What language do machines speak? LLMs and the simulated social

Abstract Large Language Models (LLMs) form a ground of convergence between computational capacity and language in unprecedented ways—a convergence that occupies a growing literature. Using OpenAI’s Chat GPT, I explore a somewhat counter-intuitive concern—rather than following the emerging dilemmas between human language, cognition, and humanist philosophical conundrums on one hand and on the other, ‘literate’ machines, I ask what language do machines indeed speak. By tracing the computational crafting of that language, my intention is to undergird a social imaginary in LLMs that is coherent and systemic, amenable to being called a simulated social, where simulation is not false nor farcical, but rather a technological craft of meaning making that coheres a machinic reality around a machinic language. This intent begins to address what notion of the social begins to emerge in the ubiquitous universe of LLMs. Two perspectives so far underexplored in the emerging literature are privileged here. Recognizing language as both a social fact par excellence and as a sign system that is anchored in the social, the discussion proceeds, first, by aligning the linguistic sign with computing technique in granular detail. Second, it calibrates sign systems with meaning making by exploring how metaphoric and metonymic constructs in LLMs simulate a social. This, I expect, leads us to paying good heed to what language machines do indeed speak and what social they create.