The algorithm's hidden layers: Nurturant matrices, quantitative poetry, and the religious ethics of language technology

Abstract This paper builds on conversations at the intersection of artificial intelligence research and the linguistic anthropology of religion by exploring a mathematical technique that connects contemporary Large Language Models [LLMs] with the history of Islamicate occult sciences. In both cases, transformation algorithms are used to entextualize linear streams of discourse as geometric embeddings in higher dimensional arrays. Focusing on an algorithmic poem from a nineteenth‐century Palestinian manuscript on the healing magic of midwives, I argue that the entextualization of language in computational space emerged as a ritual practice of self‐transformation. A historical perspective on mathematical dialogues suggests that both older and newer algorithmic language technologies can reengineer selves and societies; but the comparison also exposes the peculiarity of our own assumptions about the opaque, quantified, and gendered language of math. By separating the technical affordances of semiotic transformations from the larger ideological frames into which they are taken up, this paper seeks to model a multilayered method of comparative analysis that can inform future linguistic anthropological studies across diverse contributions to the history of LLMs.