Generative AI in structural education: preliminary column design in architecture

Abstract

This article investigated the impacts of using generative artificial intelligence, through ChatGPT, in the teaching and learning process of preliminary design of reinforced concrete columns by Architecture and Urbanism students. The study was conducted with a class from the “Concrete Structures II” course at the Federal University of Santa Maria (UFSM), divided into two groups: one with access to AI and the other using only traditional materials. The analysis combined students’ technical performance in the practical activity with perceptions collected through pre- and post-test questionnaires. The results showed that only a minority of students rated AI as “very useful” (11.11%) or “useful” (11.11%) for the specific task of column preliminary design, while the majority considered it “slightly useful” (55.56%) or “not useful” (22.22%) in this context. Paradoxically, however, the group that used AI achieved better technical performance, with a higher number of correct answers compared to the control group. Although some students used the tool to review concepts and convert units, difficulties in prompt formulation and the lack of restrictions on accessed sources compromised the effectiveness of responses. It is concluded that, without prior training and teacher mediation, generative AI is not effective as a central teaching strategy. Nonetheless, three key guidelines were identified for its responsible integration into technical education in architecture: prior student training, use of AI in controlled environments with course-specific materials, and active teacher mediation.