This paper demonstrates the use of a media arts approach to literary criticism – a kind of “media art reading”, which engages with the new technology of Large Language Model (LLM) artificial intelligence. I show that despite, and perhaps because of, the shortcomings of current LLMs as literary writing tools, the boring prose they produce does offer something new to reading. This work emerges from the thinking in my book Glitch Poetics (2022), which shows that the contemporary writing that best reflects the real social conditions of today – that is ‘literary realism’ – is specifically not that coherent literature of unity, fluency, and symphony with which the genre is often associated, but rather a broken, faulty mode of writing; divergent from norms in ways that accumulate and correlate with our experience of the world.

One of the problems with reading for literary divergence, is the lack of some comparable ‘right’ literary norm to compare and contrast it with. Ironically, and perhaps too conveniently, LLMs offer such a thing. We can train GTP-3 on innovative literature, and it can believably ape a style. But it lacks the ability to torque and twist this style against the emergent realities of today; this is the quality, I claim, that turns literary stylists into literary realists for the current circumstances.

In this paper I introduce the technique of text generation for comparative reading, frame its potential through the notion of contingency from systems theory-influenced literary theorist David Wellbery (1992), and present some example comparative readings using noteworthy contemporary literature.

REFERENCES
David E. Wellbery, ‘Contingency’, in Ingeborg Hoesterey, Ann Clark Fehn, and Maria Tatar (eds), Neverending Stories: Toward a Critical Narratology (Princeton: Princeton University Press, 1992).
100
Jones, Nathan, Glitch Poetics (Coventry: Open Humanities Press, 2022)
Osborne, Peter, Anywhere or Not at All (London: Verso, 2019)