Strategies of Interconnecting Deep Learning Models in AI-Driven Design Systems

Incorporating deep learning (DL) models into architectural design presents challenges, despite the potential to inform new design processes. Developing DL-driven systems requires identifying components and relationships between these constituents in the new design workflow. The study tests a novel DL-driven design workflow, employing systems theory to deconstruct the design process into its component parts. The proposed workflow considers design an “exploratory activity” involving the modification and evolution of problem goals and methods used to achieve those goals and determine the types of connections and combinations of those connections. DL model connections tested here involve sequential, parallel, branching, with designers choreographing the interaction between human agents and DL agents. The research involves testing the proposed workflow prototype through three case studies of different strategies. Evaluation of those work-flow strategies was performed through the linkography method, to assess how different connection strategies yield different processes. The study's significance lies in crafting “hybrid” processes that leverage human intuition and machine intelligence, yielding a creative state when the design space becomes ever-changing, in a process that unfolds new possibilities of design exploration.