Thesis Title
The Emergent Properties of Latent Space: Generative Artificial Intelligence and Simulation in New Computational Culture.

Research Description
Overall, this project will deal with the mechanization of textual and visual culture using simulation theory to yield a novel understanding of GenAI as a technology of simulation. I argue that theories of simulation provide the necessary theoretical tools to understand the imbrication of computation in the process of textual and visual production. 

This thesis is a continuation of recent work in media studies that focuses on the increasing role of media technologies as part of our physical and social environment (Hansen, 2015). The environmental turn of media artifacts facilitates the collection of mass amounts of data that have been put to work in the training of Generative Artificial Intelligence (GenAI) systems. The result of the training of GenAI on such vast datasets is their advanced mimetic and simulation abilities which can be seen in their increasingly realistic content generation. The generation of increasingly unique texts and images is facilitated in part by what is known as the latent space – a space in which mass amounts of data are compressed and clustered in relation to their content, allowing for the generation of a text that draws on not one, but many source images or texts. Latent space and its generative capabilities present a fascinating object of study for simulation theory.

As things stand, simulation theory – often calling to mind the work of critical theorist Jean Baudrillard (1929-2008) – is used in current scholarship to understand the novel affordances of GenAI. However, this scholarship is often lacking in depth in its application of simulation theory and is often content with the concept as it is, moreover, it often ignores the concept of simulation as an action undertaken in a computational format in a range of scientific disciplines.

The task of this research is to bring these two halves of simulation scholarship together to better understand GenAI across three case studies: Sudowrite, a creative writing application, Sora, a text-to-video application and Websim.ai, a community that utilizes GenAI hallucinations to create a fictitious – simulated – version of the internet. To do this, this work uses technographic methodology (Bucher, 2016) enriched by deep reading of simulation and mimetic theory to better understand these GenAI tools as they manifest in popular culture.

Supervisors
Assoc Prof Tim Graham, Queensland University of Technology
Prof Peta Mitchell, Queensland University of Technology
Dr Ben Nicoll, Queensland University of Technology