Automata at Court: A Logistic Framework for Working with LLMs

Friday 28 October // 14:15 // Drawing Studio

Despite the rapid adoption of Large Language Models (LLMs) in the realm of visualizing architecture and design projects, their practical and theoretical integration into the larger schema of design processes remains largely unexplored. This study proposes a logistic framework for deploying LLMs in design workflows beyond the domain of the visual. Using the motif of automata as characterized machines, the authors theoretically frame and conduct an experiment incorporating programming and natural languages. Establishing connections between Wolfram Mathematica and the OpenAI ChatGPT API, enables programmatic access to outsource design roles through role-play to a GPT-3.5-turbo model and Midjourney V5. With this setup, “Automata at Court” can unfold as an architectural stage play, with LLMs role-playing some of the most visionary architects – among them Claude Nicolas Ledoux, Lisa Bo Bardi, Leon Battista Alberti and Margarete Schütte-Lihotzky – hosting a long series of fictional events in 32 chambers of the Palace of Versailles. These are then further synthesized into 128 generated sceneries. Apart from this playful yet practical demonstration, the proposed framework shows how multiple intelligences can work together in the context of a design project and raises questions beyond ontological distinctions of human and machine, nature and artifice, generalist and specialist, for future LLM-aided design processes.