Generative AI is used to create music more and more. This chapter explores how this phenomenon affects aspects of musical interpretation. The way in which AI generates texts, pictures, music and other media indicates that compared to human creators – it is likely to reduce the range of interpretative variants in its output. This argument is developed in three steps: firstly, a number of examples of AI-generated songs are presented in order to outline what AI can do at the moment. Secondly, the way in which large language models (LLMs) – the underlying systems of many AI generators work is discussed, showing that LLMs have a tendency to standardise their results, while also removing any quirk or idiosyncrasy that human creators would (unconsciously) include. AI outputs are “depersonalised”, as it were. Thirdly, the concept of “model collapse” is introduced. It leads to further homogenisation as AI systems will inevitably include more and more data in their processes that are already AI-created, setting in motion a spiral of self-referentiality that is likely to lead to a collapse of the systems in the long run. While this may not be inevitable it poses enough of a risk to give us food for thought and spend more time assessing the danger posed to interpretative variety by generative AI.
Building similarity graph...
Analyzing shared references across papers
Loading...
Wolfgang Marx (Thu,) studied this question.
Wolfgang Marx
Building similarity graph...
Analyzing shared references across papers
Loading...