Language in AI Art: Encoding, Folding and Transforming

Garrett Lynch IRL

This article discusses four artworks that employ artificial intelligence (AI) as practice as research (PaR) by artist Garrett Lynch IRL. These are: I’m not Garrett Lynch IRL – DoppelGANger Portraits (2021), a series of Generative Adversarial Network (GAN) portraits; TheLastStraw (2021–23), a performative generative artwork for social media; Flag for States of Damage (2018), a performative mixed-reality artwork for the web and The Traveller (2024), a four-channel video installation with artefacts. The objective of the works is two-fold. Firstly, each artwork’s use of AI is distinct yet is intended to form part of a broad ongoing exploration of how networks can be transformative to art practice. The works maintain that AI is a form of network that enables emergence. Not the emergence of intelligence as defined in the field of AI, but instead in the context of art theory a manner in which artworks are expanded, extended or activated beyond their artist/author defined forms. AI as a network is therefore defined as both the generative adversarial network, the input, employed in the formation of the work and the resulting network of artist, artwork and audience that emerges when a work is expanded, extended or activated. Secondly, instrumental in facilitating AI as a network is the use of language as a combined form of encoding and performative utterance (Austin, 2018). Building on a fundamental basis of computing that all digital media is reducible to language, code, and numbers as well as a basis of communication theory that language is a social construct, the works explore language as both form of representation and communication between human and machine. Language enables a process of folding (O’Sullivan, 2005) or flipping (Sloan, 2012) of concepts, media and artefacts between ‘virtual’ and ‘real’ spaces, between digital and materialised forms.

|Message the AI

Original

AI Generated

Edits

Source

Methodology

In the encoding, or communication and representation, of content and meaning in my AI art language becomes a process, a methodology, of folding (O'Sullivan, 2005) or flipping (Sloan, 2012) concepts, media and artefacts between 'real' and 'virtual' spaces, analogue and digital forms, reality and fantasy, truth and fiction. By folding I refer to Deleuze's concept of the fold (le pli) as an ontology of becoming, in this case artworks being created or changing over time. However, folding is equally adopted to be a technical or procedural strategy that facilitates an artworks creation or change. This would include, for example, any artwork that is interactive or generative and constantly changes because of external input (interaction or data) informing the output (visuals, sound etc.) in a continual back-and-forth process. Sloan's concept of flipping, similar to Deleuze's fold, focuses on unpacking what such a technical or procedural strategy would consist of by describing it as a "process of pushing a work of art or craft from the physical world to the digital world and back" (2012). The works discussed in this article will unpack the various ways of how this has occurred in each and how, in turn, transform the work.

To emphasise the process or methodology of encoding and folding that occurs in the creation of the artworks, the writing of this article has employed generative AI. Writing has been generated in collaboration with GPT 4.0^[6]in the following procedural manner:

1 Source writing about each of the artworks^[7] was input into the AI. The AI output a first draft based on this input; 2 I then edited or reworked the first draft as a second draft and input this into the AI for changes or additions forming a third draft; 3 Finally, if required, one more round back-and-forth between myself and the AI as a fourth and fifth draft was employed to add final improvements.

This back-and-forth dialogical process, which can colloquially be considered as 'chatting' with the AI as I have used ChatGPT, is intended to echo what has occurred in the creation/performance of the artworks discussed. However, the process is also intended to reveal new ideas by allowing AI to make connections within the writing. The article presented to you, the reader, within this online interface takes advantage of the digital, networked and non-linear nature of the Research Catalogue publishing platform. It allows you to display or hide all text created in the process of writing the article: the original text, the AI generated text and text that has been edited out. This approach encourages you to read in a non-linear manner and is intended to give additional insight into the processes involved in creating this article and, in turn, the works. However, it should be noted that this writing approach is experimental and not without issues. These include navigation of the article, in particular between source and final compiled texts, and coherence of the article due to typographic markup that potentially adds complexity to the readability of the article.

6. GPT is an acronym of Generative Pre-trained Transformer. It is a multimodal large language AI model developed by OpenAI that can respond to both text and image input. GPT 4.0 is version four of the model released in March 2023.7. Note that in the AI-generated source for this article the AI has, when referring to me by my surname, not understood my double-barrel surname and shortened Lynch IRL to Lynch. These are mistakes by the AI that have been corrected in excerpts integrated into the final article, however, the mistakes have been left unaltered in the source.

No AI generated text was generated for this section.

Tip

Tip