In a landmark decision, the Hamburg Regional Court has dismissed a copyright infringement case against LAION, a German non-profit organisation, marking the first ruling in Germany on artificial intelligence (AI) and copyright.
Photographer Robert Kneschke sued LAION for allegedly infringing his copyright by using his image in LAION-5B, a dataset of nearly 6 billion image-text pairs used to train AI models like Stability AI's Stable Diffusion. The court ruled that LAION's actions were protected under Section 60d UrhG of German copyright law, which permits text and data mining for scientific purposes.
The court determined that LAION's use of Kneschke's image, to compare with an image description, as part of compiling its database, constituted analysis for obtaining information about correlations, falling within the scope of the law's exception.
However, the ruling did not address the broader question of whether images can be used for training generative AI models. IP attorney Mirko Brüß noted that the data mining exception does not apply to research organisations collaborating with private enterprises that have significant influence and preferential access to research findings.
The case also touched on the interpretation of "machine-readable" restrictions. While not central to the judgment, the court suggested that given advancements in AI language understanding, natural language restrictions might be considered machine-readable in certain contexts.
Kneschke expressed surprise at the ruling, stating on LinkedIn, "At first glance, therefore, all in all a strange judgement, which presumably leaves enough evidence for an appeal."
This decision sets a precedent in Germany regarding AI and copyright law, potentially influencing future cases and discussions on the use of copyrighted material in AI training datasets.