Applications of AI Large Language Models in Citizen Participatory Urban Planning : Using GPT-4o and Prompt Engineering

Abstract

Can artificial intelligence (AI) represent citizens in urban planning? Owing to the continuous development of large language models (LLMs) such as ChatGPT, LLMs are expected to represent the thoughts of citizens. However, urban planning involves not only spatial particularities but also the intertwined interests of various stakeholders, where both efficiency and equity are pursued. Hence, a citizen-participatory urban-planning model that utilizes LLMs to represent the diverse opinions of citizens must be developed. This study aims to investigate the applicability and potential of LLMs for citizen-participatory urban planning and identify future tasks. The methodology involves using LLM prompt engineering to conduct discussions between citizens and urban-planning officials in a specific community. The discussions focus on living service facilities, which is a subfield of urban planning, and the prompt-engineering approach is categorized into four levels. This classification reflects the principle that the more specific the prompt is, the more detailed the response is. Through discussions spanning various fields and prompts, this study analyzes the representation of citizen opinions based on the field and prompt composition, based on which relevant implications are derived. The prompt-engineering results are analyzed, and the degree of prompt implementation is evaluated. Two key findings are identified from the analysis. First, LLMs are shown to effectively represent citizens of a specific area during urban planning. Second, this study presents a methodology for effectively addressing local-community issues in urban planning using AI agents. These findings are crucial as they suggest an approach that can efficiently reflect diverse citizen opinions regarding urban planning by leveraging LLMs.