VIRTUAL New Schools of Thought

Abstract

The relationship between descriptions of buildings and representational images has always been ambivalent (1). In De Architectura, Vitruvius “described” buildings—he did not draw them. It was only during the Renaissance that his treatise was illustrated. Neither were the Tower of Babel, Dante’s Inferno, Shangri-La, El Dorado, or Dune accompanied by images. They were first and foremost biblical accounts, treatises, and literary productions. It was only through subsequent interpretations by painters, architects, filmmakers, and other imaginaries that these texts were reborn into images. Today, another contender has emerged capable of transforming text into image: AI-powered tools. This paper shares the outcome of a funded research project on developing methods to harness AI image productions. The grant proposal included two workshops where students facilitated the process of “translation” between the textual and visual representations through rigorous contextualization of their design intentions. We incorporated geoclimatic and environmental settings, social uses, material conditions, and building techniques to articulate the “context.” Embracing the Nietzschean notion of creativity, students staged organized chaos. For Nietzsche, the creative “will to power” was animated by the tension between the Apollonian spirit of order, restraint, and control, and the Dionysian disposition towards frenzy, madness, and ecstatic intoxication (3). We juxtaposed this conceptualization of creativity with the notions of “Hallucinations and Bias” in Large Language Models (LLMs) in AI systems. (4). The goal was to establish the conditions of possibility for aleatory variations within tools such as Midjourney, Stable Diffusion, and DALL-E. We assessed the divergence between the ‘intended’ results and the ‘unintended’ variations generated by the AI image-making tools. This research underlines the significance of contextualization in the success of harnessing AI tools in architecture.