From LLM to LLH (Large Language models to Large Language Humans): Instruction as Medium in Human-AI Role Reversal

Abstract

Large Language Models (LLMs) are often framed as a tool to be “prompted” where humans command, and the machine operates. However, as AI becomes increasingly intermeshed in our lives, role divisions blur, and humans take the role of interpreting and carrying out machine instructions. This project explores the concept of Large Language Humans (LLHs), describing this role reversal where humans become the operators of AI logic. Using a Research-through-Design (RtD) approach, we developed the Large Language Human Machine: an embodied, “black box” device that prints cryptic, ritualistic instructions for the user to perform. Inspired by instruction-based performance art (e.g., Yoko Ono, John Cage), the project treats instruction as a design material to expose shifting dynamics of authority, opacity, and agency. We document the iterative process from screen-based prototypes to a self-contained physical artifact. By viewing this process through four design parameters: persistence, tempo, uncertainty, and opacity, we discuss how materializing instructions as physical “receipts” reconfigures the user’s commitment to the machine’s agency.