Recently, there has been a lot of suggestions that the U.S. Copyright Office is registering “AI-generated works.” Nonetheless, these are not actually AI-generated works, nor are they breakthrough decisions. The U.S. Copyright Office has already registered hundreds of works related to AI-generated material – the key to obtaining registration is to specify AI-generated elements in the Material Excluded field. However, the purpose of this post is to explain copyright protection for authors using (but not limited to) artificial intelligence, rather than registration of works, which is a US peculiarity.
The new report from the U.S. Copyright Office on “copyrightability” aims to clarify when a work created using artificial intelligence can be eligible for copyright protection. The most important part is that “human authors are entitled to copyright in their works of authorship that are perceptible in AI-generated outputs, as well as the creative selection, coordination, or arrangement of material in the outputs, or creative modifications of the outputs.”
What does this mean? In order for something to be protected with copyright, it must be the result of free and creative human choices that are visible in the creation (also called “a work”). So, what is a work in copyright law?
The ultimate definition of a work
One of the most common definitions of a work under copyright law is in the Guide to the Berne Convention, which was written in 1978 by Claude Masouyé (a French diplomat who worked on the Berne Convention and Director of the WIPO). Masouyé did not call it a definition, but (in my view) he gave the best explanation of a work that has ever been written (p. 13):
“True, the genesis of an artistic work (drawing, painting, sculpture, etc.) is rather different from that of the purely literary work. The latter is expressed by its words: the writer conceives the plan of his work and then makes it known; it is this expression which gives rise to copyright. With an artistic work, the plan (mock-up, sketch, etc.) is already, in itself, capable of protection, since from this moment, the idea finds concrete form in lines and colours, with a more personal and direct execution than in the case of writings: the painter makes his own brush marks and the sculptor his statue, whereas it is of no importance whether the novelist himself puts pen to paper or dictates his text to someone else. As to musical works, they are at once artistic, with the exception that the sounds replace the lines and colour, and literary, to the extent that words accompany the melodies.”
Every work is a composition (selection and arrangement) of building blocks, such as words, colors, shapes, or sounds. We say that a painting is a work, just as we say that we sell a car, not grant all right, title, and interest. These simplifications are necessary, but they must not overshadow what is really protected by copyright.
What does originality mean?
An important part of this is the aforementioned creative freedom, sometimes referred to as “originality.” Many national laws provide that, to enjoy protection, the works must be original in the sense that they possess creativity. This was because it was clear that it could not be the case that someone could claim copyright protection in a phrase as simple as “I like bananas.” However, originality should not be confused with novelty.
According to the ECJ, for something to be original, “it is both necessary and sufficient that the subject matter reflects the personality of its author, as an expression of his free and creative choice” (e.g. C-683/17, Cofemel, para. 30). Meanwhile, the U.S. Supreme Court has said that “originality requires only that the author make the selection or arrangement independently (i.e., without copying that selection or arrangement from another work), and that it display some minimal level of creativity” (Feist Publications, Inc. v. Rural Tel. Service Co., 499 U.S. 340, 358).
“Linguistic expression” in the words of the CJEU is about words whose originality can be achieved “only through the choice, sequence and combination of those words” (C-5/08, Infopaq, para. 45). A photo can be a work by choosing the framing or angle of view (e.g. C-145/10, Painer, para. 91). In the US, choosing the background, border, placement, or angle of a banana on the wall seems to satisfy the originality threshold (see Morford v. Cattelan, Case No. 21-20039-Civ-Scola, 14).
In the EU, originality is not satisfied when the expression “is dictated by technical considerations, rules or constraints which leave no room for creative freedom” (e.g. C-604/10, Football Dataco, para. 39). In parallel, in the US, if elements are “indispensable, or at least standard, in the treatment of a given topic,” they are excluded by the doctrine of scènes-à-faire (e.g. Hoehling v. Universal City Studios, Inc., 618 F.2d 972, 979). Similarly, there is no originality when “different methods of implementing an idea are so limited that the idea and the expression become indissociable” (e.g. C‑393/09, Bezpečnostní softwarová asociace, para. 49) or rather when “expression is so limited, idea and expression merge” (e.g. Zalewski v. Cicero Builder Dev., Inc., 754 F.3d 95, 102-103).
