Copyright in generative deep learning

Abstract Machine-generated artworks are now part of the contemporary art scene: they attracting significant investments and presented in exhibitions together with those created by human artists. These mainly based on generative deep learning (GDL) techniques, which have seen a formidable development remarkable refinement very recent years. Given inherent characteristics these series novel legal problems arise. In this article, we consider set key questions area GDL for arts, including following: is it possible to use copyrighted works as training models? How do legally store their copies order perform process? Who (if someone) will own copyright generated data? We try answer considering law force both United States European Union, potential future alternatives. then extend our analysis code generation, an emerging GDL. Finally, also formulate practical guidelines artists developers working art, well some policy suggestions policymakers.