Augmented agency: approaches for enhancing the capabilities of generative architectural design processes

Over the past several decades, generative design methodologies have significantly enriched the possibilities and innovations of architectural design processes, catalyzing unprecedented tectonic manifestations and spatial qualities. However, this design paradigm still faces a series of limitations preventing its extensive application in design practice when dealing with sophisticated design intentions and unpredictable design scenarios. Simultaneously, the explosive development of emerging technologies, such as Deep Learning, offers enormous and unexplored possibilities for the design methodology driven by technological innovations. Thus, the dissertation critically reflects on generative design’s inherent constraints and its correlation with emerging technologies in Artificial Intelligence (AI), which aims to push the boundaries of the methodology for achieving enhanced and previously unattainable performances. The research is positioned in a community of practice and academy focusing on generative methodology, algorithmic techniques, digital fabrications, and Machine Learning applications. Based on the main concerns and intentions, the research adopts a mode that combines practice-based research with speculative explorations of unfamiliar technologies. Specifically, the speculative explorations involve critical reflections on the inherent limitations of generative design and analysis of its correlated characteristics with AI technologies. The consequent emerging concepts and design approaches are applied to structure or augment the design processes of experimental and collaborative projects conducted in the RMIT Architecture Tectonic Formation Lab for practical validation and optimization. The dissertation establishes a methodology entitled Augmented Agency, which includes a theoretical argument of generative design’s Augmented Inclination, two original design approaches, and a series of forward-looking research directions. The argument of Augmented Inclination reveals two conceptual shifts. The first occurs in the overall intentional tendency of applying generative design, from the exploration of unprecedented generative systems to the enhancement of their capabilities and performances in practice. The second shift refers to the functional role of Machine Learning in generative design, from a generator to an augmenter. The two perspectives are proposed based on a series of propositions that encompass reviews of the evolution of generative design, identification of its inherent limitations, and its correlation with different types of Machine Learning frameworks. As one of the approaches, Physical Adapting employs a deductive method to augment generative systems’ capacity to accommodate various fabricating and material criteria. It embeds these physical requirements within a synthesized generative process negotiating with multiple hierarchies of design intentions, significantly improving the fabrication quality with broadened design possibilities. Furthermore, the dissertation proposes a Reinforcement Learning (RL) based design approach, defined as Intuitive Cultivating. It trains a generative system to form intuitions and makes optimal generative decisions by observing real-time conditions. Unlike the conventional single-run generative process, the emerging approach adopts a mode of comprehensive training and universal application, which augments generative systems with superior capacity accommodating sophisticated design intentions and unpredictable environments. Additionally, two potential research trajectories, Collaborative Intuitions and Intuitive Robotics, are outlined in the dissertation. They elucidate the possibility of applying the above-mentioned Artificial Generative Intuition concepts to the collaboration of generative systems and robotic fabrication, respectively. The critical reflections, emerging concepts, and approaches presented in the dissertation broaden the existing knowledge framework of generative design methodology. Specifically, the Augmented Inclination argument with the two conceptual shifts identifies the core concerns for the future development of this methodology. The Physical Adapting approach expands the possibilities for generative systems and enables them to accommodate various fabricating methods to achieve unprecedented geometric complexity. Intuitive Cultivating is a groundbreaking approach that systematically introduces Reinforcement Learning into generative design processes. As an emerging design paradigm, it offers solutions to address the existing limitations of generative design and enables generative systems to be more widely and efficiently applied to sophisticated design scenarios. From a macro perspective, the exploration of Machine Learning (ML) in the dissertation has significant implications for bridging the architectural discipline with cutting-edge technological achievements. Unlike its straightforward applications[ The straightforward ML applications refer to the research that directly applies AI techniques with specific usages to the architectural field—for instance, the research employing GAN-related algorithms for generating architectural images, Etc. Section 1.3 articulates critical reflections on these AI-applying methods.] to design practice, the research systematically analyses different ML frameworks' mechanismic characteristics and correlations with architectural design logic. These reflections establish a cognitive foundation, which contributes as guidelines and insights for the researchers attempting to explore a broad range of ML applications in the architecture discipline. Ultimately, Augmented Agency catalyzes the emergence of a more capable and intuitive design methodology, which signifies a closer correlation between humans and computational intelligence.