Closed Loop

2017, 2-channel digital video without sound, 200 minute loop

Closed Loop is a recording of two artificial intelligence models conversing with each other - one with words1 the other with images2 - in a never-ending feedback loop. The words of one describe the images of the other, which then seeks to describe the words with a fresh image… Two neural networks getting lost in their own nuances, sparking and branching off each other as they converse in a perpetual game of AI "Exquisite Corpse".

Closed Loop has had no textual or visual input by the artist

1) DenseCap: Fully Convolutional Localization Networks for Dense Captioning (2016)
2) Plug & Play Generative Networks: Conditional Iterative Generation of Images in Latent Space (2016)

Technical

Two neural networks: a language captioning recurrent neural network writing what it sees in the images generated, and a generative neural network generating images responding to the words generated. Two datasets to train the algorithms: a dataset of 4.1 million captioned images to train a language network, and the Imagenet dataset of 14.2 million photographs to train the image generator network.

“The conversation shifts between the logical, to the enigmatic to the frankly incomprehensible, produces constant frissons of surprise and in the end a profound sense of displacement from one’s own settled perceptual self.” - (64/1, curatorial collective)

Special thanks to Roland Arnoldt (New York based creative technologist and collaborator), and Anh Nguyen et al. at Evolving-AI for their work on GANs.

Single channel rendering of two channel video installation

Jake Elwes digital artist installation shot from Future Intelligence 2019 - Tank Museum, Shanghai Jake Elwes digital artist installation shot from Future Intelligence 2019 - Tank Museum, Shanghai Jake Elwes digital artist Installation shot from Writing the History of the Future 2019 - ZKM, Karlsruhe