Original (Dutch) version of this article      Artificial Architecture     IAAA      Remko Scha

[English translation of an article which appeared in Dutch in: Zeezucht 8 (1994). ]

Remko Scha and Eric Vreedenburgh

Towards a different architecture


Post-war housing has been characterized by uniformity in space and time. Large numbers of identically layed out blocks were built, consisting of large numbers of identical houses. And these houses seem to be designed for eternity: there is no room for change.

If we want to build in a more flexible and varied manner, the architect must give up the idea of being personally responsible for every detail of every building he designs. He must design at a higher level of abstraction -- specify rules and boundary conditions, and leave the concrete instantiation to the end user or to chance.

In contemporary visual art, this more distanced approach already has a rich tradition. The clearest example is algorithmic art, where complex, unpredictable processes are defined by means of completely explicit rules, which are executed consistently and precisely by the computer.

Inspred by the example of algorithmic art, this article will explore the possibility of computer-generated architecture -- the possibility to realise architectural designs as algorithms which do not specify individual buildings, but infinitely large classes of different possible buildings.

Positivist architecture.

Like all other cultural activities, architecture is first and foremost a matter of imitation and convention. A house, a department store or a museum, must look like other houses, department stores or museums. But at the same time (like in all other cultural contexts), the designer and his client often feel the subjective desire or the economic necessity to distinguish themselves from their colleagues. Therefore, buildings may be designed which are intended to be demonstrably different from their predecessors, or even demonstrably better.

The architect who introduces smaller or larger departures from traditional shapes, lay-outs, materials or construction methods, often indicates his reasons. Usually, these refer to the efficient realisation of the intended function of the building. Architecture thus seems a positivist discipline, explicitly articulating the goals to be achieved by a building, deducing its design decisions from these goals, and evaluating the decisions in terms of a purely economical kind of rationality. Buildings appear to be designed just like machines. The ideal house is a dwelling machine, in which every detail has a fixed, functionally motivated place.

Post-war housing displays a blatant discrepancy between ideal and reality. The analytical, goal-oriented approach, intended to yield provably optimal results, is completely counter-productive. A building's actual functions cannot be specified explicitly -- they only emerge in the interaction with the actual users, they keep changing in the course of time, and they need not be fixed unambiguously at any moment.

The functions employed by the functionalist architect are largely fictitious. The economical component of the functionalist method has therefore taken on a life of its own. The notion of a house was reduced to the specification of a minimal set of requirements, and economic reality equated the minimal description with the maximal one. Functionalism served as an excuse to eliminate "superfluous" elements. The serial production method of car manufacturing became the guiding example.

Postmodernist philosophy.

The hybris of rational thought that we observe here is equally widespread outside of architecture. That thinking tends to overestimate itself is almost inevitable: the rational mind would like the world to be an orderly and thereby controllable complex. For this reason we can never be sure where the boundary lies between an open-minded search for structures and an obsessive projection of them. But we do have some experience with this issue now. Numerous times we have witnessed that plausible-sounding ideas about methodology or social structures were implemented all too literally and flopped completely. This has been the fate of all explicitly formulated political philosophies, and of all scientific approaches which involved explicitly formulated methodologies (in fields like sociology, anthropology, psychology and linguistics). The failure of functionalist architecture and rationalist urban planning is just another example of this phenomenon.

The illusion of an individual subject that can articulate what it knows and justify what it does, has been a constitutive ideal in our culture. This over-estimation of rationality explains the increasingly comprehensive technocratization of the world. And it explains the hybris which is often displayed by newly launched artistic, scientific and social movements which can precisely explain their principles and methods right away, before they have produced any results.

But perhaps a turning point has come in this development. Philosophers like Jacques Derrida have popularized notions which allude to the boundaries of rationality, such as ebb-sense (the withdrawal of meaning and the meaning of withdrawal) and odder-ness (the importance of intrinsically elusive things which are incommensurable with our mental framework). Reality is different from what we know. Life is more capricious than the schemas which try to capture it. We therefore cannot control it. Who does not want to admit that, will find it out all the more painfully.

