2# EMERGENT FEATURES: Architectural Geometry
Part Two of My Series on Architectural Form in the Age of Artificial Intelligence
The question is: how to avoid a drift towards determinism when applying numeric methods in computer-aided architectural design? I once began my research by investigating the potential of contextualization as an effective tool to avoid determinism in numeric design methods. Of course, I was neither the first nor the only one addressing this issue.
Over centuries, the field of architecture has been cultivating a method which is still the most common strategy used for non-deterministic design. This method uses collected typologies of exemplary architecture² lexically to support decision-making processes without strictly imposing design rules. For architects who strive to avoid the imperative character of design automata, these quotable sources of inspiration have gained importance to the point of becoming indispensable. The method of compiling the necessary typologies and their role in the design process has been thoroughly investigated over the few last years (Abel, 2014). The central problem of these compilations and topologies is the manual work required to collect and sort the necessary data. The decisions made during this process naturally affect the knowledge architects can acquire when working with these compilations, and thus they affect the design process itself. Consequently, many architects and architectural researchers have begun to compile and publish their own typologies. As a result, the present abundance of compilations and typologies increases the risk of architects being misled by subjective opinions.
The main limitations of manual typology creation, the data volume on the one hand and subjective sorting strategies on the other, have given rise to different approaches to automating this very process. The proposed need for quasi-objective typologies emerged at the border area between architectural typologies and computer-aided design methods. This need has been investigated within CAAD for many decades now; as Christian Kühn observed already in 1995 (Kühn, 1995). In this sense, architecture retrieval became an alternative to the popular design automata paradigm. And once again, the functionalist perspective plays an important role in this endeavor.
Historically, typologies in architecture were not only built on the concept of “styles”. They primarily used the building’s function as the dominant category for classification. These brute approaches distinguish between sacred and secular buildings and their potential subcategories. Thus, the position of a building in the typology, which ultimately describes its similarity to other buildings, is determined by the function it serves. Evidently, this is once again nothing less than “form follows function”. The digital equivalent of this strategy can be found in the CBR¹ movement of the late twentieth century. Here, the detailed descriptions of functions were used to retrieve architectural elements from databases. In this way, the strategy propagated a latent functionalism, because the retrieval query was reduced to functional descriptions. Furthermore, as in functionalism itself, the description of functions was soon substituted by more abstract performance figures, thus allowing for the data upon which such retrieval systems are based to be created automatically. Instead of typologies, digital indices currently continue the functionalist tradition of describing architecture solely according to computable performances. The most recent works of Schneider (2016) and Dillenburger (2016) exemplify this way of defining architectural similarity by similar arbitrary performances regarding physical or human perception simulations. The mannerist core of functionalism, which also entails being able to derive the function from the form, was thus continued almost seamlessly. As such, architects’ critique of functionalism still holds for these digital retrieval approaches: needing to reduce architecture to its performance necessarily implies that there is more to architecture than
pure performance to begin with.
Post-modern theorists took the same line when focusing on the symbolic dimension of architecture in their counter-movement. They postulated the symbolic significance of certain architectonic forms, which is an aspect of architecture post-structuralists later tried to deconstruct. Following an Aristotelian tradition, these deconstructivists wished to free architecture from possible symbolic subtext by focusing on the processes (Standfest, 2013) instead. Hence, they argued from a meta-functionalist perspective. In addition, it was not enough that they continued to focus on function: even the dialectic between the symbolic meaning of architecture and functionalism was only virtual. The problem with this dialectic stems from the way in which post-modern architecture used the relationship between symbolised meaning and architecture to impose a language like formality on the design process. They interpreted the symbolic dimension as the dominant function behind architecture. While the discussion about the importance of this symbolic dimension has occupied architecture for decades, all participants in this debate nonetheless agree that architecture has a symbolic character. State-of-the-art methods in retrieving architecture miss this perspective when they index solely by arbitrary performance indicators which cannot describe this cultural aspect. These methods do not succeed in this regard because effective computational strategies to solve this problem of semantic interpretation have only recently been developed. Thus, when confronted with a choice between manual semantic labelling and automatic performance evaluation, the latter typically won out. This is a problem with which other domains have already dealt: machine translation, image recognition, and audio processing have successfully managed to fill the gap of automatic semantic interpretation in the last few years.
This sparked my personal motivation to use these implementations’ shared underlying principle to approach the same gap in automated architecture retrieval. Stay tuned for more. (This is the second edit of “Applying Emergent Features or Architectural Geometries” which I try to put online on a monthly basis.)
¹CBR, Case Based Reasoning
² I have thoroughly investigated this drift in a previous publication “Karten statt Kurven” (2011). There, i used the Actor-network theory as a framework for interpreting the chain of reference in parametric design. Hereby the concept of drift as introduced by Bruno Latour described a way of how non-human actors influence human ones (Latour, 2002).
Abel, G. (2014). Sammlungen als Epistemische Objekte und Manifestationen von Ordnungen des Wissens. In U. Hassler & T. Meyer (Eds.), Kategorien des Wissens (pp. 109–132). Zürich: Institut für Denkmalpflege und Bauforschung der ETH Zürich and Zürich : Vdf Hochschulverlag an der ETH Zürich.
Dillenburger, B. (2016). Raumindex: Ein datenbasiertes Entwurfsinstrument (Doctoral Thesis). ETH Zurich, Zurich.
Kühn, C. (1995). Der Begriff der Architekturtypologie und seine Bedeutung für die Theorie des CAAD(Doctoral Thesis). ETH Zurich, Zurich.
Latour, B. (1999). Pandora’s Hope:Essays on the Reality of Science Studies. Harvard University Press, Boston.
Schneider, S. (2016). Sichtbarkeitsbasierte Raumerzeugung: Automatisierte Erzeugung räumlicher Konfigurationen in Architektur und Städtebau auf Basis Sichtbarkeitsbasierter Raumrepräsentationen (Unpublished doctoral dissertation). Bauhaus University Weimar, Weimar.
Standfest, M. (2013). Architektur als Prozess. In M. Blaschitz, E. Hemmrich, & K. Köglberger (Eds.), Immer noch lernen von Las Vegas (p. 24). Wien [u.a.]: Springer.
