Abstract
Designing successful interactions requires identifying optimal designparameters. To do so, designers often conduct iterative user testing andexploratory trial-and-error. This involves balancing multiple objectives in ahigh-dimensional space, making the process time-consuming and cognitivelydemanding. System-led optimization methods, such as those based on Bayesianoptimization, can determine for designers which parameters to test next.However, they offer limited opportunities for designers to intervene in theoptimization process, negatively impacting the designer's experience. Wepropose a design optimization framework that enables natural languageinteractions between designers and the optimization system, facilitatingcooperative design optimization. This is achieved by integrating system-ledoptimization methods with Large Language Models (LLMs), allowing designers tointervene in the optimization process and better understand the system'sreasoning. Experimental results show that our method provides higher useragency than a system-led method and shows promising optimization performancecompared to manual design. It also matches the performance of an existingcooperative method with lower cognitive load.