Re-Animating the Archive: Performing a Machine Learning System as Living Memory

Jonathan Reus

Proceedings of the International Conference on New Interfaces for Musical Expression

Abstract

This paper examines machine learning systems as a form of performable archival memory, operating within long-durational musical instruments. It introduces re-animation as a conceptual framework for understanding how ML systems trained on voice and text operate as unstable, probabilistic memory systems that continually reinterpret the past through inference and training interventions. Drawing on the year-long generative radio broadcast In Search of Good Ancestors / Ahnen in Arbeit, the paper grounds this framework in a concrete artistic case study. Central to the conceptual work of this paper is a reframing of instrumental liveness away from immediacy or low-latency control, and towards a temporally distributed liveness unfolding across dataset creation, iterative retraining, and durational listening. The paper further positions dataset creation as a form of musicking, where collaborative data creation workshops used in the work frame voice and text contributions as gifts rather than fungible resources, embedding reciprocal address, care, and future-oriented intent directly into the instrumental system. This orientation recognizes voice as an inherently relational act whose ethical dimensions persist beyond its digitization, and offers counter-practices to extractivist data epistemologies prevalent in mainstream AI development. By treating the training and curation of an ML model as a continuous, performative, and social practice, this paper contributes a reorientation in thinking about ML instruments, arguing that musical significance in ML-based instruments can emerge through duration, care, and collective intervention rather than virtuosic immediacy.

Citation

Jonathan Reus. 2026. Re-Animating the Archive: Performing a Machine Learning System as Living Memory. Proceedings of the International Conference on New Interfaces for Musical Expression. DOI: 10.5281/zenodo.20784499 [PDF]

BibTeX Entry

@inproceedings{nime2026_164,
 abstract = {This paper examines machine learning systems as a form of performable archival memory, operating within long-durational musical instruments. It introduces re-animation as a conceptual framework for understanding how ML systems trained on voice and text operate as unstable, probabilistic memory systems that continually reinterpret the past through inference and training interventions. Drawing on the year-long generative radio broadcast In Search of Good Ancestors / Ahnen in Arbeit, the paper grounds this framework in a concrete artistic case study. Central to the conceptual work of this paper is a reframing of instrumental liveness away from immediacy or low-latency control, and towards a temporally distributed liveness unfolding across dataset creation, iterative retraining, and durational listening. The paper further positions dataset creation as a form of musicking, where collaborative data creation workshops used in the work frame voice and text contributions as gifts rather than fungible resources, embedding reciprocal address, care, and future-oriented intent directly into the instrumental system. This orientation recognizes voice as an inherently relational act whose ethical dimensions persist beyond its digitization, and offers counter-practices to extractivist data epistemologies prevalent in mainstream AI development. By treating the training and curation of an ML model as a continuous, performative, and social practice, this paper contributes a reorientation in thinking about ML instruments, arguing that musical significance in ML-based instruments can emerge through duration, care, and collective intervention rather than virtuosic immediacy.},
 address = {London, United Kingdom},
 articleno = {164},
 author = {Jonathan Reus},
 booktitle = {Proceedings of the International Conference on New Interfaces for Musical Expression},
 doi = {10.5281/zenodo.20784499},
 editor = {Benedict Gaster and João Tragtenberg and Anna Xambó and Tom Mitchell},
 issn = {2220-4806},
 month = {June},
 note = {},
 numpages = {9},
 pages = {1317--1325},
 title = {Re-Animating the Archive: Performing a Machine Learning System as Living Memory},
 track = {paper},
 url = {http://nime.org/proceedings/2026/nime2026_164.pdf},
 year = {2026}
}