These themes also appear in recently developed branches of mathematics and physics which deal with unpredictability. Non-linear dynamics, for instance, shows that arbitrarily small deviations in the initial conditions of a physical system may give rise, in the course of time, to ever increasing variations in the resulting state. In the beginning of the twentieth century, physicists had discovered already that it is not possible to observe all details of a physical system precisely at the same time. This gave rise to quantum mechanics -- calculations with wave functions which only describe the occurrence probability of different states. Non-linear dynamics now shows that the quantized nature of matter is not the only source of uncertainty when we try to predict future states of a physical system. In classical, deterministic physics, we cannot predict everything either -- not even approximately. Our measurements of current states always have a limited accuracy, and in a non-linear system this produces an ever-increasing uncertainty about the future.

The issue is now, what we should do with such ideas. What kind of gay science is on the horizon if we make the dialectical step to a rational consciousness which does not assume an articulation of its rationality, to a subjectivity which can act sensibly without trying to control and predict everything. In dealing with this question, architecture may find inspiration in recent developments in visual art.

Kantian esthetics.

The insight that reality does not coincide with our conceptual models of it is not new. It was formulated with inescapable clarity by Immanuel Kant. And the idea that our rationality is more limited than our cognition, that we know more than we can articulate, was articulated quite explicitly by the early nineteenth century romantic artists and philosophers. Since those days, art defines itself against the increasing rationality of main-stream culture, by being the symbol of everything that rational thought ignores. Art celebrates intuition and direct experience.

Whose intuition? Whose experience? Some artists, and many collectors, critics and other fans would immediately point to the Divine inspiration of the artistic genius. But there are other traditions in contemporary art which emphasize the richness of all individual and collective human experience. Marcel Duchamp, for instance, was very explicit about this: "The spectator makes the picture."

Kant viewed the esthetic as a dimension of perception: perception which becomes conscious of itself, when the process of input interpretation does not yield a definite final result, but nevertheless creates a coherent experience. When Duchamp assigned the status of "art work" to existing readymade objects, he drew a radical consequence from Kant's point of view: that the input doesn't matter much, as long as the observer's process of esthetic reflection can take its course.

Kant himself already pointed out that works of art constitute sub-optimal input material for this process, because the artist's intentions deflect this process: they do induce a definite interpretation of the art work, which ends the interpretive process before the cognitive resonances which constitute the esthetic experience could have been built up. For Kant, the paradigmatic esthetic experience does not involve art, but natural phenomena.

Since Duchamp, increasingly many artists have accepted the challenge that is implicit in Kant's ideas: to create a non-intentional art, an art that can be experienced as a natural phenomenon. Several rather different artistic movements have developed procedures for generating art through more or less autonomous processes, initiated by an artist who would not be able to predict the final result: écriture automatique, action painting, physical experiments, biological processes, systematic, conceptual, and stochastic art. Sol LeWitt: "The artist's will is secondary to the process he initiates from idea to completion. [...] The process is mechanical and should not be tampered with. It should run its course." [1]

Perhaps the clearest example of this development is algorithmic art, where a process is defined by completely explicit rules, executed by the computer with extreme consistency and accuracy. By employing mathematical simulations of chance, the unpredictability of the outcome can be maximized.

Art is often viewed as a medium that an artist employs to transmit profound thoughts to his audience. But what an observer considers important or meaningful in an artwork is often independent of the artist's intentions. That a computer has no intentions at all, is thus no reason to doubt the possibility of fully automatic computer-generated art. Precisely the iron consistency and the relentless ardor of the inhuman computer yield results which people may find interesting.

Analytic art.

But algorithmic art did not come about as an immediate result of considerations in philosophical esthetics. There is another source, which is the tradition of analytical thinking about the structure of the image -- the kind of thinking we find already in the art and writings of the Renaissance painter/mathematician Piero della Francesca, and which in the twentieth century became one of the most important driving forces behind the artistic developments.

The pioneers of abstract art (Kandinsky, Malewitsch, Mondrian) and their disciples (such as Lissitzky, Rodschenko, Van Doesburg, Vantongerloo) always constructed their images by means of a limited repertoire of elementary building blocks and operations. It almost seems as if the individual images only served as means to discover an increasingly pure and sharp visual language. But the results of this exploratory process never became quite explicit, because, as it happens, painters make paintings rather than languages.

This tautology became invalid in the sixties, when some new painting movements were launched who employed all media except painting. It became increasingly popular, for instance, to make descriptions of visual situations, rather than actual paintings or sculptures. Sometimes these descriptions were not meant to be executed, but to define an art which would only exist at a conceptual level.

The nice thing about a description which is executed (a "visual score"), is that it can often be realized in many different ways. In that case, the artist cannot exactly predict the result of the execution of his work. He only fixed certain properties, but left other aspects to the "performer" or to chance.

