It is often said that some of the most successful among us are those who think creatively and communicate artfully. The adage holds true for artists and other content creators. As the influence of AI in artistic creation continues to increase, a litany of unique legal issues have arisen that require intellectual property attorneys to meet these new challenges.
As AI has reached a new level of maturity, a litany of AI tools for facilitating the generation of creative content have flooded the market. The new AI tools are empowering expressive people to create a wide variety of digital art and other creative works including images, videos, and music.
Often, these artworks are not based on intrinsic artistic talent in a particular media, but based on the ability of the creator to work with the AI tool in the form of inputting prompts or other information to the tool that ultimately determines the output generated by the tool.
While information-based tools such as ChatGPT serve up answers to questions and prompts nearly instantaneously, new character AI products are capable of constituting digitized fictional personas that interact with humans for conversational and pseudo-social purposes.
Other AI products have enabled the generation of art works, photographic quality images, videos and even music. Generative AI has expanded to the music industry in the form of Suno, a Microsoft-financed creation available through the software giant’s Copilot branch.
By taking a deep dive into Suno, and entering creative prompts, you might hear slight or even strong resemblances to popular artists. Though generative AI is trained on existing content made by human beings, the output is arguably “original.”
However, there is vigorous and ongoing ethical and legal debate as to whether the output of generative AI violates intellectual property law.
Generative AI Copyrightability
U.S. Court rulings have held that the content generated by artificial intelligence cannot be copyrighted as AI itself cannot be considered an “author” within the meaning of the Copyright law.
Judges concur that the output of generative AI is derivative, meaning spawning from an original source of creation as opposed to being unique and distinct on its own, although inquiring minds may wonder whether AI, now or in the near future, might be capable of generating original work products without a base of original sources.
However, for now, the output of generative AI is considered to constitute a form of derivative authorship known under the law as a “derivative work.” In the human world, derivative authors use portions of another’s work and provide their own unique contribution, while AI essentially creates spinoff content through a technological adaptation process.
Derivative works involve the alteration of an initial creative work. It is possible to obtain Copyright registration for the authorship of a derivative work based on an altered original work if the derivative work has what is considered to be a sufficient level of originality created through human thought.
However, AI generated outputs are not considered derivative works in that they are considered as not involving human thought in their creation, despite the thought that went into the AI algorithm or any training of the AI that may have occurred.
As an example, consider the United States Copyright Office’s decision to refuse authorship of Ankit Sahni’s AI-generated art. Sahni created an AI tool dubbed “RAGHAV” with the help of software developers.
Using a neural network, which is modeled/named for the action of the human nervous system’s biological neurons that convert signals into outputs, the RAGHAV AI tool transforms inputs into artistic renditions. It is noteworthy that
Sahni converted a picture from his personal camera into a modified image using the art style of Van Gogh’s famous Starry Night painting. That Sahni’s AI tool relied on training from unique paint data sets to create the new image precluded human ownership and also precluded use for commercial purposes.
And the notion that Sahni’s prompting or input could be considered human “thought” or creative input was unpersuasive. Specifically, the U.S. Copyright Office determined the RAGHAV AI tool achieved the rendering based on its machine training as opposed to Sahni’s unique combination of input prompts.
Upon a second request for reconsideration, the Copyright Review Board affirmed the denial of registration as can be seen in the response published at the U.S. Copyright Office webpage Microsoft Word – 2023-12-11_SURYAST Review Board Decision Letter_final (copyright.gov). U.S. Copyright Law was deemed not applicable to protecting Sahni’s computer-enhanced generation.
It is worth noting however that RAGHAV AI was listed as a co-author of a work successfully registered in November 2020 in the Indian Copyright Office (ROC No A-135120.2020) and in December 2021 in the Canadian Copyright Office (Reg No 1188619) as discussed in an article published in May of 2022 in the University of British Columbia’s Allard School of Law, Intellectual Property Law 422 course blog: RAGHAV: First (Registered) AI Author | IP Law 422 001 (ubc.ca).
The Issue of Legal (Infringement) Liability for AI Output Generation
The United States Copyright Office has affirmed four consecutive refusals for the registration of generative AI outputs. However, courts have been hesitant to impose liability on the developers of AI as they are merely training computers to provide feedback based on existing material.
