In the field of Architecture, the expanding array of generative artificial intelligence (genAI) tools being developed and applied to the five phases of design thinking are rapidly being adopted into architecture education and practice. This research investigates methods of integrating genAI tools in architecture design studios. Specifically, we explore how text-to-image AI can accelerate students’ iterative design processes during the ideation and prototyping phases of design thinking.

Our research method involved three activities: 1) Testing Tools, which included testing and cataloging several text-to-image AI tools; 2) Deploying Pedagogy, in which we developed two 1-week design Modules, and 3) Fostering Dialogue, which involved dialogue with academics and practitioners and hosting a Symposium + Poster Session.

Our findings from ~150 students’ work revealed that using text-to-image AI in the production of conceptual perspective and elevation images enabled students to develop between 4-10 times more than the number of design iterations produced without text-to-image AI, and we discuss the numerous transformative effects this has on the design process. We also found that fostering various forms and formats of dialogue, including student-to-student, faculty-to-student, and practitioner-to-academic, provides needed feedback into the design thinking process, and is critical to the successful integration of AI with architecture studio pedagogies.

How architecture education programs will incorporate AI-based workflows—and to what ends—is one of the foremost questions facing the academy today, and the answers will have profound implications for the future practice of architecture. We expect the pedagogical approaches we have developed, along with the findings from their implementation, to enrich and advance the discourse on the integration of AI in design thinking. We also identify potential lines of further inquiry into this process, including how responding to ethical concerns with genAI could foster productive parameters for future pedagogical explorations.

Building Information Modeling (BIM) has become a dominant design authorship platform within the building construction industry, whose standards are governed by codes and regulations with complex interrelationships. Ensuring compliance with these industry standards during design and construction is a challenge regardless of advancements in BIM technology. Manual design reviews for code compliance are subject to human errors and are not a scalable effort. Natural language processing (NLP), deep learning methods, Convoluted Neural Networks (CNN), and Deep Convoluted Neural Networks (DCNN) have all revolutionized Artificial Intelligence and provided improved efficiency and accuracy of textual interpretations. However, these methods are still time and resource intensive. Large Language Models (LLMs) can offer a high level of language comprehension and syntax accuracy for optimizing building specifications, safety and regulation compliance through robust pre-training. This study proposes a novel approach to automating audits and quality control processes pertaining to code-reviews for building designs. This framework integrates LLM, and DCNN within a BIM interface, allowing for the automated audit of specifications based on the material and geometric relationships of object modelling in BIM, which is identified through deep learning. LLM reduces the reliance on annotated datasets, while DCNN allows for deep learning of text data from code regulations. This integration of deep learning and LLM significantly improves the accuracy of design reviews and building code compliance by automating the code review process with minimal manual intervention. This new framework provides a novel and adaptable approach for AEC stakeholders via a scalable and versatile solution for automated design audits and BIM quality checks to ensure the high integrity of building design and contracts.

Both conventional lumber and mass timber building products reduce the natural forms and grown intelligence of trees into dimensional lumber through a series of orthogonal cuts. From rectilinear geometries to filleted corners, lumber is optimized for predictable structural performance and speed and ease of human labor. New advancements that enable the cutting of non-planar timber elements present opportunities to expand the use of roundwood but require a reimagining of traditional fabrication workflows and a questioning of standardized lumber. This paper presents a specific pedagogical model for teaching novel digital construction methods with roundwood, through the format of a visiting hybrid workshop for graduate architecture students. Approaches to pedagogy, workshop format, assignment specifics, and fabrication tools and methods are discussed. Student work is presented, taking form as functional furniture prototypes – a series of stools and a collective table – suggesting larger architectural assemblies. The workshop pedagogy choreographs a series of learning objectives and hands-on experiences that enable students to reimagine lumber and engage with both analog and digital fabrication methods for roundwood. As the cutting surface transitions from a planar to a ruled surface, new formal, structural, and spatial potentials are revealed.