Design Book Review issue 27, winter 1993

ALEXANDER TZONIS AND LIANE LEFAIVRE

The Two New Sciences of Representation

Gyorgy Kepes once called science "the angel with the sword, evicting us from the smaller, friendlier world in which we once moved with a confidence born of familiarity, and plunging us into a bigger, alien world where our unaccustomed sensibilities are forced to cope with a formidable new scale of events."1

The invention of perspective came like one of those angels, causing a change of mind and an eviction into a strange new world. The change was slow in coming, starting at the end of the classical period in Greece, and never fully realized until the early mid-15th century. In the course of these centuries the path of perspective was halting and full of reversals. Looking over the whole period, however, a progressive pattern does emerge. But progressive in what terms?

One of the obvious answers is that, as the years went by, a system of representation of increasing realism was developed. Descriptions carried out through this system be- came increasingly matched with the "way the world is out there." This criterion of progress might be called "externalist," as Hilary Putnam termed it in her book, Reason, History, Truth (Cambridge: Cambridge University Press, 1981).

But there is a second criterion of progress, what Putnam would refer to as the "internalist" one. Looking at things in relation to it, images created through various stages in the perfectioning of perspective became increasingly rational - in other words, increasingly coherent in the way that all their parts fit together, independent of their correspondence to an external reality judged by God's eye.

Accordingly, with few exceptions, every successive century can be seen as producing pictures that become internally free of contradiction. Visual lines, the cone or the pyramid of vision, the horizon - all the devices employed in the practice of perspective - help not only in developing more naturalistic icons, but also in informing more consistent constructs. Given this model of development, at the beginning we ought to find pictures composed of fragments, and ending with pictures that are totally systematic. Indeed, historical evidence confirms this hypothesis. Frescoes, mosaics, and pottery of late Antiquity all contain isolated islands of systematic foreshortening - a ceiling here, a table there, part of the floor in one case, a combination of a side-wall with the ceiling in another - until suddenly, these islands merge into a unified landscape, and all the aggregates come together and are subjugated, without exception, to the rigor of the intercisione della piramide visiva, or convergence of the visual pyramid.

Without doubt, the appearance of these new objects created through perspective must have come as a shock. Was this effect unpleasant? Was it akin to the shock of finding a stranger in the intimacy of a hortus occlusus, to use another of Kepes' similes? Far from it. It was a surprise, but a very pleasant one, to the degree that it was greeted by Paolo Uccello with the exclamation, "Oh, che dolce cosa e questa prospettiva!" ("Oh, how sweet a thing perspective is!"), as Giorgio Vasari reports.

The thrill felt might be explained by the illusionistic impact of perspective - based realistic pictures. People were drawn to such new idols as the birds were to the fruits in the painting of Zeuxis, as Pliny relates in his Natural History. Yet applying Putnam's second criterion-that of internal coherence - to the case, we might find another reason why perspective-based pictures were such an indulgence to create and so hedonistic to behold. It was because of how they were structured, because of their coherence which reflected the way the mind was made; it was because the visual pyramid projected natural objects onto the canvas, and the constitution of the mind onto the world.

Pictures made according to the new system of representing perspective were descriptions within which every form to be found was tagged with categories of the mind, or, more specifically, spatial categories. The tripartition that characterized the organization of perspective-based pictures corresponded to the tripartite cognitive framework of front-middle-back, up-middle-down, right-middle-left - categorical structures internal to the mind. Consequently, what the viewer recognized in such paintings was nature categorized, humanized. Perspective paintings were not only naturalistic images, but also mental images.

To quote Putnam once more, the forefather of this "internal realist" position is Immanuel Kant, whose impact on the history of science, technology, and art has been enormous. His influence among German-speaking art historians in the first part of the 20th century - in particular, his magisterial philosophy of symbolic forms - was carried out through the writings of Ernst Cassirer.

Erwin Panofsky's book, Perspective as Symbolic Form, tells, or rather outlines, the story of this "anthropocratic" takeover of the world, "the translation of psycho-physiological space into mathematical space (or logico-mathematical space).. .an objectification of the subjectives." Panofsky's text, originally published in 1927 in the Vortrage der Bibliothek Warburg 1924-25 (Leipzig & Berlin, 1927) has been published in English by Zone Books, decades after it was translated into Italian and French. Scholars using the Fine Arts Library at Harvard University might have encountered a precious, unauthorized, typewritten translation of the text, accompanied by a short initialized note by Panofsky himself warning the reader, in Latin, of potential errors. Perspective as Symbolic Form is a short text followed by a formidable barrage of notes. It is a pity that the newest publication of this brilliant and very influential document lacks an index.

