A Review of Generative Artificial Intelligence-Assisted Scenario Development Approaches in Landscape Architecture

In this article, we offer a comprehensive review of the evolution, progress and current state of Artificial Intelligence supported scenario generation research in Landscape Architecture. This technological development is enables a deep paradigm transformation centred on the shift from designing static “objects” to fostering dynamic “processes” and “systems” capable of self-adaption and evolve over time. The authors cover key generative methods such as Parametric Generatives Generative Design, focusing on rule-based optimization, Generative Adversarial Networks which leverage training via competition between networks and Text-to-Image models that generate instant conceptual visualisations. These are not merely tools to make artefacts of the design but sophisticated instruments for informed site analysis and performance-based optimization—the results of which involving hundreds if not thousands of design alternatives within minutes, are captured and tabulated efficiently. In addition, the study helps to understand crucial concerns regarding creativity, algorithmic bias and transparency. It emphasizes that Artificial Intelligence systems have the potential to exacerbate human biases present in training data and that these models can be so opaque as to create significant trust issues. In order to tackle these challenges, the paper calls for Explainable Artificial Intelligence techniques that provide interpretability, as well as User Interface/User Experience design tactics to minimize cognitive biases such as Confirmation Bias and Automation Bias. The dialogue progresses to discuss more sophisticated notions of Reinforcement Learning and World Models, in which Artificial Intelligence plays an active role in the design of dynamic, adaptive and regenerative landscape systems.