AVIARY

(2024)

5:05m, 4K video, custom-trained AI-generated images, birdsong, and music, color, sound

Aviary is a short film that roosts together speculative female, intersex, and queer birds and birdsong. Bird datasets from specimen collections to birdsong repositories have been demonstrated to contain male biases. Consistently, female, intersex, and queer birds are undercounted and understudied, leading to incomplete environmental understandings. Fancifully filling in these data gaps, the birds and birdsong in Aviary were created by custom-training AI on the limited number of real bird data avaialble for birds presenting as female, intersex, same-sex pairs, and non-nuclear family units—generating more, imaginary birds.

The image and birdsong data was collected from the Cornell Lab of Ornithology’s Macaulay Library, the largest respository of avian data in the world, and used to custom-train multiple image– and audio-generation AI models (neural networks). In addition to training the models on real birds, they were also trained on two supplementary datasets: (1) The entirety of John James Audubon’s The Birds of America, a well-known book of bird illustrations which also features male and heterosexual biases and surfaces the racist and colonial histories of birding; (2) The canon of Western classical music inspired by birdsong, itself dominated by a privileged few, which was revealed by Livio’s research to similarly reference primarily male birds. Intervening in these rich scientific and cultural datasets, Aviary speculatively writes missing birds back in.

Supported by:

CalTech-Huntington Art + Research Residency
New Music USA Creator Development Fund

Notes:

Aviary is a proof-of-concept for the feature-length documentary Salvaging Birds, which features these speculative birds, birdsongs, and music alongside interviews about bird datafication and avian AI.
With special thanks to the Cornell Lab of Ornithology’s Macaulay Library.