What can artificial intelligence do with copyrighted text? And who pays for it? This is the basic question behind the lawsuit filed by the Gema collecting agency against the ChatGPT operator OpenAIwhose decision is still pending before the Munich Regional Court today (10 am). In particular, the legal dispute revolves around the use of nine texts, some of which are very famous songs – including “Atemlos”, “Männer” by Herbert Grönemeyer, “Über den Wolken” by Reinhard Mey and “In der Weihnachtsbäckerei” by Rolf Zuckowski.
The texts are used for training GPT Chat used and, according to Echo, returned exactly or at least largely identical in response to a simple request to the system. Companies that represent authors against users consider this a copyright violation. OpenAI denies this. ChatGPT is a computer program that uses a language model to provide answers to all types of questions.
It is not yet clear whether the regional court will rule on it or refer the case to the European Court of Justice. Both sides proposed this because it concerns very fundamental questions regarding generative artificial intelligence and it is likely that the regional court’s decision will be challenged by the losing side.
Of fundamental importance also for literature, art and photography
Expert Silke von Lewinski of the Max Planck Institute for Innovation and Competition classifies it as follows: “The assessment in this test case will be of fundamental importance for all works, be they literature, journalistic texts, music, fine arts, photography or any other work used for generative AI. It is about how existing laws should be interpreted.”
If the final verdict is that Echo is correct, this would shift the balance of power between the creative industries and technology companies in favor of authors and other rights holders, von Lewinski said. “Before a text can be used for generative AI, rights holders must provide their consent and have the opportunity to receive compensation for it.”
If there is a verdict in Munich, the artists’ chances are quite big. The presiding judge, Elke Schwager, indicated in her initial assessment at an oral hearing in late September that she preferred to follow Gema’s arguments on nearly all key points.
Memorized or not?
The lawsuit (file number 42 O 14139/24) also concerns technical details. There’s no denying that the text is used to train the AI, but what happens after? Echo accused OpenAI of memorizing text in its system – ultimately storing it in a certain way – and then reproducing it. OpenAI, on the other hand, argues that it does not memorize data, but simply reflects what it learns during training. As his argument, he cited, for example, that some texts had been edited.
In his initial assessment, Schwager was a bit unsure. In simple terms, he argues like this: If a system is trained with text and then outputs almost or completely the same answers to simple questions, it is not realistic that this is a coincidence.
© dpa-infocom, dpa:251111-930-276208/1
