Some characteristics are present in Typogenetic Design that make it an interesting tool for designers as an intelligent design system. The design system adequately supports decision
Figure 6.4: Progression of design system properties
making in architectural design by incorporating integral properties of creative aspects. In contrast to responsive design systems that consist of hard-coded rules determining pre-programmed reactions during specific design cases and site conditions, an intelligent design system possesses adaptive characteristics. By using learning algorithms to flexi- bly adjust to a variety of expressions, conditions and choices, the geometric expression adapts to designer preferences in real-time. In this section, I describe the characteristics of Typogenetic Design as an adaptive system for aesthetic decision support.
Adaptivity The adaptivity of Typogenetic Design is of manifold nature. On the one hand, evolutionary search adapts to the criteria and constraints that define the design case and the restrictions of the chosen site. On the other hand, the learning algorithm, as part of the Aesthetic Support System, adapts to the preferences the designer exhibits in choosing some design solutions over others. The system shows properties of an intelligent system that “senses, reacts to, learns from, and subsequently adapts its behaviour to its present environment.” Atmar[1976, p. viii] This intelligent behaviour, learning functionality and adaptivity only lead to an intelligent system if the underlying generative engine is non- deterministic. As Fogel states inFogel[1995, p. 22], no learning behaviour arises from an apparatus with predefined reactions to external stimuli. As a consequence, the rule-based system in the adaptive framework occupies a central place in regard to the emergence of intelligent behaviour in the overall system.
Non-deterministic Behaviour One prerequisite for non-determinism in the design system is the presence of conflicting criteria in the design task, so that no single solu- tion could solve the optimisation problem. This characteristic is inherent in most design problems when there are no clearly defined problems that could be solved directly. The processes that lead to non-deterministic behaviour include the evolutionary operators which generate a variety of designs from the simple rules of the GD by mutation and crossover. Two other layers increase the degree of entropy in the system: the guidance mechanism inside the Aesthetic Support System and the emergent behaviour of the shape grammar itself. The resulting HITL system merges the capabilities of human and machine to extend the non-deterministic behaviour of generative algorithms during iterative design generation. Therefore, an oscillation of divergent and convergent behaviour of the system was observed.
Subjectivity As creativity is subjective by nature [Kazerani 2014], the interactive ap- proach of IC is a prerequisite for the ability of a designer to fully engage with architectural optimisation. In addition, the interactive guidance mechanism allows the designer to steer the computational design process in a creative and intuitive way by opening up the possibil- ity to direct the search. Therefore, internalised heuristics, implicit and explicit knowledge can be applied to the system’s evolution without changing the mode of interaction. By mixing well-performing and subjectively desired shapes in the selection procedure, it is likely that the solutions which emerge will fit the defined performance criteria and sub- jective design intent which the designer follows during the directed search. Repeating the set of symbiotic activities over and over again leads to a negotiation between AI and the designer.
Iterative Nature A collaborative communication between both symbiotic partners - the AI and the human designer - transcends the mechanistic analysis-synthesis-evaluation model typically used to describe optimisation processes, as defined by Radford and Gero [Radford and Gero 1988]. Even the similar design model offered by Kalay Kalay [2004, p. 10] falls short of describing the communicative process to its full extent. Therefore, the iterative process of evolutionary design also includes a collaborative aspect which must be discussed further elsewhere. At this stage, the sender-receiver model put forward by Claude Elwood Shannon [Shannon and Weaver 1949] leads to an understanding of how both partners in the symbiosis of Typogenetic Design benefit from the abstraction process of encoding a message for the conversation partner. A systematic decoding of the classifiers that support the decision making that occurs during the iterative design process, for example, might provide input for AI to learn about the decisions of the designer. On the other hand, a pareto-front clustering 8 approach might be an efficient way to provide
8
Usually solutions are clustered based on a similarity in fitness. In the case of architectural design, a spatial fitness measure seems more appropriate. It would allow the machine to cluster solutions based on
the knowledge generated about efficient shape generation by the AI as design input. As the iterative aspect in Typogenetic Design is obviously present in in the feedback loop, I shall proceed to a summary of this section, as presented in Figure6.4.
