Interactional foundations for critical AI literacies — Mark Dingemanse (2026)

Mark Dingemanse, Centre for Language Studies, Radboud University Nijmegen. "Interactional foundations for critical AI literacies", preprint, under review for a volume on critical AI, 2026. DOI 10.5281/zenodo.19452872. The author applies a CC BY licence to the accepted manuscript. The PDF is filed as sources/ingested/Dingemanse-2026-Interactional foundations-for-critical-AI-literacies-preprint.pdf.

The wiki reads this text on Does a Work Still Need to Be New?.

Opening

Dingemanse opens with the source code of Claude Code, released in March 2026, whose prompts read like incantations ("Report outcomes faithfully"), and compares them to an Azande witch-doctor's apprentice addressing his medicine (Evans-Pritchard 1937). His aim, following Lucy Suchman, is "to slow down discourses of the 'smart' machines" and to trace how people have been "thinking with things, outsourcing judgement, and making sense of interactively presented non-sense".

Divination and horoscopes

"Formulating a prompt for a large language model is not so different from posing a question to an oracle." Divination involves "ritualized procedures that generate chance outcomes and subject them to human interpretation", from casting lots to tarot. It can be described "as the generative use of chance" (Morgan 2016). Its experts preside over the procedure without being party to the question, which gives it "ostensive detachment" (Boyer 2020). Astrology adds personalisation. Studying horoscope columns, Theodor Adorno found "apparently specific references" so general they always fit; Dingemanse, after Garfinkel, calls this "specific vagueness", whose role "is to offload interpretive work to participants".

Sense-making in interaction

In Harold Garfinkel's experiment (1967), students put yes/no questions to a "counsellor" who read the answers from a random list. They found the exchanges "helpful" and left "with a feeling of received knowledge". Even when the answers are random and the whole burden of sense-making falls on one person, that person can experience the exchange as meaningful. Joseph Weizenbaum's ELIZA showed the same with a computer program.

Contemporary language models

Reinforcement learning with human feedback tunes model output to human preferences and gives it its "chatty and docile appearance"; so-called reasoning repeats, at scale, Ellen Langer's experiment on "placebic information". The models are "finely adapted to human sense-making".

"Prompt engineering" as divination

Prompting guides and accounts of spider divination (Zeitlyn 1990) are hard to tell apart, because both are "interactional practices evolved in response to the challenge of working with chance outcomes": unpredictable answers, "lore" about best practice, and questions shaped for the consultation. A longitudinal study (Alcaras & Ricci 2025) found that regular use restructures work: people "discretize their activity", deal with "scattered layers of work", adapt to the model, and shift "toward logistical manipulation of outputs and away from forms of engagement that sustain a sense of accomplishments"; "a change happens in the texture of labor, as it loses differentiation and alters the distribution of agency". Dingemanse's formulation: "in prompt engineering, it is not only the prompt that is being engineered, but also the prompter."

Conclusion

People go to great lengths to make sense of output that is "specifically vague, personalized, and designed to conceal a lack of understanding"; "we are both primed for language models and prompted by them". Citing Margaret Boden on ELIZA, such programs "respond to, rather than understand, language". Ada Lovelace warned against overrating a new engine's powers, and also against the reaction that comes after, "to undervalue the true state of the case". Dingemanse closes: "Rather than letting ourselves be distracted by machines of mediocrity that have been built to capture our preferences and sell them back to us, we should focus our efforts on better understanding and celebrating the astonishing creativity and flexibility of human sense-making."