Designing and Evaluating Landscape Proposals with Generative AI ChatGPT - Urban Plaza Design via Landscape Firm Style Emulation with Educational Potential -

This study empirically examines the process by which ChatGPT, a generative AI, autonomously learns and analyzes various landscape design theories, urban design guidelines, and the distinctive styles of renowned landscape architecture firms to generate and self-evaluate design proposals. By comprehensively referencing accumulated literature and design case studies, ChatGPT produces design proposals that reflect spatial contexts, green space arrangements, pedestrian flow, and publicness while incorporating the unique aesthetic characteristics of specific firms. Furthermore, based on domain expertise and relevant data, ChatGPT establishes six evaluation criteria—publicness and openness, design creativity, urban context and harmony, environmental sustainability, feasibility, and social functionality—and conducts autonomous assessments accordingly. The evaluation results were statistically aggregated and validated to assess the reliability and validity of AI-generated design evaluations. The analysis revealed that ChatGPTs design proposals, which emulate the styles of landscape firms, consistently received high scores for creativity and publicness. However, evaluations concerning practical aspects, such as budget, construction technology, and harmony with urban context, showed lower scores or significant variances. This suggests that implementing AI-generated creative designs in practice requires the on-site experience and comprehensive judgment of human experts. The study confirms the potential of generative AI to generate and evaluate design proposals using reliable literature and accumulated design data. The study also indicates that, in landscape design education and practice, generative AI can serve as more than an auxiliary tool, playing a pivotal role in expanding creative ideas and advancing design development.