Creative works generated by artificial intelligence (“AI”) have become increasingly popularized in recent months. Popular AI art generators, such as DALL-E 2, Lensa, and Stable Diffusion, are programs that allow an end user to express otherwise basic parameter inputs, such as genre, style, and content, and then direct the AI software to generate a new creative work based on those inputs.
In recent months, the topic of AI-generated artwork has raised some important legal questions; the most interesting and relevant to our work are questions related to the rights of copyright owners. For instance, AI generally becomes functional through training datasets curated in such a way that the computer system can “learn from” and apply that knowledge to new inputs and queries.
Where AI is used to generate artwork, it raises a significant question as to what data was sourced and used in training the AI software, and importantly, whether the owner of that data authorized the use of it for that purpose.
AI-generated works may occasionally contain elements of an artist’s “signature” or known style, suggesting that the AI tools were likely trained on works available on the internet, and it’s ultimately unclear from a user perspective whether or to what extent the creators of the sourced artwork gave permission to use their works as training data.
Indeed, Lensa from Stable Diffusion has run into this issue recently, and a core question concerns whether the artwork used to train the AI is potentially infringing upon the original artists’ copyrights.
One such example of the issues we consider in this article is arising out of a high-stakes matter currently before the Southern District of New York between The New York Times (“NYT”) and OpenAI, the creators of ChatGPT. In that case, NYT is alleging that OpenAI, and its investors and subsidiaries, leveraged the digital content created by NYT to train its large-language model (“LLM”) that is deployed in the ChatGPT product.
Importantly, NYT is alleging that OpenAI did this without a license to use NYT’s content for such purposes, and that given certain parameters from a user, ChatGPT “will output near-verbatim copies of significant portions of Times Works . . .” or at least relies heavily on NYT data in its LLM to generate its output.
It is anticipated that OpenAI will put forth a “fair use” defense and argue that its use of the NYT content is sufficiently “transformative” to escape financial liability.
Whether that defense will be successful remains to be seen from a legal standpoint, but it is clear that the outcome of this case has the potential to significantly impact the underlying technology, and may depend, in a material way, on experts’ ability to evaluate the technical components and uncover relevant facts.
Under the Copyright Act, copyright owners enjoy the exclusive right to create derivative works (17 U.S.C. §106 (2). Derivative works are works that are based on one or more existing works1. Common examples of derivative works include translations of literary works from one language to another, adaptations of existing works into new media, such as films or screenplays based on novels, or elaborations upon existing works.
In such examples, the new derivative work is its own original work based upon the existing work that has been modified or subsumed to create a new expression. Generally, the copyright owner enjoys the exclusive right to create such derivative works or permit their creation by someone else.
However, copyright law also provides an exception to this right that otherwise permits third parties to create derivative works without permission of the copyright owner: fair use.
Guided largely by public policy considerations, fair use doctrine expressly permits use of copyrighted material without permission from an owner, for such purposes as, for example, news reporting, commenting/criticism, or scholarship (17 U.S.C §107). Certain types of use, such as parody or satire, have been accepted as fair use for some time, such as the song “Pretty Woman” by 2 Live Crew 2, or much of the music created by Weird Al Yankovic.
A common inquiry among fair use analysis is whether the derivative work is transformative; whether the derivative uses the original work for a new purpose or for a different manner. Indeed, this was a key consideration in the Supreme Court’s decision of Google’s use of Oracle’s Java code in the development of its Android platform 3.
Thus, derivative works that sufficiently transform the original work for a new purpose or different manner, may be considered fair use.
Similar questions arise in consideration of AI-generated works. For example, one immediate question is whether works that are generated from AI can be considered derivative works under the copyright law.
There is likely an argument to be made that the AI-generated artwork is based on, in some capacity, the preexisting works used to build the training data. However, there is likely also an argument to be made that the deployment of AI in that generation of a derivative work may be sufficiently transformative to constitute fair use.
Additionally, using AI to generate creative works presents a question of authorship and ownership of the resulting work. Is the AI software being wielded like a camera or a paintbrush to render the end-user’s vision, or, is the AI actually “creating” as a pseudo-author in place of a human? Has the training data artwork been appropriately licensed from the original creator?
Questions of authorship and fair use are not novel considerations under the Copyright Act; however, rapidly evolving technology like artificial intelligence invites considering these issues with a fresh perspective.
AI source-code-generation tools are also starting to be used in the creation of new software works. DisputeSoft’s copyright infringement experts have years of experience in analyzing and comparing between software works for evidence of copying to support or refute alleged infringement upon intellectual property rights, such as for Copyrights or trade secrets. We anticipate that soon we will soon be called upon to analyze similar issues related to code generated with AI tools, such as the recently-popular ChatGPT.
DisputeSoft’s software experts closely monitor evolving software technologies and how they fit in with established legal frameworks as determined by the courts. We find it likely that the source code for such AI-art-and-code-generation tools will soon be subjected to expert examination in litigation to aid a trier of fact in determining whether a work, or the whole system, is creating improper derivative works, or rather, transformative new works, or if each generated work should be evaluated on a case-by-case basis.
The US Copyright office has updated its guidance as of March 16, 2023, to state that AI-generated works that are purely generated from user text input, where all generation and creative expression is performed by the AI, that such a work will not be eligible for US Copyright protection.
The new guidance does not speak to the creative expression used as the input text to provide to the AI for generation of a new work. Similarly, the guidance appears to treat the AI-generated work from input text similar to a work made for hire, with a special exclusion of eligibility for protection.
The provided rationale for the ineligibility of purely-AI-generated works is that US Copyrights are designed to protect creative creations generated by humans, not other creatures or machines. The new guidance specifically references a selfie photograph taken by a monkey that was ruled ineligible for US Copyright protection in 2014.
It has yet to be decided what impacts incorporating AI-generated works into other larger works will have on those newly created or derived works.
The US Copyright Office has launched an AI Initiative to hear experts and the public weigh in on this new frontier of creative expression from April 19 to May 31, 2023. The full newly-issued guidance from the US Copyright Office is available here.
- A “derivative work” is 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 work consisting of editorial revisions, annotations, elaborations, or other modifications which, as a whole, represent an original work of authorship, is a “derivative work”. 17 U.S.C. §101. ↩︎
- See Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569 (1994) (finding 2-Live Crew’s commercially successful parody of Roy Orbison’s song “Oh, Pretty Woman” could be considered a fair use). ↩︎
- See Google, LLC v. Oracle America, Inc. 593 U.S. _ (2021) (finding Google’s use of Java code in Android sufficiently transformative to constitute fair use). ↩︎
