Copyrights and AI: A Solution
(Source)
Copyrights are everywhere. There’s a copyright on the Gideon the Ninth book, the Eras Tour movie, the Thomas the Tank Engine theme song, and so much more. Virtually every book, movie, painting, recording, script, etc. that you can think of has a copyright. But what is a copyright? And how do you get it? According to the US Copyright Office, a copyright is a form of protection on an expression of an idea. With a copyright, you, and only you, have the right to reproduce, sell, distribute, and perform the work. Copyrights are also automatically created. You don’t need to register the copyright with the US Copyright Office to own it. Although you should if you want to ensure legal protection. So you own a copyright on every email and text message you have ever sent. To automatically create a copyright you must satisfy three basic requirements: originality, creativity, and fixation.
Originality means the idea must be original to you. The idea doesn’t have to be something novel, it just can’t be a copy of something else. Creativity only requires a “modicum” of creativity. And fixation means the idea must be expressed in a tangible medium. For example, an idea in your head is not fixed, but once you write it or draw it, it is fixed. A speech you have spoken out loud is not fixed, but once it is recorded, it is fixed. Copyrights not only value ideas, but also the time and effort spent manifesting that idea into reality.
However, there is another requirement that is not often talked about: the author must be human. It was first brought up in Burrow-Giles Lithographic Co. v. Sarony, where the Supreme Court decided that copyrights were the “exclusive right of [humans]” for their creations. This rule has been interpreted to mean that animals, nature, and machines do not satisfy the authorship requirement. So you can’t get a copyright if you train an elephant to paint a mural, find a cool rock in nature, or have a machine convert a work from analog to digital.
But what about using machines as tools? If you take a picture with a camera, did you create the picture, or did the camera? Burrow-Giles Lithographic answered this exact question by deciding that photographs were “representatives of original intellectual conceptions of the author.” This permitted the use of machines as tools, which is why Photoshop is allowed, just like a paintbrush or a chisel. But with the rise of AI, courts have found themselves asking where it falls into the current scheme.
A recent case, Thaler v. Perlmutter, dealt with the issue of AI generated art and authorship. But this case had a clear answer. An AI autonomously generated art without any human input or direction. Therefore, the court found that there was no human author, so the art was not copyrightable. But what happens once you start mixing human and AI contributions into one work? Both the courts and the US Copyright Office have given little to no guidance on this question. To answer this question, there are two big issues we have to resolve: conception and creation.
The first issue deals with how much of the finished product needs to have been conceived by a human author? On one end of the spectrum, works without any AI are conceived entirely by humans and are copyrightable. On the other end of the spectrum, we have cases like Thaler v. Perlmutter, where there was no human conception and the AI generated the work autonomously, so there is no copyright. But what happens if there is a mix of human conception and AI autonomous generation? For example, what if you want to draw a tree, and you can picture every detail, but you don’t have the necessary artistic skills to draw it, so you use AI to create exactly what you imagine? What if you have a general image of a tree and use AI to create a tree, but the specific details are left to the AI to autonomously generate?
The copyright dispute of the comic book Zarya of the Dawn gives us an answer. Here, an artist created a comic book, but used AI to create some of the panels and artwork. The US Copyright Office gave the artist a copyright for the collective work, but not for the individual artworks and panels that were created using AI. They said when an AI “determines the expressive elements of its output, the generated material is not a product of human authorship.” While the US Copyright Office didn’t say it in clear terms, what they mean is that they highly value human creativity. So while only a “modicum” of creativity is required, that creativity must also be entirely human. Going back to the examples, if you can imagine exactly what you want to create and the finished work would look like, then you have satisfied the conception requirement. But if you have a general or rough idea of a finished work, then you have not satisfied the conception requirement.
The second issue deals with how much of the actual creation or “fixation” of the work must be done by a human? Clearly, if the work is fully done by a human author, then it is copyrightable. But what if we use an AI to fix some details? Or what if we use an AI to finish a work based on a rough sketch made by a human? And how is this different from using tools like Photoshop to edit works? The US Copyright Office has given some insight. In the Zarya of the New Dawn letter, they said the key distinction between tools like Photoshop and AI generated works is predictability. They also said that making small edits to an AI generated work is insufficient to satisfy the creativity requirement for copyrights. However, there is no guidance on the flip scenario of using AI to make small changes to a human’s work. But it is clear that copyrights are meant to value the effort and time spent by an artist to create a work. Therefore, AI assistance should only be allowed when the AI’s contribution is both predictable and small in degree.
Together, the human authorship test for AI assisted works should be that (1) the work was entirely conceived by the human author and that (2) the AI’s contributions were (a) predictable and (b) small in degree.
AI, just like Photoshop, can be a useful tool for creators. While AI does pose unique issues compared to traditional tools, this test would remedy those concerns. Furthermore, this test aligns with the traditional authorship test and provides clear guidelines that would be beneficial to courts, the US Copyright Office, and creators.
The only concern with this test, and with all tests in this area, is that we are unable to perfectly determine whether a work was created by a human or an AI. The new cases about AI generated works only arose because people admitted to using AI. But nothing is stopping someone from lying and saying that they created a work without any AI assistance. And issues have already risen for magazines and authors because of this. But we’ll have to let the computer scientists solve that one.
Suggested Citation: Akshat Shah, Copyrights and AI: A Solution, Cornell J.L. & Pub. Pol’y, The Issue Spotter (October 28, 2023), https://live-journal-of-law-and-public-policy.pantheonsite.io/copyrights-and-ai-a-solution/.
Akshat Shah is a second-year law student at Cornell Law School. He graduated from Rensselaer Polytechnic Institute with degrees in computer science and economics. Aside from his involvement with Cornell Law School’s Journal of Public Policy, Akshat is the co-vice president of the South Asian Law Student Association.