AI'nspired: Data for evaluating AI-generated vs. web images as sources of design inspiration

The resource contains data from an experimental study involving a design task supplemented by three questionnaires. The experiment generated a dataset comprising 26 tram designs. These projects were developed as part of a ”Trams of the Future” competition organized by the Faculty of Architecture at Poznan University of Technology, Poznan Technological Institute Łukasiewicz, and Modertrans. The design task was undertaken by 52 participants, working individually or in pairs. As part of the design process, participants were required to gather inspiration from two sources, selecting up to 10 images from open online sources and up to 10 generated using the Midjourney V6 tool. In total, 1050 images were gathered for analyses (474 web inspirations, 417 AI inspirations, and 159 final project images). These are available in the AI'nspired images.zip file. Precomputed color, contrast, Resnet embedding, and DINO embedding similarities between inspiration images and final designs images are available in the AI'inspired image similarities.csv file.

Participants also completed two questionnaires detailing their experiences with AI tools and their preferences for the images used as inspiration. The first questionnaire was completed before the projects started, whereas the second one was completed after submitting the final designs. Finally, a separate survey was conducted among 33 experts in architecture, interior design, industrial design, and visual arts. The experts filled out an online questionnaire to rate the tram designs in terms of creativity and innovation. Data from all three surveys (along with group identifiers for matching images to surveys) are available in the AI'nspired surveys.zip file. The surveys are in two languages: Polish [PL] - the original language in which they were gathered, and English [EN] - translations of the original survey questions and answers.