Twice,again: Virentis – Bounded (Un)Supervision as World-Building Method
Proposal Type
Individual Talk
Location
Hypertexts & Fictions
Start Date
July 2026
End Date
July 2026
Abstract
"Twice,again: Virentis" sits within a wider ecosystem of world-building via a parent project ("Twice, again"), tied together via a narrative canon, which was written without revision as a form of 'automatic' writing. Via established distant reading methods (parts of speech extraction, frequency mapping and key word concordance extraction), fragments of the 50,000 word canon are extracted and used to influence the output of local large language and image generation models. Two LLMs are set in conversation with each other, one, taking the role of the planet Virentis (game-master), the other taking the role of a cartographer (player) mapping the planet. A framework of text-based adventure games is set up through system prompts, that come together with the canonical fragments to create a logic for the discussion to emerge within.
John Potts (Potts 2023) has described my previous e-literature work as positioning the author as demiurge–setting out the initial conditions and then standing back to see what emerges. In world-building terms, this can be seen as supervising the laying of the foundations from which a world can evolve, unsupervised, to find its own logic from within the fragments and the framework it is provided, using the knowledge embedded in the weights of the model as a wider (unknowable) archive.
Through close readings of generated dialogues, the network flow within this generation and the resulting cartographic artefacts (created via machine generated ASCII maps and local text to image diffusion models), this individual talk shows how “supervision” shifts between controlled authorial narration (what texts are admitted, what rules are enforced, and what infrastructures are employed) to emergent consequences of the relational networks. By releasing agency distributed decision move into a much wider ecology. I argue this produces a form of "bounded (un)supervision" where authorship shifts into the role of demiurge and medium. Inputs (canon, retrieval, prompts) become sites of distributed guidance where human and more-than-human creativity re-emerge in the resultant texts.
Twice,again: Virentis – Bounded (Un)Supervision as World-Building Method
Hypertexts & Fictions
"Twice,again: Virentis" sits within a wider ecosystem of world-building via a parent project ("Twice, again"), tied together via a narrative canon, which was written without revision as a form of 'automatic' writing. Via established distant reading methods (parts of speech extraction, frequency mapping and key word concordance extraction), fragments of the 50,000 word canon are extracted and used to influence the output of local large language and image generation models. Two LLMs are set in conversation with each other, one, taking the role of the planet Virentis (game-master), the other taking the role of a cartographer (player) mapping the planet. A framework of text-based adventure games is set up through system prompts, that come together with the canonical fragments to create a logic for the discussion to emerge within.
John Potts (Potts 2023) has described my previous e-literature work as positioning the author as demiurge–setting out the initial conditions and then standing back to see what emerges. In world-building terms, this can be seen as supervising the laying of the foundations from which a world can evolve, unsupervised, to find its own logic from within the fragments and the framework it is provided, using the knowledge embedded in the weights of the model as a wider (unknowable) archive.
Through close readings of generated dialogues, the network flow within this generation and the resulting cartographic artefacts (created via machine generated ASCII maps and local text to image diffusion models), this individual talk shows how “supervision” shifts between controlled authorial narration (what texts are admitted, what rules are enforced, and what infrastructures are employed) to emergent consequences of the relational networks. By releasing agency distributed decision move into a much wider ecology. I argue this produces a form of "bounded (un)supervision" where authorship shifts into the role of demiurge and medium. Inputs (canon, retrieval, prompts) become sites of distributed guidance where human and more-than-human creativity re-emerge in the resultant texts.
Bio
Andrew Burrell is a practice-based researcher and educator exploring virtual and digitally mediated environments as a site for the construction, experience and exploration of memory as narrative. Their ongoing research investigates the relationship between imagined and remembered narrative and how the multi-layered biological and technological encoding of human subjectivity may be portrayed within, and inform the design of, virtual and augmented environments. Andrew's networked projects in virtual and augmented environments have received international recognition. Andrew Burrell is a senior lecturer in the School of Design, University of technology Sydney.