Generative Artificial Intelligence and Copyright Current Issues

The recent rise in popularity of generative artificial intelligence–powered applications such as ChatGPT poses important copyright issues for individuals and businesses with respect to content creation, including the scope of rights with respect to commercial use, content publication, potential liability for infringement, and content enforcement.

AFTER YEARS OF DEVELOPMENT AND PROMISE, AN OVERNIGHT ARTIFICIAL INTELLIGENCE HIT

Recently, numerous applications powered by artificial intelligence (AI) have gained immense popularity with content creators around the world who are learning to leverage the power of generative AI. AI-powered models like ChatGPT, Dall-E, and Midjourney, among others, are in the news daily and many people have already become both dependent on, and amazed by, their capabilities. These applications are revolutionizing the ease in which many types of content can be created quickly, including website copy, marketing campaigns, social media posts, lists of all kinds, poetry, and even software code.

One of the more popular AI tools, ChatGPT, is a chatbot created by OpenAI in San Francisco and is capable of generating human-like text in response to any given prompt based on deep learning techniques, which involve training the AI model on a large dataset of text to predict patterns and relationships of words and phrases. In fact, the opening paragraph above started with a solid draft from ChatGPT, which we prompted to describe what ChatGPT is and is capable of doing. After a bit of editing, we had what we wanted.

ChatGPT’s free language model, GPT-3, was released in June 2020. However, OpenAI just unveiled its fourth iteration of the software, GPT-4. [1] OpenAI’s website [2] explains that GPT-4 is “more creative and collaborative” and even allows the AI to “generate, edit, and iterate with users on creative and technical writing tasks, such as composing songs, writing screenplays, or learning a user’s writing style.” The ABAJournal recently published an article titled, “Latest version of ChatGPT aces bar exam with score nearing 90th percentile.” [3] While ChatGPT is limited to generating text responses, GPT-4 now allows users to provide images as input and can generate comments or even code based on the image provided. Other OpenAI applications such as Dall-E and Midjourney offer text-to-image models.

THE FINE PRINT

Like most applications, terms of use are the starting point to understand the scope of rights granted to the user. Contractual provisions for the use of ChatGPT address the following:

• Content Ownership: “Content” is defined in OpenAI’s Terms of Use (Terms) for ChatGPT as anything input to or output generated by the service. With respect to ownership, the Terms state: “[a]s between the parties and to the extent permitted by applicable law, you own all Input, and subject to your compliance with these Terms, OpenAI hereby assigns to you all its right, title and interest in and to Output.” OpenAI retains limited rights to use content for the purpose of providing and maintaining the service, complying with applicable law, and to enforce OpenAI’s policies. [4]
• Similar Outputs: The Terms warn that output will not always be unique or proprietary to a single user:
  • “Due to the nature of machine learning, Output may not be unique across users and the Services may generate the same or similar output for OpenAI or a third party. For example, you may provide input to a model such as “What color is the sky?” and receive output such as “The sky is blue.” Other users may also ask similar questions and receive the same response. Responses that are requested by and generated for other users are not considered your Content.”
• Confidentiality: The Content input to ChatGPT is a disclosure, and as explained in the Term’s ownership provision, it may be used to train its models, and some amount of input Content may become part of ChatGPT’s responses to others. The lack of confidentiality associated with using ChatGPT is a major reason why companies should be cautious about its use by employees.
• Publication Requirements: OpenAI’s Publication Policy [5] directly responds to questions about further publication of output: A company that wishes to publish “written content (e.g., a book, compendium of short stories) created in part with the ChatGPT are permitted to do so under the following conditions: (1) the published content is attributed to the company, (2) the AI’s role in creating the content is clearly disclosed (3) content topics do not violate OpenAI’s Content Policy [6] or Terms of Use, (4) [OpenAI] kindly ask that you refrain from sharing outputs that may offend others.”

OpenAI further explains that “the role of AI [must be] clearly disclosed in a way that no reader could possibly miss, and that a typical reader would find sufficiently easy to understand.” The Terms provide a sample disclosure:

For instance, one must detail in a Foreword or Introduction (or some place similar) the relative roles of drafting, editing, etc. People should not represent API-generated content as being wholly generated by a human or wholly generated by an AI, and it is a human who must take ultimate responsibility for the content being published.

ChatGPT even provides sample language to describe a typical creative process:

The author generated this text in part with GPT-3, OpenAI’s large-scale language-generation model. Upon generating draft language, the author reviewed, edited, and revised the language to their own liking and takes ultimate responsibility for the content of this publication. [7]

POTENTIAL RISKS OF INFRINGEMENT

Although OpenAI is not likely to challenge a company’s use of content generated by ChatGPT (unless a user violates the Terms), that does not mean use of such content is without risk of copyright-related challenges.

