Copyright law and training of generative AI models: Professor Dornis presents ex-pert opinion in the EU Parliament

09/05/2024

Generative AI (e.g., ChatGPT) can generate texts, images, music, and videos. The technology is based on extensive training with copyrighted data, usually ta-ken from the Internet without a license. In the USA, AI developers invoke the “fair use” defense, according to which unauthorized use of protected material is permitted for educational, scientific or research purposes, among others; the dis-cussion in Europe, on the other hand, has only just begun.

Professor Dornis and Professor Sebastian Stober (Institute for AI (IKS) at Otto von Guericke University in Magdeburg) have presented an expert opinion in the EU Parliament.

“The main argument against the permissibility of generative AI training under European and national copyright law is that the AI models extensively access “expressive elements“ of the works exploited during training - and not just the “pure information“, as has been consistently claimed to date,” explains Professor Dornis when asked. “We came to this conclusion after a thorough look into the engine room of AI technology.”

The expert opinion has been published as an open access book.

Go to list