What Language Does AI Speak? Expressive AI and An Anthropology of Machine Languagestext
Yunus D. Telliel
Worcester Polytechnic Institute
Generative AI poses a challenge to anthropological understanding of human language. While AI outputs mimic human linguistic forms, AI's expressive capacity does not fully operate like that of humans. Human languages are rooted in embodied and embedded histories, whereas LLMs inherit those histories as decontextualized data rather than as conditions for new enunciations. Drawing on my collaboration with computational designers and researchers, I propose adapting insights from expressive range analysis (a computational method for characterizing and visualizing a generator's capabilities in a low-dimensional space). Focusing on the category of expressiveness across human and machine languages, I propose a framework that shifts the comparative basis from what a language produces (the actual) to the boundaries of the sayable (the potential). This framework offers anthropologists a third way of studying AI, one that requires neither anthropomorphizing it nor reducing its expressive power to mere statistics.