Potential liability for copyright infringement still exists to the extent that the AI models rely on existing content to generate content. Both artists and businesses have asserted claims in federal court concerning the use of copyrighted images for training AI models and creating resulting images that are alleged to infringe their copyrights.

For example, in a class action comprised of “at least thousands” of other creators, Midjourney, Stability AI and DeviantArt are being sued for copyright infringement, violations of publicity rights, and unfair competition. [8] Getty Images has also filed suit against Stability AI both in the United States and separately in the United Kingdom based on Stability AI’s alleged use of more than 12 million images from Getty’s database for machine learning purposes.[9] AI software code generating tools have been similarly challenged. [10]

In these cases, defendants have claimed that the use of copyrighted material in this manner constitutes fair use. These cases are still in an early stage, but whether ingesting copyrightable material for machine learning purposes is an infringement or a fair use may ultimately need to be decided by the US Supreme Court. Accordingly, it is possible that OpenAI could be sued based on its use/ingestion of third party content to train ChatGPT, and depending on the output in any given situation, it is also possible that specific users could be sued if the content generated by ChatGPT (and later used by a specific user) is too similar to a third party’s content.

PROTECTING GENERATIVE AI OUTPUTS

As to whether and to what extent companies can actually protect content created using ChatGPT and other similar AI tools, the recent US Copyright Office decision regarding the copyrightability of an AI-generated graphic novel, titled “Zarya of the Dawn,” is instructive.

“Zarya of the Dawn” is a comic book with original text authored by a human but comprised of individual images created by Midjourney, a prompt-based AI program for image generation. The images generated were then selected, arranged, and partly edited by the human author. The Copyright Office originally granted the registration of the work but––upon learning that AI-generated images were used––limited the scope of the copyright to cover only the original text and final product in its entirety as a compilation, while revoking the copyright protection for the individual AI-generated images.

The Copyright Office requires human authorship for creative works to be deemed copyrightable. In cases such as “Zarya of the Dawn” where AI is used, the Copyright Office will look closely at the extent to which a human, and non-human, contributed. While the author argued that each image was the result of a creative process in which a human guided the AI iteratively through hundreds of variations of intermediate images until a desired result was achieved, the Copyright Office explained that a person who provides text prompts to an AI machine “does not actually form the generated images and is not the ‘master mind’ behind them.” In other words, “[t]he information in the prompt may ‘influence’ the generated image, but prompt text does not dictate a specific result.”

The Copyright Office’s decision further explained the issue of human control over generative AI in terms of the comparative control a photographer has over a camera. The Office notes that, unlike a photographer, AI models lack the same human level of control over the final images produced, including the photograph’s framing, lighting, exposure, depth, etc. The author of “Zarya of the Dawn,” as explained in the Copyright Office’s decision, does not dictate how the subject or the subject’s presentation appears, and cannot control what the AI creates.

The Copyright Office stated, “While additional prompts applied to one of these initial images can influence the subsequent images, the process is not controlled by the user because it is not possible to predict what Midjourney will create ahead of time.” Moreover, the Copyright Office noted that the author’s editing of the images was “too minor and imperceptible” to satisfy the necessary creativity for copyright protection.

The Copyright Office recently issued a policy statement providing additional guidance on whether and to what extent artistic works created with the help of generative AI are eligible for copyright protections. [11] In direct response to its decision regarding “Zarya of the Dawn,” the Office clarified that moving forward, “the Office will consider whether the AI contributions are the result of ‘mechanical reproduction’ or an author’s ‘own original mental conception, to which [the author] gave visible form.’” More specifically, “when an AI technology determines the expressive elements of its output, the generated material is not the product of human authorship.”

The Office’s guidance specifically noted that, “[b]ased on the Office’s understanding of the generative AI technologies currently available,” users currently lack the requisite creative control. Whether GPT-4 or other AI tools going forward will provide users with sufficient creative control over the AI’s output for that output to be deemed registrable at the Copyright Office is yet to be determined. However, unlike traditional issues related to joint creation of works, in the case of AI output, there will be a perfect written record of the creative process. Will AI service providers be subpoenaed someday for chat records to prove or disprove sufficient human authorship?

CONCLUSION

Given their simplicity of use and, in many cases, impressive output, ChatGPT and similar AI services are likely here to stay. As generative AI platforms expand and evolve, it will be important for businesses to think proactively about how these services should and should not be used in the workplace. Morgan Lewis stands ready to help readers develop AI content use policies appropriate for their businesses.