. In the case of Thaler v. Perlmutter, Thaler attempted to register a work created entirely by an autonomous AI and the Copyright Office rejected that application. The court noted that rejecting the application in this scenario was an easy call because there was no human authorship. But how much human intervention is necessary to achieve human authorship? The Copyright Office has stated that if AI is used to generate a work for which copyright registration is sought, the applicant must identify the human-created elements of the work (for which registration may be available) and the AI-created elements of the work (for which no registration or protection would be available).
What is considered fair use?
Fair use is a big topic that has generated tens of thousands of pages and years of heated legal discussion. In general, fair use is a doctrine under U.S. copyright law that allows for the limited use of existing works without the need to acquire permission from the copyright holder. Previous court decisions dictate that use of an existing work is a fair use if the resulting work is “transformative” (significantly differs) of the original work. This is an oversimplification as there are tomes of legal scholarship and much debate concerning what makes a work “transformative” of the original work. The Supreme Court recently addressed this, making an attempt to clarify the issue in Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith, but exactly what makes a work “transformative” continues to be determined on a case-by-case basis and remains uncertain.
The fair use defense is offered in The New York Times’ recently filed copyright infringement case against Open AI and Microsoft in which it alleges that Open AI and Microsoft illegally used “millions” of copyrighted articles to help develop their AI model. Open AI and Microsoft assert that their AI outputs are transformative and thus a fair use. Essentially, Open AI and Microsoft are asserting that it is not an infringement to copy existing works as a preliminary step to develop a new, different, and non-infringing product. Whether the fair use defense is available to AI providers, and the extent of its availability, remains to be seen as the courts sort through these issues.
Should AI-generated outputs be labeled?
In October 2023, President Biden issued an executive order that called for clear labeling of AI-generated content produced by or for the government, in part to help people determine whether communications that appear to be from the government are authentic. Such labeling advises end users that material was generated by AI, but that does not necessarily mean that the AI output is false or inauthentic. And, as noted above, the Copyright Office has stated that a copyright registration applicant has a duty to disclose any content that is generated by AI in the application so that the identified material can be excluded from protection and registration. So, while a general AI labeling requirement communicates how an item of content was produced, labels don’t provide a qualitative judgment about the content or otherwise reliably warn against use of or reliance upon that content. But an informed consumer is a good thing and I, for one, would like to see a broader labeling requirement for content generated in whole or in part by use of AI. I suspect that, in time, we will be surprised at the amount of AI used in the content we consume.
About Owen Seitel
Attorney Owen Seitel is a principal and co-founder of Crown®, LLP, a boutique intellectual property law firm in San Francisco. If your work involves creating, distributing, or securing intellectual property, Crown can help you. Crown clients range from individual artists and producers to large corporations, and they come from myriad industries, from fashion to film, music to marketing, beer to Broadway.