AI-DRIVEN, EVIDENCE-BASED GENERATIVE DESIGN: A TEACHING FRAMEWORK FOR LANDSCAPE ARCHITECTONICS INTEGRATING RESEARCH FINDINGS, DESIGN PRECEDENTS, AND INTELLIGENT RECOGNITION AND JUDGMENT

Abstract

Artificial intelligence (AI) has entered design courses, but its use often stops at tool operation. Taking Landscape Architectonics, an undergraduate course in landscape architecture programs, as the vehicle, this paper builds a teaching framework for generative design that integrates four elements: research findings, design precedents, AI-based recognition and judgment, and AI generation. The core claim is simple. Teaching students to “use AI” means teaching them to run an evidence-constrained generative design chain. “AI-driven” therefore carries two meanings: at the tool level, AI executes generation; at the teaching level, evidence steers generation, and the work of teaching lies in leading students from the former toward the latter. Among the four elements, research findings and design precedents are static knowledge that travels across sites. Research findings supply criteria for what counts as good. Design precedents supply examples of what to do, distilled from 13 street design guidelines into 114 retrievable situation packages. AI-based recognition and AI generation produce dynamic outputs anchored to a specific site. Students mobilize these elements through three stages—reading the site, grounding the context, and giving form—and a chain of evidence disciplines each stage. The framework follows three design principles: complementary elements, partial embedding, and explicit judgment. Three classroom examples show how the framework operates. This paper focuses on framework construction and does not report learning-outcome data.