The Complements of an AI Architecture
Buildings are primordial forms of Artificial Intelligence. They are artificial to human thinking; buildings complement and compute what people can’t: airflow, temperature, shade, shelter. Architecture has always been a mineralogy of artificial thinking. Buildings compute with their programmable doors, gates, or passages. Autonomous from program, ownership, and use, buildings do not contain but compute. Older, and not bounded to electric circuits alone, information forms with anything that bonds, links, or joins. Gens transmit biological information; memes carry cultural information; walls bear building information.
With the overview of a historian, Yuval Harari identifies as artificial intelligence that what is artificial to the ingredients of its medium. Take language: the intelligence of culture, or simply memes, are artificial to the words they are written in. Today, memes belong to the internet; they can’t be authored by one subject alone. Artificial to its basic grammar, words, and language, memes are built from their intersubjectivity. Intelligence is its correlates. Computers correlate from data, and buildings correlate from its elemental physics. By fusing through networks, graphs, trees distributed elements AIs think plural.
For an architect, thinking plural is the uttermost unthinkable. On the opposite, limited to the capacities of a single rational, human-centered architecture evolved as a history of notations that aimed to reduce pluralities. The notational tools of an architect, from axes, symmetries, grids, types, to models reduced the number of individual parts until one could cast and build with single objects only. Incapable of thinking space plural, a humanistic architecture fragmented space into single snippets filled with single purposes. Space framed projections of a centered human subject. Subsequently, by chaining space through time, or better the architect’s path, buildings depleted to sequences of singular programs. However exceptional or complex, it staged one view only. The notations of an architect couldn’t render all views at once. The resulting architecture framed, limited, marked, contained, or expressed a series of single moments. But it could not compute. Stripped away from their capacity to compute, buildings today can barely be reprogrammed or reused. Most buildings even can’t house multiple programs. Consequently, buildings’ lifecycle becomes ever shorter, empty real estate increases, and urban fabrics further expand.
Now, Artificial Intelligence complements existing notations with plural capacities. With its capability to design consensus, like the memes of a crowd, AI creates new ways to interact with institutions, which buildings are, synthetically. Augmented by Artificial Intelligence, buildings turn into an ecology of building parts as open worlds. Unframed, the plurality of entities remains independent but can communicate with each other. Recognizing multidimensional relationships, Artificial networks propose in higher variability more performative buildings. Simply by inference from plural reasoning. Departing from all possible variants of a plan, from a panorama of all perspectives, or the datasheet of every smartmeter at once, those models build artificial intelligence as an inter-subjectivity between its data. Translated into architecture, the peer-to-peer propagation of a network turns into the physics between part-to-part. The whole of a building turns into a fused consensus between its building parts. Through its plural inference, it is possible to form figurations that are already explicit with the attributes of their parts. The whole is projected as a specific part for the part as a whole. When wholes and parts are nested into each other, orders are turned upside down. Collapsing into particular qualities of a part, the previous grids, axes, symmetries convolute into linkages between building elements. Beyond geometrical elements, parts superpose building figurations within their figurative qualities. One to one in so many ways, it is the scale of a part where artificial and human design interlinks. Augmented with regulations and structural performance, computable parts can be carried, assembled, and figurated quickly. At people’s scale, parts carry loads, temper space, and compute seats, stages, doors, and passages. Profoundly partly, parts depart from their context in the same way as they situate their context. Fundamentally recursive, parts have to convolute their environments.
Mereologies
Borrowed from philosophy, the term Mereology, literally meaning the study of parts, offers an umbrella to gather architectural knowledge around parts. Mereologies begin by filtering contexts in search of those particular physics of spatial computation. Part design deviates from sampling spatial entanglements. What is a node when a building turns into a network? What physics captures which data? Previous building notations reduced peer-to-peer relations between building parts into extra artifacts, like staircases, facades, grounds. What articulation can we achieve, when we maintain resolution one to one? Ultimately partial, mereologies value parts, not through their geometrical figure but their figurational capacities.
In an iterative design workflow, the Mereology itself becomes an architectural notation by drawing parts not as a singular figure but in its plural assembly. Not to be confused with a building, the assemblage of a mereology displays all combinatorial complexities of a figure at once. Mereologies as architectural drawings represent the plural characteristics an artificial network computes with. By exhausting all computable complexity of a part, they collapse all states within that part. In that way, the drawings open a recursive interplay between the design of a part and its computability.
