K21 in Düsseldorf into a training ground where humans and AI co‑produce choral music, making the labour, datasets and protocols behind “intelligent” systems tangibly audible. The project stages AI not as an autonomous agent but as a coordination technology, inviting visitors into live vocal recording sessions that feed a public AI choir trained on a songbook derived from Hildegard von Bingen’s Ordo Virtutum.
Starmirror proposes a scenario in which artificial intelligence models coordinate humans to generate intelligence for collective benefit, extending Herndon and Dryhurst’s longstanding work at the intersection of music, machine learning and experimental organisational forms. In collaboration with the design and architecture studio sub, the artists turn the main hall at KW Institute for Contemporary Art into an immersive sound and light installation that doubles as a recording studio, listening environment and living archive for new choral works.
On selected Sundays, choirs, an ensemble and visitors are invited into call‑and‑response recording sessions based on a songbook developed from Hildegard von Bingen’s twelfth‑century morality play Ordo Virtutum, where a soul navigates between virtue and temptation. These sessions produce a public choral dataset that trains a “Berlin AI choir” to be premiered as a work in the second part of the exhibition at K21 Kunstsammlung Nordrhein‑Westfalen, Düsseldorf.
Before entering the main hall at KW, audiences encounter Arboretum, centred on Public Diffusion, an image model trained entirely on public‑domain material and built on the PD40M corpus of around 40 million images, where each motif appears at least 600 times to count as a general concept. Together, Arboretum and the choral training ground form a public AI protocol in which human and nonhuman agents contribute to a shared context, foregrounding questions of authorship, deepfakes, the platform economy and the possibility of “agentic social mediation, not social media”.
