Application Optimization Research on Full-Process Low-Error Integration of Generative AI and AIGC in Architectural Design

Abstract

Entering the 2020s, the discipline of architecture is amidst an epistemological crisis, transitioning from “experience-driven” to “algorithm-enhanced.” Although Generative Artificial Intelligence (AIGC) has demonstrated astonishing creativity in the conceptual divergence phase, technical hallucinations and logical fallacies caused by the “Image-Information Gap” have encountered immense resistance in structural rationality and construction implementation. This study aims to fill this gap by proposing ALEF (Application-Level Error-free Framework), a full-process closed-loop system built through hybrid reinforcement learning and physical constraint engines. This study first traces the historical evolution from traditional design to digital twins, revealing the essence of information loss. Subsequently, through empirical case comparisons between the ALEF paradigm and traditional paradigms, it quantitatively analyzes the specific impact of the ALEF framework on efficiency (an increase of about 30%-40%) and error rates (a decrease of about 60%-65%) in Text-to-BIM transformations. The research finds that by introducing cutting-edge physics-aware tools such as Neural Concept and PhysicsX, AI’s role can be reshaped from a mere “painter” to a “digital construction partner” equipped with engineering rationality. Finally, this paper demonstrates that under the new paradigm of human-machine collaboration, the architect’s role has not perished but has achieved a dual reconstruction of design ethics and creativity through the identity of a “curator.”