If the description of an artwork or a class of artworks is specified by mathematically precise algorithm, this is called algorithmic art. In principle, a simple algorithm can be executed manually by a human person. But usually, algorithmic art is realized by a digital computer.

An algorithm may define a large class of different images with complete precision, for instance by indicating that all variations within a certain pattern must be enumerated systematically. Or, if the number of possible choices is too large to be realized one by one, the algorithm may indicate that random samples are to be drawn from a set of possibilities. In that case, every new execution of the algorithm may yield new images.

The algorithm is a "meta-artwork": the mathematical characterization of a set of possible artworks. The visual language of an artist is no longer implicitly suggested by an œuvre consisting of individual works. The language is explicitly specified in the algorithm which generates arbitrary examples from the œuvre.

Algorithmic art in the nineteensixties ties in with the analytical movements in early abstract art, but it defines visual languages which are less complex than those of Kandinsky, Malewitsch or Mondrian. The neo-constructivist chance art of François Morellet and Herman de Vries, for instance, employs algorithms which put a particular shape (for instance a square, a circle or a line segment) on randomly chosen positions on the plane. In fact, these algorithms were still executed by hand; random choices were made by throwing dice or by consulting random number tables.

Artificial [2]

The constructivist tradition was concerned with harmony and purity. Today, that seems a somewhat arbitrary and limited ideal. Expressionism taught us the esthetics of ugliness. Duchamp demonstrated the esthetics of indifference. The current challenge is an esthetics that encompasses everything: beautiful, ugly, and indifferent.

Art is not a means of communication. It is meaningless raw material, interpreted in an absolutely arbitrary way by a culturally heterogeneous audience. There are no serious reasons for wanting to make certain artworks rather than other ones. An artistic project that wants to face this issue, must avoid choices, transcend styles, show everything: generate arbitrary examples from the set of all possibilities.

An individual, spontaneous artist cannot live up to this challenge. What is needed, is a deliberate technological/scientific project, with a sensible division of labor between man and machine. Human artists/programmers should develop an algebraic definition of the space of all possibilities; the computer can then choose and display random examples from this space.

The image-generation project Artificial uses this approach to realize the Kantian ideal of an art without artists. The algorithmic techniques that Artificial deploys for this purpose are based on the neo-constructivist chance art mentioned above.

One of the prototypical algorithms in sixties chance art, for instance, puts tokens of a given visual shape on randomly chosen positions on the plane. A similar algorithm creates arbitrary closed shapes by combining line segments. These two algorithms can be combined in an obvious way, so that both the shape and the position of the image elements are determined at random. Other algorithms generate a multitude of different regular patterns or regular shapes; these can also be integrated. We may thus gradually abolish choice, by avoiding the exclusion of any choice -- by affirming every choice, and by putting it on a par with all other choices inside an all-encompassing probabilistic system.

The ultimate consequence of this approach would be a computer program generating all possible images, with probability distributions that yield maximal diversity. It will not be easy to develop this program. But the Artificial programs show that it is possible now to make significant steps in this direction.

Cage: "I was driving out to the country once with Carolyn and Earle Brown. We got to talking about Coomaraswamy's statement that the traditional function of the artist is to imitate nature in her manner of operation. This led me to the opinion that art changes because science changes - that is, changes in science give artists different understandings of how nature works." [3]

In their metonymic symbolizations of "chance", "nature" and "objectivity", the process artists of the sixties manifested a deeply felt emotion -- the desire for an art that does not originate from the whims of the individual, but from a deeper necessity. The project of a total, all-embracing chance art is now initiating the process of actually satisfying that desire -- by treating absolute randomness as the deepest necessity.

Cage: "Is man in control of nature or is he, as part of it, going along with it? [...] Not all of our past, but the parts of it we are taught, lead us to believe that we are in the driver's seat. With respect to nature. And that if we are not, life is meaningless. Well, the grand thing about the human mind is that it can turn its own tables and see meaninglessness as ultimate meaning." [4]

Artificial architecture?

Obviously, Artificial constitutes a very constructive contribution to contemporary autonomous art, where nothing is created any more without the painful awareness that there are no good reasons to make exactly this rather than something entirely different. Artificial explodes this impasse. It posits a stimulating technical challenge which takes our current relativistic insights seriously: to show everything.

