Your Song: Epistemological Foundations of Music Recommender Systems

With the massive popularisation of music streaming platforms, music recommender systems (MRSs) have become a prominent technology and an important topic of research in data sciences. However, they are still poorly understood within the humanities. Applications of MRSs include the use of Artificial Intelligence (AI) to analyse enormous amounts of data such as music audio content (e.g. duration, key, loudness), user-related data (e.g. listening patterns and preferences), and metadata (e.g. images and texts associated with each song). These complex analyses of billions of data are transformed into recommendations, for example in the form of automatically generated playlists that are distributed and personalised on a daily basis to millions of listeners worldwide. Hence, MRSs have been directly and quickly transforming how music is experienced in the 21st century. However, so far, we do not have a comprehensive humanities-based characterisation of MRSs.

In this project, I will provide the first comprehensive, critical, historically and philosophically informed analysis of MRSs. I will do so by shi ing the focus from the computational and sociological/cultural aspects of MRSs to their epistemological status as a science and technology of music in the 21st century. First, I will construe the epistemological foundations of MRSs by identifying and addressing the following underpins: (i) the knowledge presupposed by and supporting MRSs research and applications; (ii) MRSs’ objects of enquiry; (iii) the methods MRSs employ to analyse their objects; (iv) the knowledge MRSs generate; (v) the way MRSs represent and communicate this knowledge; (vi) the validation or justification of the knowledge generated; (vii) and the uses and purposes of this knowledge. It is expected that this analysis will disclose the foundations of a specific understanding of music aesthetics, a new way of approaching musical objects and subjects (e.g. songs and experiences). I will then investigate whether the same epistemic patterns are to be found in other technologies such as AI soundscape generators and emerging prompt-based music generators. Then, I will introduce and develop the hypothesis that this techno-scientific approach to music repeats epistemic patterns that were central in many 18th century music theories and aesthetics within the so-called affective-rhetoric tradition. Finally, I will situate those underpins within trends in the epistemology and ethics of AI to address what could be the corresponding musical aesthetic challenges of MRSs such as diversity, criticism, and gender bias.