Forum Bauinformatik 2025
26.09.25, Aachen, Germany
On September 26th, 2025, Research Associate Max Zorn will present his work on trust-enhancing visualizations at Forum Bauinformatik 2025.
Abstract:
Design Space Exploration (DSE) tools enable designers to assess numerous design variants without the extensive computational demand of traditional simulation by utilizing Machine Learning (ML) methods. These advanced ML-based tools empower designers by providing real-time feedback and allowing interactive evaluation of promising design candidates. However, the inherent uncertainties and potential errors associated with such models often lower the designers’ trust, hindering a broader application in practice. This paper presents the application of Gaussian Processes to model and visualize the uncertainties inherent in ML-based DSE tools, evaluating whether visualizing the model uncertainties, specifically through confidence bounds, positively impacts designer trust. We conducted a pilot within-subject user study with architectural designers, investigating how their understanding of these uncertainties influences trust and decision-making. We compare three visualization modes with different levels of embedded information and assess the designers’ perceived trust, decision-making confidence, and overall usefulness of each visualization in supporting exploratory tasks. We find that designers prefer visualizations that include uncertainty information, reporting higher trust, confidence, and decision-making effectiveness compared to modes lacking such cues.