Deep Ruin: Climate Disasters in Imaginary Lands

Australia Council practice-based research + creative development support material.

The themes I explore in this creative development + practice-led research proposal develop long-term artistic preoccupations I have with landscape, romanticism and data-driven art. In several interesting and important ways these themes merge in the fields of artificial intelligence and human-machine creativity, both directly and as metaphor.

For the last few years I have been exploring the technical aspects of machine learning (AI) systems and their applications for immersive media. I wish to take this research further into the creation of new artistic works by establishing a creative dialogue with AI systems that extends beyond a simple linear application of machine learning models. 

The emergence of machine learning in 2014 applied to art and pictorial representations was in some senses shocking, as it seemed to subvert what were regarded as uniquely human capabilities in imagination and visualisation. But this had been a long time coming, since the earlier autoencoder artificial neural network (ANN) research of Hinton, LeCun, Bengio, Schmidhuber and many others working on machine intelligence. I recall a similar perceived threat to the artistic process in the 1980’s, when early paint programs and applications like Photoshop emerged. These are now a common part of the artistic toolset. But they were unintelligent tools, so there is arguably something different about machine learning, as it seems to replicate and displace certain human cognitive functions and ask questions about exactly how ‘uniquely’ human they are.

Deep Learning (DL) models are high-dimensional representations of datasets (which represent some information about something observable), from which intermediary or future states can be inferred. This draws the analogy with imagination: extrapolations or inferences from observations we make about the world or our internal states. 

How DL systems create representations is fascinating in itself. They can generate striking ‘inhuman’ interpolations in the ‘latent space’ of representations – intermediary states that we would not normally be aware of or would filter out, by interpolating through the ‘latent space’ between one image representation and another:

BigGAN Image Space Interpolation. Peter Morse (2019).

This research proposal revolves around exploring AI processes with a specific application to artistic visions of landscape and monumental ruins, inspired by the work of C17th ‘ruin’ painters Panini, Robert, Bloemen, Piranesi and others of the romantic era (eg Turner, von Guerard, Chevalier etc.). These will be hybridised with contemporary Australian and Antarctic landscapes, derived from my extensive photographic catalogue. The hybrid latent space will be subject to a variable data-driven climate model, producing novel interactive landscapes of architectural and landscape decay. Needless to say, one of my objectives is for viewers/users to reflect upon the pressing contemporary issue of climate change and our relationship with the natural world.

P. Zucker. Ruins – An Aesthetic Hybrid. The Journal of Aesthetics and Art Criticism, 1961.

The research will draw upon interdisciplinary philosophical perspectives upon landscape, refracting ideas from Malpas’ “The Place of Landscape’ (MIT, 2011), Twitchell’s “Romantic Horizons’ (UMP, 1983) and the literary study of ruins. Of great interest to me is the great Australian ’emptiness’, a sort of post-colonial mythography that could all too easily be embedded in the algorithmic prejudices of AI systems. My work questions this, evoking computer-imagined inhabitation that has disappeared, leaving monumental traces in imaginary lands ruined by climate disasters. The research will explore metaphors for a post-human future in an uninhabited/uninhabitable world. Arguably, this is a new place that a machinic imagination provides us access to: a damaged landscape of AI hallucinated monuments, eroded by a data-driven climate model. 

The accompanying images below map out some of the technologies, experiences and aesthetic resources I will be drawing on, with specific reference to my work in Tasmanian, Australian and Antarctic panoramic landscape photography. These novel climate-affected landscapes of ruin will form an interactive ‘explorable’ terrain for immersive media such as fulldome and VR. 

DomeLab system (EPICEntre, UNSW). Weather simulation (Peter Morse, 2017)
ML Training Schematic. (Peter Morse, 2019)

Panoramic Landscape Photography

Examples of my panoramic landscape photography in Australia and Antarctica. To be used for AI training.

Cape Denison. Antarctic Panorama. (Peter Morse, 2010)
Seascape, Tasmania. (Peter Morse, 2016)
Desert Landscape, Northern Territory. (Peter Morse, 2017)

Tasmanian Romantic Landscapes

Examples of Tasmanian Romantic landscape imagery for Style-Transfer training

William Piguenit. Tasmanian Lake Landscape (Mt. Olympus, Lake St. Clair). 1878.
John Glover. Classical Landscape. c.1820.

Classical Ruins

Some examples of classical ruins dataset for ruin-generator AI training.

Giovanni Battista Piranesi. Arch of Titus in Rome. 1748.
Giovanni Battista Piranesi. Temple of Minerva. 1756.

Stills from “Destruction of Germania”, Peter Morse, 1998.

You may also like...

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.