The Impact of Large Language Models on Functional Diagram Generation: An Extended Cognition-Based Investigation of First-Year Architecture Students

Abstract

This study investigates the impact of Large Language Models (LLMs) on the functional
diagram generation process of novice architecture students within the framework of extended
cognition theory. Through a quasi-experimental design involving 41 first-year students, the
research compares topological networks generated through traditional sketching with those
produced through GenAI interaction for a weekend house project. Quantitative analyses of
spatial nodes and edges reveal a significant cognitive expansion: LLM assistance increased
the number of spaces by 11.4% and connections by 17.9%. However, the findings also indicate
a qualitative shift toward instrumental dominance. Students made statistically insignificant
curatorial reductions to the AI-generated complex networks, acting more as passive editors
than active co-creators. Furthermore, extended dialogue loops resulted in additive topological
expansion rather than hierarchical refinement. The study concludes that while AI offers
substantial capacity enhancement, architectural pedagogy must cultivate spatial critical
thinking to filter the hyper-productive nature of AI.