In architecture, the same issue is at stake. In many situations it is inappropriate for the architect to force his individual taste upon others, but it seems almost unavoidable. Inspired by the Artificial approach to autonomous art, we therefore propose an architecture of chance, where the architect works at the right level of abstraction. The designer is no longer concerned with expressive details. He only defines the "rules of the game", which determine which situations are possible at all. Rules which specify the size of the playing field, which pieces are in the game, what groupings of these pieces are possible, which moves can be made -- just as in chess. In a particular context the architect can thus (if necessary!) make decisions about scale, rhythm, or the repertoire of applicable elements. In this way, a specific situation may get its own morphology.

Concerning the functional dimension of architecture: we view the function of a building as variable. We choose not to fix the function as the starting point of the design process. The concept of "function" is replaced by the concept of "potential". To take the potential of a building into account, chance architecture algorithms must be somewhat more complex than algorithms for autonomous art. To deal with constructive and functional aspects in an optimal way, they must be integrated with automatic design techniques from Artificial Intelligence. (Automatic design is one of the most successful areas in A.I. In designing VLSI circuits, for instance, programs are by far superior to human designers.)

The Palladio-machine.

The idea of rule-based architectural design is not new. It occurs implicitly in the very first theoretical essays about architecture that were written in our cultural tradition. Especially relevant in this context is the rediscovery of Vitruvius in the Renaissance, and the interpretation of his works by Palladio.

Palladio was the first important architect who constructed his designs by means of rules. This enabled him, for instance, to produce a large number of variations on the theme of a 'villa'. "I Quattro Libri dell' Architettura", published in 1570, mentions several such rules. Other rules can be reconstructed by analyzing his villas, as was done recently in a study by George Hersey en Richard Freedman. [5]

Hersey and Freeman tested the correctness of the reconstructed rules by implementing them in a computer program for generating new designs for Palladian villas. In this way, they found many mistakes and inaccuracies in the studies about Palladio from the last 400 years. It also turned out that Palladio was not consistently precise in applying his own explicit rules. Palladio's Platonic villa did not always survive its materialization.

The rules employed by Palladio and his contemporaries were formulated in terms of elementary transformations such as translation, rotation and reflexion. The same rules were used for designing the overall structure and for designing component parts. In nature we also see that simple transformation rules can yield complex results. And Artificial displays this phenomenon as well.

The game of architecture.

Let us get back to the issues of today's architecture and urban planning. The great challenge in this area consists in the necessity of flexibility and the attractiveness of plurifomity. It is not a good idea to fix someone's living environment forever, and it is even worse to do that for everyone in the same way.

The building process must not be based on an unduly precise definition of the function of "living", if our houses and cities are not to get alienated from actual life. The very fact that the building process is organized in a particular way, implies that certain aspects of "living" will be ignored.

The designer only knows his own preferences, and usually he does not recognize these as limitations or fixations. There is every reason, therefore, to split design decisions into several levels. On a higher level, boundary conditions are specified for the lower level. The way in which these conditions are satisfied, may then be determined by chance, or by the end-user, and thus reach beyond the fixations of an individual designer.

Designing a house or a neighborhood should only consist of articulating structure-defining rules (for instance about space, material and capacity), and a specification of the dependencies between the different rules and decision levels. In this way, for instance, the designer specifies a set of potential partitions rather than one actual one. Within the limits of the rules, the game can be played in different ways, with different possible outcomes, not necessarily anticipated by the designer.

For a rule-designer who wants to explore the possibilities and limitations of his game, the computer will be indispensible. The next step would be to design algorithms which transform or generate rules. Formulating new rules would then become part of the game. This may engender growth processes which go much beyond the creativity of the individual designer.

If we look at the increasing complexity of the 'games' which are being played in algorithmic art, artificial intelligence, and biological simulations ('artificial life'), then such a playful architecture seems possible in principle. But is that going to help things? Is the realization of a more pluriform architecture not precluded anyway by practical kinds of problems?

Flexible production methods.

The current practice in housing-construction suggests a very clear pessimistic answer to this question. Every step in the direction of a more flexible procedure tends to be viewed as absolutely impossible, because it seems incompatible with industrial serial production. Industrial serial production is considered as the only way to take advantage of technology in the production process, to be used whenever it is too expensive to manufacture products individually by hand.

