Machine learning systems are increasingly becoming part of virtually all aspects of networked literary culture. Chat bots, virtual assistants, and news-generators are just some examples of the ubiquity of automated agents. Most of these systems are designed to represent humans, but on a more technical level, they actually (mis)represent datasets. In this chapter, I investigate David Jhave Johnston’s augmented poem, ReRites, written in collaboration with a machine learning system, in an effort to articulate a poetics of misrepresentation. This poetics circulates the seemingly paradoxical circumstance that the outputs of machine learning systems both do and do not represent their datasets. They are the direct result of a data-intensive computational process, but they also exist independently thereof. I articulate this poetics of misrepresentation using a critical vocabulary of mimesis, specifically the concepts of figura and simulacrum, vis-à-vis contemporary discussions of machine learning. The poetics of misrepresentation points to the simultaneity of sameness and difference between dataset and model in machine learning-based text generation and shows that pattern recognition is poetic — as well as poietic, i.e., a process of creation. Literary experimentation prompts a critical understanding of machine learning in literary media, which also points to the necessity of considering machine learning as a new paradigm of mimesis that affects the very conditions of mimetic practices as such.
|The Routledge Companion to Literary Media
|Astrid Ensslin, Julia Round, Bronwen Thomas
|Udgivet - aug. 2023