Appreciable vs De Minimis: A Creator’s Guide to AI and Copyright
By Angelica Ruzanova
Note: This article is meant to be a supplemental educational resource, not official legal advice.
How we manipulate content is not always intuitive in the eyes of the law. Before Midjourney, there was Photoshop, and before Photoshop, there were darkrooms. If we trace history, image enhancement tools go back to Daguerre’s camera obscura that allowed humans to duplicate our world on a sheet of steel.
How, then, has AI changed the creator’s legal right to enhance their work with emerging tools? I watched the Online News Association’s panel on “AI, Copyright and Legal Considerations for Journalists, Freelancers and News Organizations” conducted by associates from Davis Wright Tremaine LLP to answer my questions.
How are copyright law and generative AI interacting in the media industry?
Since the release of ChatGPT, several lawsuits have been filed against AI companies alleging copyright infringement. These cases often involve two types of allegations.
The first is lawsuits challenging how AI models are trained. Filed by authors and creatives, these lawsuits allege that AI companies infringed copyright by using their protected works to train AI models without permission.
For example, Getty Images alleges that nearly 12 million photos from its database, including those that are watermarked, were used to train Stability AI, a text-to-image generation model.
Reference included in the February 2023 complaint filed by Getty Images against Stability AI.
Similarly, The New York Times claimed that OpenAI infringed copyright by using over 3 million of its articles to train ChatGPT. The news organization also argues that ChatGPT has evaded the site’s paywall, negating OpenAI’s fair use.
The second type is lawsuits concerning the output of AI models. This category of lawsuits arises when the output generated by an AI model, in response to a user prompt, appears to infringe on the copyright of existing works.
An example given by the associates from the law firm was a case where users could prompt an AI to create images in the style of a particular artist, resulting in outputs that closely resembled the artist's work.
While this specific functionality has been modified, the potential for AI models to produce infringing outputs remains a legal concern.
What is the US Copyright Office's current guidance on disclosing the use of AI in creative works?
The key principle discussed by the associates at Davis Wright Tremaine was that any "appreciable" use of AI must be disclosed in the copyright application. Failure to do so can result in the rejection of the application, as seen in the case of the graphic novel "Zariah of the Dawn," where the initial registration was revoked after the USCO discovered the artist's use of Midjourney to generate images.
However, the USCO's guidance on what constitutes "appreciable" use of AI is somewhat ambiguous. It clarifies that "de minimis" use, meaning minimal or insignificant AI contribution, does not require disclosure.
To illustrate the distinction, the USCO provides a hypothetical scenario involving an image with a car, a deer and a forest background:
Appreciable use: If AI generated the forest background, this would be deemed "appreciable" and require disclosure, rendering the forest element ineligible for copyright protection.
De Minimis use: If a human created all elements of the image but used AI for minor edits, such as sharpening the background or adjusting the lighting, this would be considered "de minimis" and would not necessitate disclosure.
The legal firm also offered additional examples:
Using ChatGPT to generate an idea that a human author then develops independently would likely be considered "de minimis."
Employing AI for tasks like grammar and citation checking is also likely to fall under "de minimis" use.
However, instructing AI to write an article on a specific topic, even if providing human input and guidance, would likely be considered "appreciable" use.
Entertainment industry guilds, such as the Writers Guild of America (WGA), Directors Guild of America (DGA) and Screen Actors Guild-American Federation of Television and Radio Artists (SAG-AFTRA), are negotiating provisions related to AI in their collective bargaining agreements. These negotiations reflect the concerns of writers, directors and actors about the potential impact of AI on their jobs, while also recognising the potential benefits of using AI in their work.
How can I stay up to date with ongoing developments that may affect my work?
According to Aristotle, our response to objects of vision depends primarily on their arrangement in front of us. In that respect, our reliance on them being true, unsurprisingly, lies in faith.
The existential way forward for a public that has faith in each other is appropriate disclosure of the tools used to generate or modify work. The Davis Wright Tremaine law firm encourages discussing agreed-upon terms with your newsroom, as well as understanding the rights you are granting to the AI company and the potential liability you are assuming.
As some of these issues remain ongoing, it is useful to also be aware that the terms of conditions can change frequently. The Davis Wright Tremaine law firm recommended periodically checking on forums, such as the Intellectual Property Law Server, to stay updated on emerging issues.