Until plaintiffs begin highlighting specific content and technical idiosyncrasies supposedly used to generate artificial creations, AI developers will likely continue to sidestep legal liability.
As an example, if a musician were to take legal action against Suno, claiming its outputs were similar to pop songs already in existence, Suno would likely prevail. The makers of Suno trained the tool using a wide variety of music styles, yet the outputs constitute completely original instrumentals and vocals.
Suno users can enter millions of prompts and never replicate actual songs. Suno is designed to generate music through artificial intelligence dependent on the subtleties of prompts, user interactions and idiosyncratic scenarios.
It is tantalizing to consider whether an artist’s recognizable sound could somehow be protected if a generated AI output created a senseon the part of an audience of “recognition” of the artist’s touches in a work that was generated using the artist’s work as an input.
A similar issue arises in original audio generated from AI that used, for example, a speaker’s actual voice as an input and generated new vocal utterances. These audio (and to a limited extent video) works are now known as “deep fakes.”
Deep fakes can be distinguished somewhat from generative outputs that use a collection of inputs to generate a unique output in that a deep fake is specifically intended to mimic or “clone” the voice of the original speaker for the purpose of generating original (new) utterances/output.
However, technology is under development to combat or at least recognize whether utterances are genuine or cloned: New Research Combats Burgeoning Threat of Deepfake Audio | UC Berkeley School of Information.
At the same time, there is a push, especially among performance artists, for copyright protection involving deep fake “clones” of their voices. Some artists have accused popular social media platforms such as TikTok of using deep fake renderings in their content: Performing artists push for copyright protection from AI deepfakes | Reuters.
The Future of the AI Legal Landscape
At present, at least in the United States, the courts and Copyright Office have ruled in favor of original content creators engaged in legal battles against the users of generative AI based on generative AI outputs.
However, in the long run, as the technology and the legal framework evolves, we believe there is the potential for a growing acceptance of copyright protection, even if in a limited form as a special class of protection, for generative AI works.
Such special protection could be along the lines of Mask Work Rights, under 17 U.S.C. Section 901, which provides protection for computer generated three-dimensional circuit topographies.
A legal precedent might eventually be set that states the use of a copyrighted work as the origin of an iteration in a generative AI work is no longer protected, or at least that the resulting work involving extensive modification, especially modification prompted by actual people as opposed to machines would be afforded protection.
On the other hand, things could turn a different direction in which the use of an original copyrighted work as an input to a generative AI platform could constitute an instance of infringement that would not be eligible for any affirmative defense of fair use .
Topics to Watch for in the Near Future within the AI Legal Landscape:
Acceptance of Generative AI Work for Copyright: As mentioned, there is the potential for a shift in the legal landscape where generative AI work may be accepted for copyright protection. Such a shift would involve future legal battles and court rulings revolving around the copyrightability of AI-generated content.- Extent of Modification: The text highlights the question of how much modification is required for AI-generated content to be considered distinct from the original and thus not infringing of the original work under the law. The specifics of how and under what thresholds courts interpret generative AI originality could be a point of contention in future cases.
- Modification Prompted by Humans: The distinction between modifications generated by AI algorithms themselves and those based on prompts provided by human intervention is likely to be a significant factor in future legal discussions, including whether the human prompting itself can be a copyrightable work. Determining the level of human influence required to afford copyright protection to the generated work could become a key issue.
- Legal Precedent: Ongoing and upcoming cases can and will shape the future of AI-related intellectual property law and will set the short and medium term legal precedent. The influence of these cases on the legal landscape will be closely watched to determine how practices will be modified or what new procedures and rules must be followed.
- User Intellectual Property Protection: Court rulings in favor of original content creators will create legal challenges for AI tool users that are attempting to claim intellectual property protection on generated works and/or prompts. How these cases are resolved will significantly impact future behavior in the generative AI space and will be worth monitoring closely.
- Lawsuits Involving AI Tools and Copyright Infringement: Lawsuits involving AI tools like Stable Diffusion and ChatGPT and the alleged infringement of the output of these tools on intellectual property rights are ongoing. The outcomes of these cases, as well as any similar cases that arise, will have significant implications for AI developers and users.