It is interesting to note that even in car manufacturing, industrial serial production is no longer considered the only game in town. It was introduced in the beginning of the twentieth century by Ford, but was rejected by the Japanese immediately after the Second World War, when Toyota developed the lean production [6] method, which combines elements from handcrafted production with elements from mass production. Production lines are used, but unlike in classical production lines, the machines are suited for different operations. And the machines are designed so that they can be easily and quickly set up and adjusted. This integration of craftsmanship and automation makes it possible to produce smaller series with more variety without significant additional costs.

Lean production involves innovation at a technological as well as an organizational level -- concerning issues such as logistics (avoiding stock-piling), quality management, administration and machinery, as well as the coordination between designer, producer and supplier. The flexible nature of the resulting production process makes it possible to produce customized series -- which was altogether out of the question in Ford's production lines.

And lean production is not the final step. Production machinery is increasingly computer-controlled, and therefore increasingly flexible. As a result, industrial manufacturing of completely individualized products is becoming a real possibility. Automatization used to be equivalent to uniformity. In the future, it will be the prerequisite of diversity.

The Components House. [7]

Also in architecture, steps have been made in the direction proposed here. A beginning that we can build on was the development of the Components House. This involved a set of project-independent components that can be used to put together different series of houses.

For the Music District in Almere, Archipel Ontwerpers realized two housing projects with very different kinds of houses, combined in a variety of configurations. This approach was developed further in a system called the Components House, where the building blocks are components of houses rather than complete houses. The Components House is based on a number of very simple rules concerning production, space and function. Nevertheless, the components can be used to put together a wide variety of houses and living environments. At the time we did not yet have the possibility to develop architectural results by means of generative algorithms. We therefore employed handcrafted designs.

To facilitate controlling the logistics of the design and construction process, the Component House is divided into three subsystems. The first subsystem contains the elements which determine the space of the house: the concrete elements [8] for building the basic construction. The second subsystem contains various systems for finishing and materializing the (outside) space. The third subsystem contains systems for technical functionality, including "house equipment" [9], intermediary conduit systems and inside walls.

The Components House as a system thus consists of subsystems which in their turn consist of further subsystems, components and elements. But because these different subsystems are connected with each other in complex ways, they do not constitute a simple tree structure. They form a heterarchy rather than a hierarchy.


The Components House fits in well with the modus operandi of an art generation algorithm such as Artificial, that we discussed above. Artificial works with a specification of a complex set of elements and patterns, and of all manners of combining these elements within these patterns; from the set of possibilities defined in that way, the program then draws random samples.

The Components House also specifies a set of precisely defined elements. The 'patterns' for the Comp0onents House are put together out of descriptions concerning the architectural language, the materials to be used and their properties, the desired spatial capacity of the houses, the scale and the degree of differentiation of the composition, the financial possibilities, the constraints defined by building regulations and urban planning, etc., etc. The computer program can generate series of variants of houses that comply with these descriptions. Because of their formal complexity, these variants will have a much more organic "look and feel" than the unequivocal housing blocks which human architects currently press through the sieve of traditional production


[1] Sol LeWitt: "Sentences on Conceptual Art." Art-Language 1,1 (May 1969).

[2] The Artificial programs were designed and implemented by Remko Scha. Contributions to various stages of software development by Anthony Bijnen (Metaform Software, Amsterdam), Vangelis Lykos (Academie van Bouwkunst, Amsterdam) en Boele Klopman (Technische Universiteit Twente, Enschede).

[3] John Cage: Preface to: 'Where are we going? And what are we doing?' In: Silence. Lectures and Writings by John Cage. Middletown, Connecticut: Wesleyan University Press, 1973, p. 194.

[4] From the same preface by John Cage, pp. 194/195.

[5] George Hersey en Richard Freedman: Possible Palladian Villas (Plus a Few Instructively Impossible Ones). Cambridge, Mass.: The MIT Press, 1992.

[6] The concept of lean production was introduced in the MIT research reported in: Womack, Jones en Roos: The Machine that Changed the World. New York: Macmillan. 1990.

[7] The first version of the Component House was developed by Archipel Ontwerpers in collaboration with Nevanco housing.

[8] This first version was based on a concrete construction by Heembeton. Wood, steel or bricks may be applied as well, though every material will impose its own conditions concerning span, stability and production.

[9] By "house equipment" we mean furniture-type realizations of functions such as cooking, washing, and waste disposal. This equipment is connected through an intermediary conduit-system to the central conduit-tube. This creates complete freedom to rearrange the appartment lay-out. Applying Matura prefab-elements, Archipel Ontwerpers has recently designed two example projects on the basis of this concept.