Collaboration between two large language models for design concept generation

Abstract Recent advances in generative artificial intelligence, especially in large language models (LLMs), offer new opportunities to overcome the major challenges, such as human cognitive and creative limits, in design concept generation. Research results show that a well-trained LLM is able to generate design concepts as good as, or sometimes even better than, human designers. However, design concept generation using one LLM also suffers from several limitations, such as a lack of diversity and novelty in generated design concepts. It is well known that human collaboration in design concept generation drives higher quality and more original ideas by fostering serendipitous insights and diverse perspectives. However, it is not clear whether LLM collaboration can lead to similar outcomes and overcome the limitations of design concept generation by an individual LLM. This research assesses the effectiveness of collaboration between two LLMs in design concept generation through two design studies. Three representative LLMs are first employed to generate design concepts for two engineering design problems, and then they are asked to improve their own design concepts and the design concepts generated from the other two LLMs, respectively. The results of these studies indicate that the collaboration between two LLMs improves the novelty and variety of the initial design concepts generated by an individual LLM. Among all performance metrics evaluated, variety benefits the most from the LLM collaboration. These findings suggest that collaboration among LLMs can serve as an effective tool for concept generation in the early stage of product and system design.