Zwischenheit: A Human-Machine Dialogue on Music Understanding and the Situatedness of Listening

A phenomenological exploration of the limits of machinic accounts of creativity is offered in this first case study of the panel, the audiovisual performance Zwischenheit (2025). Performed live during the panel, the work is inherently discursive and selfexplanatory. It consists of an improvised conversation between the composer and a language model, in which the latter is invited to reflect on soundscapes it "listens to" via an audio-captioning algorithm. Whereas neural-network engineering typically treats "music understanding" as an a priori computational task, Zwischenheit seeks a speculative, empirical, and situated alternative. The live exchange unfolds contingently, yet consistently foregrounds the gap between embodied cognition and learned statistical patterns. In this way, the unsituatedness of machine listening emerges not as a contingent limitation, but as an epistemological impossibility of genuinely understanding music.