Towards Explainable and Responsible AI: Examining the Role of Large Language Models in Participatory Mixed-Use Zoning

Mixed-use zoning (MUZ) has emerged as a key alternative to address the shortcomings of conventional Euclidean zoning. Given the need for nuanced negotiation among multiple stakeholders, finer-grained considerations, and its multifaceted implications, participatory planning has attracted growing attention in the MUZ process, thereby demanding more sophisticated planning support systems (PSS). While existing PSSs, which integrate evolving computational techniques, have shown promise in supporting participatory MUZ processes, they often fall short of engaging diverse stakeholders as technical complexity increases. To bridge this gap, large language models (LLMs) hold significant potential. However, a contextualized examination of how to realize such advantages of LLMs while addressing related socio-technical concerns remains largely underexplored. Therefore, this research examines how different LLM application approaches, as well as the interplay between human intervention and LLMs, influence the generation of MUZ outcomes through a comparative experiment evaluating four distinct LLM-based models. The key findings reveal: (1) how differences in LLM applications shape the interpretation of stakeholder inputs and lead to divergent MUZ outcomes; (2) the potential of LLMs to facilitate more explainable uses of artificial intelligence (AI) in supporting MUZ; and (3) a responsible AI related paradox in incorporating human intervention into LLM-based models. Building upon the findings, directions for future research are also outlined.