Proto (4AD) is the culmination of a series of experiments and songs utilizing machine learning. As well as it being the first album of its kind, we also staged large performances at major festivals and institutions, created videos, musical tools, and a novel framework for the integration of AI techniques and the data they require into an art practice. Unlike most music and ML projects, we focussed on generating sung voice, rather than perpetuating historical attempts to produce an automated composing system. We view Spawn (our ML system, imagined as a baby hungry to learn) as a collaborator, performing sung interpretations of material we composed. 
This was an attempt to depart from kitsch 20th century narratives of AI as displacing the human, instead focusing on my belief that AI, trained on a chorus of human contributions, is an evolution of human coordination technology with origins in group singing techniques themselves. 
To present a 21st century narrative of AI, we assembled an ensemble of singers and developers to train Spawn, all of whom were accredited and compensated for their data. In performance, contrary to regrettably barren electronic music standards, my stage was filled with humans singing. We included our audiences in our training, by inviting them to sing in euphoric call and response with Spawn. We recorded tens of thousands of voices, and Spawn can now sing in the group voice of those who trained her. 
The project continues as we write new music and develop new concepts for vocal exploration and protection in relation to machine learning. Our vision of data interdependence in relation to ML is cited far outside the field of music. During lockdown we started a podcast, “Interdependence,” to speak directly to decision makers encompassing Google Brain, OpenAI, and Spotify on the necessity of accrediting and compensating those whose data pivotally contributes to their AI aspirations.  The ultimate goal is for our musical experiment to impact policy and public perception around machine learning.

Links: https://www.hollyherndon.com/proto

Composed and produced: Holly Herndon, Mathew Dryhurst 
Ensemble members: Annie Garlid, Marshall Garret, Houwaida Goulli, Stine Janvin, Jules LaPlace, Roman Ole, Josa Peit, Lyra Pramuk, Soriane Renaud, Albertine Sarges, Evelyn Saylor, Colin Self, Spawn, Elizabeth Wood 
Additional production: Ville Haimala 
Additional arrangement: Evelyn Saylor 
Recorded in Berlin by Guy Sternburg, Paul Pilot, Holger Jost, Holly Herndon 
With support from #bebeethoven

Holly Herndon (US) operates at the edges of avant-garde and pop music to establish a dynamic, disruptive canon of her own. On her most recent album  PROTO, Herndon fronts and conducts a choir comprised of both human and A.I. voices. Herndon’s music comments on how technology affects humanity in thought-provoking, and often unsettling ways.

Mathew Dryhurst (GB) releases music, research and artworks solo and in conjunction with creative partner Holly Herndon. He teaches at NYU’s Clive Davis Institute of Music, and hosts discussions with figures shaping 21st  century culture on the Interdependence podcast.