Some Reflections on the Potential and Limitations of Deep Learning for Automated Music Generation

Abstract

Deep Learning and Artificial Intelligence are slowly revolutionizing many fields of applications, having the potential to replace humans in a variety of tasks and jobs. Nevertheless, creativity has always been considered something inherently human: recent research shows that, however, this may not always be the case. From this standpoint, this paper focuses on music and on the recent advancements in deep learning applied to the generation of musical content. We argue that, while those models are able to produce results that could actually be considered music, the role of the human musician still remains preponderant in the production of a musical piece. We here reflect on such limitations, directing our efforts to imagining new tools and instruments that may allow to experience new forms of interaction while supporting novel processes of creativity and music production.