Human-in-the-Loop co-Design

This study aims to enhance understanding of the performance of human-learning HITL approaches in human-AI co-design processes. The conversation mode of human-learning HITL in human-AI co-design processes and usage patterns performance in AI-learning HITL were explored. Subsequently, the effects of the two different HITL approaches on human-AI co-design processes were investigated. A total of 204 participants were recruited to complete two design tasks using human-learning HITL and AI-learning HITL approaches in human-AI co-design processes, respectively, with the help of AI. Thirteen judges were recruited to assess the solutions generated from the design tasks. The results of this study revealed that the human-learning HITL approach achieved significantly higher performance than the AI-learning HITL approach in novelty and esthetics. The AI-learning HITL approach showed higher mean scores in environmental friendliness. AI can support designers by providing additional information and resources during design processes. These findings highlight the importance of human input in enhancing creativity-related aspects of design, as well as the complementary role of AI in human-AI co-design processes. This contributes insights for developing more effective AI tools and guidelines to support human–AI co-design processes. In addition, the study identified that people may have difficulty expressing their entire requirement in one round of conversation with AI. Designers are more likely to use funneling conversations and exploring conversation modes to communicate with AI during the human-AI co-design process. These findings allow developers and designers to consider how to develop more efficient and effective guidelines for communicating with AI in human-AI co-creation.