Voices and Voids | collaboration
Responding to current concerns about the ubiquity of voice assistants, this artistic research focuses on building a series of performative artifacts that aim to challenge AI and ML technologies, and to examine automation through the prism of “ghost work” that constantly support these systems. By allowing AI agents to listen to our most private conversations, we become receptive to this mediated care, while forgetting or ignoring how much these automated interactions have been pre-scripted. While these interactions cultivate a sense of familiarization with the non-human, they also corroborate the impact of Late Capitalism and the Anthropocene. Within these contradictions we see an opportunity to reclaim, examine, and ultimately transcode this data through an interdisciplinary performance project, by developing embodied experiments using a combination of design, data-driven art, cyber crafts, found-object and traditional percussion instruments, spoken word, and movement.
– from Voices and Voids blog
For the Voices and Voids project, led by Audrey Desjardins, Afroditi Psarra and Bonnie Whiting, I used Google Assistant training data, provided by Amazon MTurk workers in exchange of a remuneration, to train text-generating neural networks, namely GPT-2 and textgenrnn. Over 20,000 lines of data were collected, cleaned and then used to produce AI-generated poetry. The result below:
sound art by Danny Clay // video by Audrey Desjardins // poetry by GPT-2 and textgenrnn, trained by Gabrielle Benabdallah
for their training data