Questionable devices: Applying a large language model to deliberate carbon removal

This paper presents a device-centred approach to deliberation, developed in deliberative workshops appraising methods for removing carbon dioxide from the air. Our approach involved deploying the Large Language Model application ChatGPT (sometimes termed “generative AI”) to elicit questions and generate texts about carbon removal. We develop the notion of the “questionable” device to foreground the informational unruliness ChatGPT introduced into the deliberations. The analysis highlights occasions where the deliberative apparatus became a focus of collective critique, including over: issue definitions, expert-curated resources, lay identities and social classifications. However, in this set-up ChatGPT was all too often engaged unquestioningly as an instrument for informing discussion; its instrumental lure disguising the unruliness it introduced into the workshops. In concluding, we elaborate the notion of questionable devices and reflect on the way carbon removal has been “devised” as a field in want of informed deliberation. • Explores roles for large language model applications, like ChatGPT, in participatory research and policy deliberation. • Develops the concept of “questionable devices” to analyse how ChatGPT introduced unruliness into deliberative practice. • Argues that closer attention to devices may empower participants to engage politically with deliberative processes. • Reflects on the way that carbon removal has been “devised” as a field in want of informed deliberation.