Self-attention (Vaswani et al., 2017) is the defining mechanism of today’s Generative Pre-trained Transformers (GPT), known as Large Language Models (LLMs). Self-attention is the extraction of long-range predictive relationships between every token (or other data element) in a dataset. Algorithmic self-attention clearly differs from embodied human attention; yet it enables the production of texts that superficially resemble human-written texts. Attentional processes are also deeply involved in the production and reception of human literary texts, notably poems. Indeed, Alford (2021) defines poetry as ‘an instrument for tuning and refining the attention’. This leads us to ask, to what do AI-generated poems tune our attention? Over time, might attentional biases—potentially introduced by LLM self-attention architectures—modify poetry readers’ and writers’ attentional attunements to language, self, and world?

To begin to explore these questions, my research turns to the Neurocomputational Poetics Model (NCPM; Jacobs, 2022), which sits at the intersection of linguistics, poetics, cognitive neuroscience, empirical aesthetics, and reader reception theory. NCPM offers an integrated empirical framework with which to examine the creative outputs of AI models and their attentional structures. In this work-in-progress report, I examine statistical features of the language of 360 poem versions generated by commercial LLMs (Claude, Gemini and GPT-4o) when directed to continue poems in the style of Emily Dickinson, starting from the first four lines of 40 original poems. Results indicate substantive differences between original and AI-generated poems with regard to poetic dimensions including lexical-semantic content, use of end-rhyme and enjambment, with implications for attentional attunement.

This exploratory and critical-creative paper maps Australian poetry written between 2000 and 2023 which engages with concept of the algorithm, as identified via the Austlit database. Much of this corpus of poetry references the post-digital ecology in which it has been composed, which is especially the case in Jordie Albiston’s collection, Euclid’s Dog: 100 Algorithmic Poems (2017), and also the history of mathematics and the evolving concept of the algorithm. Algorithms serve as metaphors, and as metonyms, for the pervasive predictive algorithmically-inflected environment in which readers are guided towards online news, culture and media. This paper performs a close reading of a selection of these poems, while also studying word frequency and mapping shared images and metaphors across the corpus. The paper culminates in the (playful) creation of several poems via large language models, based on existing creative work putatively about algorithms, and examines the ethics of such compositions.

Writing has never been a purely solitary endeavor, nor a purely human act. Discussants, readers, pens, paper, keyboards—these have always been part of the apparatus of authorship, shaping not only how we write but how we think. This paper reframes the novel as a cyborg creation: an assemblage of human intention, technological suggestion, and emergent meaning, situated within an ecology of forms and practices. Drawing on Donna Haraway’s cyborg figuration and practice-led methodologies, I explore how generative AI participates in this lineage, not as a radical break, but as a reminder and continuation of our longstanding entanglement with technologies of thought and expression. AI does not simply co-author; it unsettles. It mediates possibility, interrupts habits, and reveals how writing has always been relational, iterative, and more-than-human. Within this ecology, the novel becomes a site of attunement, where agency is distributed, and creativity emerges through friction, resonance, and feedback. Rather than clarifying authorship, AI’s presence destabilizes it, inviting us to ask not just who writes, but what writing becomes when its boundaries blur. In embracing the novel as a cyborg creation, this paper gestures toward writing as an intertechnological and intersubjective practice grounded in partial perspective, speculative openness, and ongoing transformation.