Abstract: The growing field of “critical algorithm studies” often addresses the cultural consequences of machine learning, but it has ignored music. Te result is that we inhabit a musical culture intimately bound up with various forms of algorithmic mediation, personalization, and “surveillance capitalism” that has largely escaped critical attention. But the issue of algorithmic mediation in music should matter to us, if music matters to us at all. This article lays the groundwork for such critical attention by looking at one major musical application of machine learning: Spotify’s automated music recommendation system. In particular, it takes for granted that any musical recommendation – whether made by a person or an algorithm – must necessarily imply a tacit theory of musical meaning. In the case of Spotify, we can make certain claims about that theory, but there are also limits to what we can know about it. Both things – the deductions and the limitations – prove valuable for a critique of automated music curation in general."