The Hamburg Regional Court has dismissed a stock photographer’s lawsuit against the non-profit organization Lion. Lion provides training data sets that can be used to train AI models. For this purpose, the provider creates link lists including related data. The stock photographer sued: he said the process infringed his copyright. Lion is represented by the law firm Hedrich Rechtsanwalte, with Georg Hedrich Heise being legal counsel for Median.
Advertisement
The court confirms the legality of Lion’s actions. The freely accessible data set “Laion 5B” includes an image of the photographer. It would be better if it included a link that leads to the image on the Internet. There is also an image description and information in text form about what can be seen in the image. To check, Lion used software to check images to see if the content and text fit together. For this purpose, Lion downloaded the image, analyzed it and, according to its own statements, removed it again.
The photographer’s image was watermarked and said could not be reproduced. The case was based on this. So the question was whether Lion’s creation of the data set was a legitimate copy or not. Heydrich lawyers write in a press release: “The reproduction of an image for the purpose of selecting training data – according to the court – is text and data mining within the meaning of the statutory exceptions.”
Processing of images is permitted for research
The reasons for the decision also say: Specifically: Lion reproduced the photograph, but this fell under §60d of copyright law. There is an exception for paragraph research purposes. This means that permission was given to process the image for research purposes.
Against this background, according to the court, there is no need for a final assessment as to whether the limitation regulation in section 44B of the Copyright Act applies – “Whether the defendant can rely on the limitation regulation in section 44B URHG appears doubtful Is.” 44B regulates so-called text and data mining.
The question of the legality of AI providers using Lion’s data set for training AI models is also not addressed in the decision. However, in this context the judge cites a study by the Copyright Initiative. It is also said that when the training data set is created, it is impossible to predict what will happen downstream – during training and while generating images.
The judge also cited European copyright law (DSM) in the reasons for the decision. The AI regulation states that providers of general purpose AI models are obliged to find strategies to comply with applicable copyright laws.
Representatives of Lion EV welcomed the decision in a press release: “The decision gives us legal certainty and enables us to continue promoting innovative and scientifically based projects with a high level of transparency and reproducibility that benefit the whole of society. benefit.” The association will continue to provide free data sets for research in the field of artificial intelligence.
The plaintiff can appeal within one month.
Disclaimer: The law firm of Heise’s legal advisor Georg Heidrich represents LAION in the proceedings. He was not involved in the creation of the article,
(EMW)
