Exhibition floor talk | Uncanny Valley
Loading...
Output Type
Other
Publication Title
Exhibition floor talk | Uncanny Valley
Publisher
Edith Cowan University
School
School of Arts and Humanities; ECU Galleries
Description
Exhibition Statement | Masahiro Mori’s uncanny valley hypothesis describes the relationship between the human-like appearance of an object and the emotional response it produces. Affective responses such as unease and eeriness can arise when replicas and representations of humans appear or behave in non-human ways (MacDorman & Chattopadhyay, 2016). Exploring this phenomenon in a machine learning art space, Tynan curated images of synthetic human faces created using Nvidia’s StyleGAN2, a Generative Adversarial Network or GAN (Karras et al, 2020). This form of generative image modelling is a zero-sum game of unsupervised machine learning in which a discriminator and a generator compete against each other to produce highly realistic images. A generator’s role is to produce fake data that is recognised as plausible by a discriminator, which assesses the fake data for realism according to the training inputs. A GAN could be described as a kind of “peer review process” (Altavilla, 2017) for data forgery. The basic premise is that a network is trained to examine patterns in a large dataset, with the intention to generate new data with the same characteristics as the data it was trained on (Wood, n.d., Rayne, 2021; Vincent, 2018; Karras et al, 2020). In this case Flickr-Faces-HQ was used, an image dataset of 70 000 human faces collected under permissive licenses (Rougetet, n.d). To create morphing video portraits of human-like faces, Tynan animated the latent space of images produced by Nvidia’s StyleGAN2. Latent space, which remains ‘hidden’ until it is extracted by an algorithm, is a “representation of compressed data in which similar data points are closer together in space” (Tiu, 2020). Human facial features, learned by a GAN, were extrapolated and controlled to produce new synthetic facial data in the form of digital images Photorealistic computer-generated imagery contributes to the eroding distinction between reality and fiction in digital media spaces. Images of human faces produced by GANs could be described as uncanny in their ability to provoke feelings of eeriness when a viewer recognises their ‘almost perfect’ human likeness. There are numerous hypotheses for the uncanny valley effect. The perceptual mismatch hypothesis suggests that negative experiences of a stimulus are caused by “an inconsistency between the human-likeness levels of specific sensory cues. Clearly artificial eyes on an other- wise fully human-like face—or vice versa—is an example of such inconsistency” (Kätsyri et al, 2015, p. 7
Artist Bio | Darren TYNAN is an interdisciplinary artist, researcher, and educator from Perth, Western Australia. He has exhibited work in solo and group exhibitions and collaborated with artists in Australia and abroad. Tynan’s research interests include Anthropocene studies, eco-horror, glitch art, aleatoric composition, and applications of machine learning in digital media. As an artist, Tynan uses chance based processes and explores technical malfunction as a mode of creativity. He has recently incorporated machine learning technology into his artistic practice, such as generative adversarial networks, text-to-image generation, and autoregressive language models.
Additional Information
Exhibition dates: 10 Mar - 7 Apr 2022 | ACDC Floor Talk: 23 Mar 2022 12.30 - 1.30pm