Deep Copyright: Up‑ and Downstream Questions Related to Artificial Intelligence (AI) and Machine Learning (ML)

Artificial intelligence (AI) systems are increasingly capable of taking over tasks, that until recently required cognitive abilities such as the creation of works. Already commercial applications reach the market, whose outputs would arguably be considered as creative, if produced by a human author. Also, systems have become capable of translating all kinds of text including prose between dozens of languages. One enabler for this development is (deep) machine learning (ML), which may require input from copyrighted works to train the respective models in becoming »creative«. The present essay discusses some of the legal and philosophical issues that arise from these developments, looking both at the status of the »downstream« generated works, and the possible constraints that copyright law might impose on the materials the systems need for learning and hence for their »upstream« modelling.