Publications

Text and data mining (TDM) for AI training can be regarded as the starting point of a complex process that impacts the market for human literary and artistic creations in different ways. The machine is only capable of mimicking human content after it had the opportunity to derive patterns for its own productions from myriad human creations that served as training resources. Once AI training has been completed and a generative AI (GenAI) system is brought to the market, AI output may support fruitful human/machine collaboration. However, it may also kill demand for the same human creativity that empowered the AI system to become a competitor in the first place. In the terminology of the ubiquitous three-step test in international and European copyright law, this latter challenge raises the question whether copyright exceptions permitting TDM for AI training cause a conflict with a work’s normal exploitation. A closer inspection of the normal exploitation test shows that the chances of demonstrating a relevant conflict are slim in the case of AI training. Rightsholders seeking compensation for displacement effects caused by GenAI systems must resort to the final criterion of the three-step test and argue that the use for AI development unreasonably prejudices their legitimate interests. In practice, this means that copyright holders can hardly employ the three-step test as a tool to erode TDM exemptions altogether. They can only insist on the introduction of appropriate remuneration schemes to avoid unreasonable prejudice in cases of commercial AI training.