Paper Short Abstract:

Based on a series of experiments with a large language model ingested with ethnographic interviews, we reflect on moments of disconcertment, re-experiencing and estranging ethnographic encounters. We argue for a speculative use of Generative AI in Anthropology.

Paper Abstract:

Generative AI (GAI), such as Chat-GPT have permeated various spheres of academic work. They have also entered anthropological research and teaching. To sort haphazard engagements with virtues and fallacies, we propose there currently are both an Anthropology of and with GAI.

Anthropology of GAI includes tracing the computational and cultural resources algorithms mobilize and their acute effects, such as investigations into cultural biases, material-ecological requirements, or creative engagements of users. Anthropology with GAI includes experimentation with how to employ GAI in anthropological practices, such as ethnographic research, analysis, writing or teaching.

Contributing to the latter, we present a series of experiments with retrieval augmented generation (RAG), combining an open large language model with a vector database of ethnographic interviews. Our ethnographic data, drawn from three distinct projects, includes over 3000 mobile interviews on European movies, 230+ interviews from the COVID-19 pandemic, and a study of cybersecurity in 30 Danish SMEs. By engaging in chat conversations with our material, we prolonged and estranged the ethnographic encounter. We offer a series of reflections on the embodied experiments with the model, dynamics of conversations, and provocations that it evoked. We argue that we experienced a sense of companionship but also disconcertment and friction (Madsen et al. 2023). The experiments allowed us to facilitate spaces of speculation and allowing others to participate in or re-experience ethnographic moments. We argue that our experiments give reasons to further explore how GAI can be employed in anthropological practice and especially their merits as tools for speculation.