Adaptive Optimization of Subjective Design Attributes: Characterizing Individual and Aggregate Perceptions

Nandy A and Goucher-Lambert K. 2023. Proceedings of the ASME International Design Engineering Technical Conferences (2023).


Subjective attributes play a significant part in the assessment of user-facing products. Unlike performance requirements, these quantities are best evaluated through human feedback. While people share commonalities in their evaluations, allowing personalization when quantifying these subjective attributes may improve the alignment between computational and human representations of design information. We investigate this topic through a study in which participants (N = 56) make a series of pairwise decisions between parameterized mugs, and indicate their perceptions of how comfortable each is to hold. Interactive Bayesian optimization is used to adaptively arrive at a design that optimizes this subjective quantity. Participants guide the model through only their own decisions or make decisions using a model that has already been trained with simulated data (N = 25) or data from the real decisions of other participants (N = 31). The resulting designs are evaluated across the different cases, showing the impact of capturing individual and aggregate perceptions of subjective quantities. The findings imply that balancing aggregate and individual-level decisions within models simultaneously results in the best alignment with human perceptions of subjective attributes. Further implications for design include the potential for personalized control over subjective attributes for designers, users, or users-as-designers.