This reasoning attempts to avoid a monopoly. However, no one knows when the expression of an idea is so limited that it merges with the idea itself or is so generic that it cannot be protected. These are just rhetorical arguments used by courts to justify a fair judgment. Whether the court uses them correctly is up to the judge.
How can I tell if something is a work?
How can a judge decide whether a combination of certain elements is a work or not? It is impossible to answer that question without examining the creative process. Copyright law protects only human creations, and sometimes it is hard to tell the difference between human and AI creations. We need to show that a specific combination was created by a person. Saying that a process was creative is not enough. A work is the result of combining free and creative human choices that can then be seen in the final product. Therefore, I would say that there are two main ways that AI can be used to create a work. The following illustrates these two possibilities with examples from the USCO report.
The first option in the U.S. Copyright Office report was called “Expressive Inputs.” This is when someone creates a work, and then uses AI to reinterpret it. The report provides an example of “Rose Enigma” (p. 23).
The input is an original drawing protected by copyright law. However, the AI-generated output is not protected in the same way. Currently, AI systems interpret the input using the prompt as a hint, but the final form is more or less the interpretation of the algorithm. Therefore, the output is protected only in the part that includes the input. This means the position of the nose, lips, and cheekbones in relation to the shape of the mask, the arrangement of the stems and rosebuds, and the shape and arrangement of the four leaves.
This protection is similar to what might hypothetically be called the reverse of a derivative work. Rights to the derivative work do not extend to the underlying work (here, the input). In this case, however, there is no derivative work in the output. The artist’s (not necessarily the user’s) copyright is protected only to the extent of the underlying work. The prompt alone does not add much to the case. At most, it may add some elements to consider, suggesting, for example, that flowers are roses, but no one said they had to be red.
The second case is about modifying or arranging AI-generated content. Anything not created by a human is not protected by copyright, but that does not prevent the author from using those elements to create a work. If a person puts certain components together, the resulting work is just a collage of unprotected elements. You can achieve this effect using the inpainting functionality (explained below) as shown in the report (p. 26).
These tools permit the author to select and arrange individual elements, unlike the statistical interpretation of the prompt itself. However, not every use of these tools will lead to an original output. The simple combination of two elements will most likely be covered by the “merger doctrine”, whether or not the relevant jurisdiction calls it explicitly like that. On the other hand, incorporating unprotected elements into a larger work will not affect the protection of that work as a whole. For example, an AI-generated image shown during one of the scenes of an original film will not cause the film to lose its protection.
Conclusions
But are AI-generated outputs economically different from other types of products? Is it not true that anything a person makes can have some value, especially if someone is willing to pay for it? Why do we need laws that do not address economic needs and protect only selected creations?
Lawmakers should not pass laws just to keep lawyers busy. Not even to encourage creation, but to secure the right to benefit from the fruits of one’s labor. We should not protect only original creations and then argue about whether something is original or not. This is the case especially since there are neighbouring rights that do not require originality. AI-generated music, whether original or not, is protected as a phonogram. Instead of discriminating against other forms of expression, we could protect any expression of information. This will help us focus on what copyright disputes should be about: similarity as mere coincidence or as a result of copying.
These are not the only problems with such regulations. There are many more, especially when it comes to enforcing copyright law in practice (regardless of the type of work). Even the jewel in the European Union’s crown, the AI Act, has more holes than Swiss cheese, and copyright is peeking out of those holes (see, for example, here and here).
For more about copyright and AI creations, especially if you prefer a more academic format or want to understand China’s position on the Berne Convention, see our new scientific article.
This research received funding from the National Science Centre, Poland, grant no. 2021/41/N/HS5/02726.
________________________
To make sure you do not miss out on regular updates from the Kluwer Copyright Blog, please subscribe here.
