Evaluating AI-generated design solutions in a basic design studio

The first-year studio in architecture programs plays a pivotal role in introducing novice designers to the complexities of design problems. Interpreting multifaceted design briefs and generating viable solutions can be challenging for novice designers, given their lack of prior experience. This study seeks a method to enhance students’ understanding in a basic design studio. By creating synthetic design solutions to the given problem definitions in a sample set of assignments, it enlarges the solution space to include a variety of responses to a design brief. Using assignment briefs from two institutions, text-to-image diffusion models were employed to produce diverse solutions that preserved the semantic organization of each brief. After an initial generation, expert feedback was incorporated, refining prompts and producing a second set of synthetic solutions, which were then evaluated alongside a control group in semi-structured interviews with design experts. This evaluation focuses on whether the explicitness of problem definitions and expert feedback separately and together impact synthetic solution generation. Initial findings indicate that AI-generated solutions perform in correlation with the brief definition. Diffusion models can rapidly generate a wide range of design solutions to briefs, particularly in the early stages of assignments. However, without feedback, later-stage solution spaces tend to be filled with arbitrary visuals. With expert guidance, synthetic solution spaces have the potential to expose students to a broad spectrum of solutions, enabling them to better interpret design problems, grasp key concepts, and develop critical perspectives on their design processes. These findings offer valuable insights into the role of AI in interpreting design problems and generating solutions, emphasizing its potential to enhance the comprehension and exploration of design problems in early design education.