Mereologies as a project
Inherently, predictive models that source from big data assume that beyond a certain threshold, any variance is given. Within such a practice, anything that claims to be new is already exhausted within a given set. Consequently, any building arrangement, however complex, is equally easy to plan, construct and use. Once rationalized, captured as data, linked, and shareable, any geometrical expression comes for free. When architecture is automated and dwells in information, value is not anymore distributed through differentiation but equates with the capacities to partake. Here, we encounter a shift from the distributional frameworks of Postmodernity towards the rationalization of computability by platforms. When a building is not anymore the machine to live in but to compute with, the spatial figuration of a building is valued by the degrees of indeterminacy; by its capacity to be indifferent to multiple rationals while partaking. Without care to the actual content but to shareability only, computational architecture is ultimately plural.
About
Daniel Koehler is an architect, urbanist, researcher, and co-founder of lab-eds. He is an assistant professor for architecture computation in the School of Architecture at UT Austin. Before, Daniel researched at the Bartlett in London and Innsbruck University, where he completed his Ph.D. He has taught at several institutions, among them the Aalto University in Espoo Finland, Vilnius Academy of Arts in Lithuania, and the University of East London. His work has been exhibited in Prague, Milan, Venice, Graz, Montreal, London, and is part of the permanent collection of the Centre Pompidou in Paris. He is the author of “The Mereological City”, a study on the part-relationships between architecture and its city in the modern period. His current research focuses on the urban implications of distributive technologies, which are being designed by means of sets, data, interfaces and their architecture.
Lab for Environmental Design Strategies is a research-based architecture studio. The studio architects through computation towards ecological forms aiming for a societal, cultural friendly environment. The studio was founded 2008 by Daniel Koehler and Rasa Navasaityte.
Large City Architecture Research is an umbrella for design research on urban form through computation. It was also a Research Cluster at the BPro Bartlett School of Architecture, UCL London from 2016-19. Led by Daniel Koehler and Rasa Navasaityte.
The work shown here are credited to:
WanderYards by Chen Chen, Genmao Li, Zixuan Wang, BPro AD RC7.
Enframes by Kexin Cao, Yue Jin, Qiming Li, BPro AD RC17.
iiOOOI by Sheghaf Abo Saleh, Hua Li, Chuwei Ye, Yaonaijia Zhou, BPro AD RC17.
Blockerties by Junyi Bai, Anna Galika, Qiuru Pu, BPro AD RC17.
Enfilade by Mengshi Fu, Ren Wang, Chenyi Yao, and Zhaoyue Zhang, BPro AD RC17.
Endscape by Yao Chen, and Zhaofeng Chen, BPro AD RC17.
Slabrose by Dongxin Mei, Zhiyuan Wan, Peiwen Zhan, and Chi Zhou, BPro AD RC17.
Comata by Anthony Alvidrez, Shivang Bansal, and Hao-Chen Huang, BPro AD RC17.
Independent by Aldryn John Matias, UT Austin.
Selected Publications
Monography:
Koehler, Daniel. 2016. The Mereological City: A Reading of the Works of Ludwig Hilberseimer. Architekturen 36. Bielefeld, New York: transcript.
Papers:
Koehler, Daniel. 2020. “From Partitioning to Partaking, or why Mereologies matter”, in: Prospectives Mereologies, 1, 1, Bartlett School of Architecture, University College of London (2020).
Koehler, Daniel. 2019. “Mereological Thinking: Figurating Realities within Urban Form”, Architectural Design (AD), issue Discrete, 03.2019, ISSN:1554-2769.
Bai, Junji, Galika, Anna, Koehler, Daniel, Pu, Qiuru, 2019. “Distributed Ledgers and their Impact on Urban Form”, ACAADIA Proceedings 2019.
Koehler, Daniel, Abo Saleh, Sheghaf, Li, Hua, Ye, Chuwei, Zhou, Yaonaijia and Navasaityte, Rasa. 2018. “Mereologies – Combinatorial Design and the Description of Urban Form.” In Proceedings of the 36th ECAADe Conference, S. 2 vols, 85–94. eCAADe 1.
Koehler, Daniel. 2017. “The City as an Element of Architecture: Discrete Automata as an Outlook Beyond Bureaucratic Means.” In Proceedings of the 35th ECAADe Conference, ed. 2 vols, 523–32 1. Rome: eCAADe.
Koehler, Daniel, Navasaityte, Rasa. 2017. “Mereological Tectonics: The Figure and its Figuration”, TxA Emerging Technologies 2016 Proceedings.
Koehler, Daniel. 2014. “The Ecological Gap: Mereological Considerations on Architectural Typology.” IaacBits 4 (4.1.2).
aiarchitects.org 