After reflecting on the wider philosophical background of the project work on and drawing out some of the implications of Typogenetic Design, I proceed to discuss the results of this PhD study in the following chapter.
their geometric similarity. Basing this process on a pareto front already provides a set with an equal level of fitness.
CHAPTER
7
Discussion
“Philosophy, art, and science are not the mental objects of an objectified brain but the three aspects under which the brain becomes subject.”
–Gilles Deleuze
In this chapter I discuss the results generated by observation, investigation and ex- plorations conducted during this PhD study. A variety of theories, technologies and communities of practice were explored during the progression of the research and finally lead to a coherent understanding of the research through the iteration of the emerging cre- ative ideas in different contexts of innovation. The collection of projects used to generate and showcase the results of this dissertation are presented in Figure 7.1. This overview diagram shows the projects in their respective position in the overall research agenda. I divided the projects in two different trajectories - the breadth and depth of the study - to emphasise the instrumental character of distinct explorations. Projects in the breadth section were used to provide a wider perspective and investigate a set of phenomena, mech- anisms and apparati in connection with an application case. During the projects that I subsume under the heading of ‘depth of research’, I translated generated knowledge into the problem area explored as a focus during this PhD study - interactive generative design for additive manufacturing. Investigating the creative ideas circulating in the projects in this application case are revealed in the following discussion, along with theoretical work produced during this PhD research.
7.1
Tectonic Articulation
In this section, I discuss the results of the Smart Nodes Project and the associated activities that involved analysing architectural optimisation as the first stage towards my emerging architectural practice.
Figure 7.1: Project position in overall research
As the first step in my research project, I analysed the practical issue that I was looking at: architectural optimisation. A literature review led to the identification of an applied potential of human-in-the-loop technology in the aesthetic decision-making process. Architectural optimisation was focused generally on quantitative criteria and on convergence toward one solution. In contrast, architectural design often incorporates qualitative criteria, involves tacit knowledge in the decision-making process (especially in the context of aesthetic evaluation) and usually explores a variety of design solutions to synthesise ideas and solutions. Consequently, I understood that the application of optimisation in architecture is often disconnected from the creative process of designing solutions. An adaptation of architectural optimisation to the characteristics and features of creative processes in architectural design was necessary. Introducing a process for initial morphological search directed by the decisions of the architect during exploration of shape spaces was the result of a range of experiments and case studies. The integration of interactive mechanism as part of the fitness function along with performance criteria was an idea that led to Typogenetic Design. The significant invention of interactive real-time browsing capabilities for emergent representations supports designer’s creative processes and decision-making by offering novel ways of creative input. Earlier, I was focused on reviewing the technological aspects of the optimisation process to get a multi-faceted understanding that could be used to extend the capabilities of designers to match the design potential offered by additive manufacturing. Those explorations shall be discussed
in the following.
During Smart Nodes Project a variety of tectonic expressions for the articulation of structural nodes were explored and tested in physical prototypes. The most promising node design for the understanding of the manufacturing, fabrication and construction constraints of additive manufacturing was tested in a full-scale prototypical structure. By integrating the node design in the structural simulation of the overall structure, the embedded indexical knowledge1 was used to generate custom-optimised nodes for specific contexts. This context-specific approach to node design linked the local dimensions and proportions of the structural nodes with the spatial configuration of the structural system. The systemic-spatial language used as part of the Smart Nodes Project combined the physical-technical aspects of additive manufacturing. Those aspects were explored on the local level of node design. In this way, the overall structure of the UABB pavilion was fixed. The architectural design team guided and orchestrated the node design in- vestigations. As a result, the engineering-based exploration was strongly committed to the spatio-morphological communication of the structural nodes as ornament. Thereby, the communication of structural trajectories, fabrication constraints and constructive sys- tem to the design audience integrated a cognitive-social function. The design potential of additive manufacturing was consequently expressed in the structural nodes.