A Case Study of Design Concept Generation Based on Knowledge Operation With a Large Language Model: Architectural Design Using Christopher Alexander’s Pattern Language

Abstract

Abstract This paper examines a case study using Christopher Alexander’s Pattern Language in architectural design to explore the impact of knowledge operation processes on design concept generation with a Large Language Model (LLM). The knowledge operation for the pattern language involves the chain of thought of defining design context, functional requirements, and physical attributes of alternative solutions. Three prompting methods were compared: 1) presenting only the design problem, 2) adding plain text of the pattern language, and 3) including both the pattern language and instructions for knowledge operation. Experiments were conducted with an off-the-shelf LLM, generating three design concepts in text and image for each prompting method. Thirty-four mechanical and architectural design students evaluated the concepts for validity, novelty, and feasibility. Results showed that the prompt involving knowledge operation led to limited feasibility but the highest novelty and moderate validity. The study suggests that the knowledge of conceptual design theories and methodologies can facilitate LLMs in effectively generating novel and valid design concepts.