LARGE LANGUAGE MODELS IN THE DESIGN PROCESS: A WORKSHOP ON THE POSSIBILITIES OF HUMAN-AI COLLABORATION THROUGH KNOWLEDGE FORMATION

With the rise of Artificial Intelligence (AI), Large Language Models (LLMs) are proving to be a valuable technology for enhancing the designer’s creativity in the design process (DP). For instance, LLMs can synthesize information, brainstorm ideas, and simulate user opinions. We analysed the implementation of LLMs in the early phases of the DP, the Discover and Define phases, to investigate AI’s impact on the designed output and on the collaborative sphere of the design team. To achieve such objectives, after a comprehensive literature review and an initial test of the technology, we organised a workshop, in which 86 design engineering students took part. They were required to redesign a medium complexity product, using ChatGPT 3.5 throughout the early stages of the DP, sided by the “Prompt-chaining Cards” tool. We assessed the participants' initial knowledge, biases, and attitudes towards the LLM, and tracked the evolution of these perceptions throughout the workshop. Moreover, we conducted a qualitative analysis of the generated data, the research paths taken by the participants in collaboration with AI, and the identified needs and traits of the fictional users. The workshop's results enabled us to elucidate the opportunities and implications arising from the collaboration between product design students and LLMs. In particular, the results highlight the importance of “prompting” the correct inputs to the AI and the need for a critical discussion and analysis of the AI-generated data within the design team. Additionally, incorporating LLMs into a design team creates a set of specific challenges to be addressed, particularly a lack of communication between the human and machine counterparts, that ask for a high level of awareness of the process, a mastery of the technology and strategic management skills to effectively take advantage of the AI. In conclusion, our knowledge contribution to the discussed topic is an “integration framework” intended to help designers maximize the benefits of using LLMs in the first phases of the DP while highlighting the technology's current limitations.