- Impact of AI Crawling the Web: OpenAI developed technology that “crawls” the web and “scrapes” website content for AI training purposes. Open AI’s web crawling/scraping led to a large, wide-ranging class-action lawsuit in the Northern District of California (P.M. et al v. OpenAI LP et al., 3:23cv03199). In view of this landmark suit, it becomes clear that the use of AI tools to scrape web site content may face significant legal challenges, the outcomes of which will set important and lasting precedents.
The near future of the AI legal landscape may involve discussions and legal proceedings related to the copyrightability of AI-generated content, the extent of human modification required for copyright protection, and the outcomes of ongoing lawsuits involving AI tools and intellectual property.
These are topics to watch for developments and potential changes in the legal framework. Stay tuned. The field of generative AI law will continue to take shape in the months and years ahead.
A Brief Summary of Relevant AI Intellectual Property Proceedings
To date, few intellectual property legal challenges have been made against AI companies as the field is emergent and dynamic. Though courts have ruled that the use of AI tools to manipulate artwork and other content does not culminate in user intellectual property protection, legal challenges extend beyond the users of AI tools.
The number of lawsuits filed in response to AI’s use of existing data will likely increase in the near future and also across posterity.
Getty Images (US), Inc. v. Stability AI, Inc., 1:23-cv-00135-JLH
Getty Images sued Stability AI, the developers of an open-source artificial intelligence art generation tool dubbed Stable Diffusion. The suit alleges Stability AI infringed on Getty Images’ intellectual property with the use of 12 million images, captions and metadata.
Moreover, Getty Images also filed a separate suit against Stability AI in which it alleges the company’s text-to-image tools use copyrighted content without permission.
Tremblay v. OpenAI, Inc. (3:23-cv-03223)
As noted above, OpenAI has been subjected to at least two class actions including the present class action case in which Authors Paul Tremblay and Mona Awad sued OpenAI, insinuating their copyrighted books were used for AI training purposes.
The complaint states the authors’ books were used for training OpenAI’s popular AI-powered chatbot ChatGPT. The suit alleges ChatGPT reproduced accurate summaries of the books, constituting copyright violation. OpenAI states its technology “crawls” the web for archived information including information published on Wikipedia.
The New York Times Company v. Microsoft Corporation (1:23-cv-11195)
The New York Times lawsuit against Microsoft and OpenAI insists its ownership of exclusive rights for reproduction, publication, display, and performance in accordance with the Copyright Act 17 U.S.C. § 101 et seq. precludes Microsoft and OpenAI from producing what it alleges to be near-verbatim copies of copyrighted material.
The complaint also alleges Microsoft’s Bing Chat and OpenAI’s ChatGPT provide users with access to copyrighted content only available to those who pay for access.
Authors Guild v. Google 721 F.3d 132 (2nd Cir. 2015)
Authors filed a lawsuit against Google after the tech giant attempted to digitize nonfiction and fiction books through an ambitious scanning process.
The authors alleged Google failed to seek permission to scan published content, ultimately violating copyright laws. The court determined Google digitization of copyright-protected content constitutes a non-infringement of fair use as the presentation of text was limited.
Closing Thoughts
In the ever-evolving landscape of generative AI and its impact on the legal industry, one thing remains clear: the boundary between human creativity and artificial intelligence-generated content will continue to challenge traditional notions of intellectual property.
While the courts and the U. S. Copyright Office have consistently ruled in favor of original content creators in their battles against AI users, the potential for a shift in this paradigm looms on the horizon.
The intersection of AI and intellectual property law presents both challenges and opportunities, and the legal community must adapt to effectively address these complexities.
As the legal system grapples with defining the boundaries of copyright in the age of AI, we will undoubtedly witness further legal precedents, legislative changes, and discussions that will shape the future of generative AI and its place within the legal industry.
Protect Your IP and Patents with METROLEX IP LAW GROUP
METROLEX IP LAW GROUP is an international full service Patent and Intellectual Property law firm with offices in Washington D.C. and Osaka. We have specific experience in protecting AI related inventions and dealing with AI-related issues.
We focus on high-quality work products, and rapid, efficient filing and prosecution across the world, including U.S., Japan, Europe and China.
