Closed Rooms and Oracles: the poetics of vector space in Sam Riviere’s Conflicted Copy

Published one hundred years after Tristan Tzara’s instruction that poets should “copy conscientiously”, Sam Riviere’s Conflicted Copy (2024), created in collaboration with GPT-2, invites us to reconsider one of the core principles of procedural poetry: that language can speak for itself. Large Language Models are the archetypal copyists, fed by vast datasets which they convert into a matrix of probability; but aside from legitimate concerns about how conscientiously these error-prone algorithms copy, Riviere’s poems pose questions about what it means for poetry to collaborate with an entity that, albeit in a highly prescriptive manner, also uses language. Building on John Searle’s ‘Chinese Room’ thought-experiment, which illustrates the so-called black box problem in AI: the fact that we do not understand exactly how artificial intelligence systems work, this article will explore three ‘closed rooms’: first, the Covid-19 lockdown in England and Wales during which Riviere produced his poems; and the enclosed space of the dialogue box in which he entered his prompts and GPT issued its responses. In the former, uncertainty collects around the location and origin of the speaking voice; in the latter, both poet and LLM are revealed as vastly distributed, networked selves with a distinct material presence. Finally, there is the ‘o-machine,’ another thought experiment, devised by Alan Turing, in which an enclosed machine consults an external oracle. Turing’s o-machine imagines intelligence as a collaboration between machine and not-machine. As such, Conflicted Copy offers a release from an anthropomorphic understanding of cognition.