Grids and Bags: Two Ways to Summarize People

Media scholars traditionally understand computation as a process of sorting the world into a few reductive categories, such as a grid of potential values for race and gender. But in the last seven years, large language models have gone in another direction: they now assemble petabytes of unsorted data without categorizing them first. This essay provides a genealogical overview of this approach, which creates what is known to mathematicians as the multiset or bag. I argue that the creation of multisets or bags derives from early modern techniques for gathering and summarizing the singularities and oddities of the world (particularly people) into a single representation. By focusing on the period between 1600–1665 and on still-life paintings that produce a statistical sense of summary out of botanical specimens and ethnographic recordings, I explore how the disordered, jumbled dataset emerges as the verso to orderly grids and Linnaean taxonomies. Because multisets and bags have historically been constructed out of alterity—out of automata, indigenous bodies, and Asian languages—they also indicate a different way of understanding the operations of race inside a large language model.