The prompt engineer of an AI as the originator

Who can be an author?

According to German and European copyright law, an author can only be someone who makes a personal intellectual creation. This presupposes a human being. If, for example, a monkey takes a selfie, there is no copyright for this photo.

There have already been legal issues that are now becoming relevant again when assessing the authorship of AI products, for example in connection with whether photographs taken by speed cameras are protected by copyright. Among other things, this was based on whether a human had previously carried out certain configurations of the speed camera.

Another - now again relevant - legal issue concerns Mandelbrot fractals. Here, fractals are created on the basis of the work of mathematician Benoît Mandelbrot, which are particularly peculiar and interesting at the same time. In this discussion - on which no case law has been passed - the possibility of authorship of such images was predominantly denied.

The question of authorship arises anew in the context of artificial intelligence. Irrespective of the question of whether artificial intelligences are themselves in breach of copyright due to the data used for their training, the question arises as to whether the product of such an AI, e.g. a text or an image, has an author.

From an abstract point of view, possible authors are

• the authors of the training data

• the manufacturer/programmer of the AI

• the operator of the AI

• the user of the AI

• the AI itself

Any rights of authors or other parties entitled to the training data are not likely to continue in the result of the AI - judged according to German and European copyright law. However, see the aspects in the section "Co-authorship" below.

With regard to the manufacturers and operators of the AI, copyright protection rights could certainly be considered. For comparison: Under copyright law, for example, there is the right of the so-called sound carrier manufacturer ("music label"). A music label has a right to a pressed CD, for example, irrespective of the copyright protection of the songs on it. The purpose of protection here is the music label's economic investment in the artist and the "music". Even the removal of a single sound from the CD can then constitute an infringement of the music label's right. Whether a single sound - or more precisely a "sample" - may be extracted in the context of hip-hop music - inter alia under the aspect of artistic freedom - without infringing copyright is a question that has occupied the courts for over 20 years (the "metal on metal" case).

Prompt engineer as author

In the present case, the question of whether the user of an artificial intelligence can be considered an author will be examined in more detail. Such a user is often a so-called prompt engineer, i.e. a person who - instead of programming - operates the prompt of a finished AI in a particularly "artistic" manner and thus directly influences the results of the AI ("prompt engineering").

This consideration has a parallel in traditional copyright law: according to copyright law, a distinction must be made between the protection of photographic works and the protection of photographic images. While a level of creation must be reached for the protection of photographic works, this does not apply to simple photographs (but a lower protection threshold). The exact difference does not need to be explained in detail here. What is relevant, however, is that ultimately a photographer operates a camera and creates a photograph. The photographer has neither produced the hardware of the camera nor developed the software for it. Nevertheless, the photographer is classified either as the author of a photographic work or as the photographer. The reason for this is that the photographer "operates" the camera in a particularly skillful way, e.g. by choosing a particular angle, selecting the scenery and, if necessary, setting other parameters such as the exposure time.

Based on this, it would be logical to also grant the operator of an AI, i.e. the prompt engineer, a property right. Whether the prompt engineer is then to be classified as an author or as a mere "AI creator" in direct or analogous application of copyright provisions is certainly a question that requires further legal examination. It may also be worth updating copyright law in this respect - at least for clarification purposes.

The advantage of such an approach is that not every prompt engineer automatically becomes a copyright holder. Rather, a person who merely operates an AI in an extremely "mundane" way or makes an input for the generation of an AI result would not be protected, whereas a person who deals with the prompt in a much more in-depth manner may be entitled to an IP right.

Post-processing

The result of the above considerations is confirmed when a comparison is made with the use of image editing software. Persons who only make simple drawings, such as circles or other simple geometric shapes, with image editing software do not become the owner of the property right. Persons who create extensive graphics, on the other hand, become authors, even if they have "only" operated the software.

In this respect, it should be noted that the results of an AI often undergo post-processing - whether by the prompt engineer himself or by another person. In any case, these post-processing operations can quickly lead to the creation of a copyright, at least for them.

Co-authorship

It should be noted that the result of the AI is based on the training data used. It is quite possible that the result of the AI corresponds in part or almost completely to a training data set, for example a specific image used for training. The result generated by the AI may therefore already be protected by copyright. Whether and to what extent this is the case may be difficult to determine. Search engines are increasingly being created to find such "pre-existing" rights.

Conclusion

In summary, there is much to be said for the creation of a copyright in favor of the prompt engineer. However, this assessment is based on a variety of legal issues that are currently still unresolved. In particular, there is still no case law. It may also make sense to amend the Copyright Act to clarify this.

However, care should already be taken now to ensure that prompt engineers are sufficiently involved when using artificial intelligence. This can be done, for example, by concluding a usage rights agreement / license agreement with the prompt engineer as a precautionary measure. Otherwise, there may be a risk of a claim for injunctive relief by a prompt engineer now or at a later date - or vice versa: the prompt engineer may demand injunctive relief and damages.

Please note that there is currently no established prevailing legal opinion on the above considerations and - as far as is known - no case law. The details of the assessment may therefore be subject to change and also depend heavily on the AI used in each case. Nevertheless, the precautionary measures outlined are already recommended.

Image above by Gerd Altmann from Pixabay.