Generative AI and copyright – a difficult relationship

The challenge:

Generative AI and copyright have a tense relationship. Copyright law deals with the question of who created a work and who has the right to dispose of it. Generative AI challenges this traditional concept. For AI to work, it must be ‘trained’ with a large number of existing works, most of which are protected by copyright. At the same time, the use of these systems creates new content that may resemble known works, come very close to them, or even combine several works. This raises the question of where exactly the copyright-relevant ‘red line’ lies and who can assert rights.

From the lecture:

At the heart of the matter is the classic legal concept of a work as a personal, intellectual creation. The law focuses primarily on the relationship between the author and the work. It is intended to protect against unauthorised use and theft and to ensure appropriate remuneration. The effort that has gone into a work is irrelevant to the scope of protection.

In generative AI, countless works are included at the training data level without it being possible to identify which traces continue to have an effect in the model. At the model level, the question arises as to whether the trained system itself can constitute a copyright infringement if it is capable of imitating or even reproducing training material. At the output level, there is debate as to whether and under what conditions an AI-generated text or image can itself be considered a new work or should only be understood as a derivative of the data sets. Added to this is the copyright issue with regard to prompts. Can the latter be protected by copyright? When humans give instructions, choose styles and motifs, and thus guide the systems in a certain direction, it must be discussed whether the prompt itself is an intellectual creation or whether the decisive creative contribution is to be seen in the model.

Another problem area is models that have long been trained with extensive data sets. It is difficult to take retroactive action against this use of data, especially since many providers are based in other jurisdictions. It remains to be seen whether government authorities should take action or whether it will remain a matter for individual copyright holders to take legal action.

Perspectives:

Generative AI challenges us to think about what we want to understand as creative creation and authorship – and how we protect it under copyright law. Legally and politically, the aim is to regulate the interaction between data, model, prompt and output.

Furthermore, the question arises as to how authors can be appropriately involved if their works have played a role in training. At the same time, the classic goals of copyright law must continue to be considered. It is intended to protect the creative achievements of human beings. The difficult relationship between generative AI and copyright law lies precisely in defining these goals appropriately under new and challenging conditions.