XX

xCoAx 2020 8th Conference on Computation, Communication, Aesthetics & X 8–10 July, Graz online

Convoluted Alterity

Keywords: Alterity, cGAN, Tinder, Image-to-Image, Data Scraping.

This work is part of an exploration of human agency within online dating networks. While understanding human agency as the capacity of the individuals to act autonomously and to experience the impact of one’s own free choices, I dare to suggest that human agency is, and perhaps it has always been, on a perpetual path to obsolescence through a continuous and inevitable merge with other forms of agency. These other forms recognized as alterity go beyond the individual, the human, the machinic and the material. The agonizing process is observable in the everyday forms of ephemeral media, of transient relations, of short-lived attention, of the never-ending representation crises. The inescapable anxiety relies on the denial that the human individual might have always been a construct, hence an illusion. Current state of the art machine learning technologies, allow to generate photo-realistic portraits from people who doesn’t exist, that are hard to distinguish from real ones. These images are generated thanks to a machine learning model-training process that integrates hundreds of thousands of portraits of real humans. Currently, these styles of machine generated imagery is being used for all sorts of commercial and political purposes, so far without any moral or ethical concerns, as it is claimed that non-real, “imaginary” or imagined-by-GAN’s people isn’t really affecting any real human directly. [1] The question then arises as to what are the role and implications of these new others and what is left there for the originals?

Convoluted alterity explores the aesthetics of sameness, using self-portraits retrieved from the popular online dating network Tinder, where people represent themselves in order to become part of an offer of desirable beings. I extract these pictures using scraping algorithms, which I have developed as part of my doctoral research and artistic processes [2] in order to conduct data scraping from specific users over the Application’s API. The portraits are manually manipulated in order to mark label-regions with RGB color-coded blobs that allow a cGAN model [3] to distinguish which parts of an image are the fragments corresponding to the self-curated human shells. On one side of the installation, the edited portraits are presented as still images inviting the viewer to speculate about who of all the possible beings could fill the uncanny void from the missing pixels. On the other side of the installation, the trained model dynamically depicts some of the possible ways in which the other and the same co-exist. The boundaries between the individuals are erased, the empty shells melt through the poetics of convoluted imagery. The produced imagery is projected as a UV Map on an amorphous blob.

Samples of the portraits scraped from the Online Dating Network Tinder and the Generated Imagery. The visible skin on the human portraits is masked by the artist in order to allow the model to be trained with the pixels corresponding to the color labeled regions.

Online version available at

Notes

  1. Generated Photos commercializes “worry-free” model photos. The technology behind it is a machine learning model called StyleGAN (Dec 2018, Karras et al. and Nvidia ). https://generated.photos/
  2. Tinder Scraper: https://github.com/andresvillatorres/tinder_light_scraper
  3. The cGAN model proposed is based on the pix2pixHD model form NVidia and the styleGAN. An implementation of it is available for RUNWAYML.

References

Join the conversation