생성형 AI를 활용한 공간디자인 프로세스의 재구조화 연구

초록

(Background and Purpose) Generative artificial intelligence has spread rapidly across the design industry, and this trend is frequently described as the automation of design labour. Such framing obscures a more fundamental shift taking place inside the spatial design process itself. Existing studies address individual courses, individual projects, or tool-level error types, while international studies decompose design outputs or specific studio tasks; consequently no integrated stage model spanning research through presentation has yet been proposed, and there is no shared frame of reference explaining at which stage, by whom, and through what judgment the process is reorganised. This study argues that generative AI has not automated spatial design work but has restructured the process by introducing new stages, new participants, and new loci of human judgment, and proposes an eight-stage integrated model to substantiate this argument. (Method) This study is based on a literature review of domestic and international academic papers, conference proceedings, and industry reports retrieved from KCI, RISS, DBpia, AURIC, and Google Scholar between May and June 2026, using the keywords Generative AI, Spatial Design, Interior Design, and Workflow. Four inclusion criteria and three exclusion criteria were applied explicitly, and ten studies, nine domestic and one international, were selected for analysis. The traditional six-stage process was compared with the AI-integrated process stage by stage. To compensate for the concentration of the sample on journal and conference papers, theories drawn from design methodology, environmental psychology, human-computer interaction, human-in-the-loop artificial intelligence, and the philosophy of artificial intelligence were reviewed as the theoretical grounding of the proposed model. (Results) The traditional six-stage process is restructured into an eight-stage model: Research, Concept, AI Generation, Selection, Validation, Re-generation, Documentation, and Presentation. Validation and Re-generation are new stages absent from prior process models. Validation requires human judgment across three layers, structural, contextual, and intent-related, and the layer at which validation fails determines the depth to which re-generation returns, whether local, intermediate, or full. This causal mechanism distinguishes the proposed model from a mere subdivision of existing stages. Human judgment is weakest at AI Generation and strongest at Validation, indicating that generative AI has not displaced the designer but has transferred the burden of judgment to a newly emerged stage, reorganising the process into a hybrid structure in which one segment cycles locally while the remaining stages proceed linearly. (Conclusions) The growing adoption of generative AI should be understood as a restructuring of the spatial design process rather than an automation of design labour. Practical implications include redistributing organisational capacity toward Validation and expanding the designer's role to that of curator and validator, while educational implications include redesigning curricula around the validation and re-generation loop and treating foundational writing instruction as a prerequisite for lower-year students. The proposed model is a hypothetical model derived from literature analysis, and its empirical verification is left to subsequent research, including in-depth case studies, Likert-scale surveys of practitioners, and controlled comparisons of task time and output error rates.

키워드