Generative AI may fundamentally transform music recommendation. Recommendation systems emerged to manage the overwhelming abundance of music made available by streaming platforms. Yet, these systems operated on the assumption that songs were relatively stable cultural objects: discrete works to be discovered, revisited, shared, and loved. The large-scale introduction of generative music into streaming platforms challenges this assumption.
Some recommendation paradigms may become increasingly ineffective. Collaborative filtering, for instance, depends on patterns of repeated listening and shared consumption; however, if musical abundance increases by several orders of magnitude, such signals may become significantly less reliable. Generative music also raises the possibility of a radical transformation in which the song itself ceases to function as the fundamental unit of musical reception as music might become hyper-personalised, continuous, adaptive, and endlessly generated - something listeners tune into and out of rather than select and replay.
Rather than disappearing, recommendation may undergo an ontological shift: from selecting works within a finite catalogue to governing generative musical flows. This transition raises critical questions about platform power, transparency, and cultural diversity. Streaming platforms already shape listening practices through recommendation infrastructures; generative AI may intensify this capacity by enabling fully adaptive musical environments optimised for engagement. In the talk, I will explores these tensions and considers what forms of resistance may emerge in response.