The accent in this seminal essay centered more on aspects of internal-coherence criteria than on external ones. Additionally, in an intriguing turnabout, Panofsky suggests that the Kantian internalist approach had its roots in the emergence of perspective as a paradigm. From the outset, the Kant-Cassirer orientation is obvious given that, in the original German title, "Die Perspektive als 'symbolische Form,'" "symbolic form" is set off in quotation marks, in reference to Cassirer's The Philosophy of Symbolic Forms (first published in German in 1921, and in 1977 by Yale University Press). Unfortunately, the quotation marks were dropped in the English translation.

There is an evolutionist view combined with the neo-Kantian one in Panofsky's study of the incremental systematization of the perspective-based artificial world through the centuries. The progress of cumulative systematization had an order. Despite the brevity of the study, there is a sufficient dilineation of the phrases. The same evolutionist order is also found in the development of other cultural systems in their progress toward coherence. The poetics of classical architecture emerged out of antiquity and reached its maturity in the Renaissance in a similar way. Related intemalist/externalist criteria can also map this path. In some cases, such parallel ways were autonomous; in others, they were intimately interlinked and reinforced each other. The investigation of such correspondences could lead to the identification of hypermaps of intercultural influences, as well as the identification of deeper forces, such as the growing mentality of bourgeois rationalization which produced such structures as bureaucracies, the systemization of economic affairs, the legalization of every- day civic life, leading a variety of cultural developments.
Unlike his later book, Gothic Architecture and Scholasticism (1951), Panofsky's study on perspective rarely refers to factors exogenous to the phenomenon of the development of perspective (one rare exception is when he notes that "the space of Giotto and Duccio correspond [s] to the transitional high Scholastic view of space."). Compared to Gothic Architecture and Scholasticism in which he deals again with the problem of incremental systematization/conceptualization of the artificial world, Perspective as Symbolic Form is a relatively single-plot narrative.
Hubert Darnisch's L'Origine de la Perspective offers us a good overview of the abundant work carried out on the topic since Panofsky's seminal essay, and goes to great lengths to establish such relationships between multiple layers of the movement. The passages about the relationship between Renaissance perspective and the ideal city are particularly intriguing.

Current appetite for multi-layered contextual studies of culture will certainly be opened by the efforts of Christopher S. Wood in his introduction to Panofsky's essay. He brings out very clearly the relation between Panofsky and neo-Kantians. Wood's introduction situates Panofsky's essay in the context of German art historical scholarship and of the tensions that existed there between an ahistorical structuralist philological as opposed to historicist-contextualist thinking. He also elaborates on the discussion of the conflict between the subjective-versus-objective identity of perspective representation system with which Panofsky's essay ends. This he links with a current debate about the history of culture, science, and art as posed by one of the most radical exponents of relativism, Paul Feyerabend.

The representation system of perspective explicitly presupposes a juxtaposition between object and subject, the world as is and its maps as they appear. Depending on the individual point of view, such a representation system leads to descriptions that can differ considerably from each other even if they stand for the same object. And none of the descriptions are necessarily more "true" than the others. This invites, by analogy, implications about the incommensurability of human beliefs. Wood belongs to a different belief outpost than Panofsky. Siding with Damisch, with whom he shares many opinions, Wood appears to regret Panofsky's "totalizing" mentality and last-minute rejection of "perspectivism" which the study of perspective could have easily implied.

The implications of the invention of perspective were not only critical or epistemological, but also technological. Constructing a perspective description according to the perspective representation system became a completely reliable and standardized procedure, an algorithm-so much so that it easily led to its mechanization through the invention of the camera. It also contributed significantly to the creation of an even more automated system of representation two centuries later: computer-image generation. But while the camera appeared to have inherited the "conservative" aspects of perspective, or those related to its "totalizing" character, the computer offered a radical, "perspectivist" alternative. Once more, however, the angel of scientific change was full of shocking surprises.

This radical alternative created new uncertainties. Perspective had its critics, two of whom Panofsky refers to toward the end of his essay: Plato, who condemned perspective as it was born because "it distorted the 'true proportions' of things, and replaced reality and the nomos [law] with subjective appearance and arbitrariness"; and El Lissitsky, who attacked perspective because it "limited space, made it finite, closed it off." In order to overcome perspective's "closed" character, Lissitsky proposed a solution which Panofsky recounts: "The conquest of an 'imaginary space' by means of mechanically motivated bodies, which by this very movement, by the rotation or oscillation, produce precise figures (for example, a rotating stick produces an apparent circle, or in another position, an apparent cylinder, and so forth)." Panofsky looked down on Lissitsky's proposal because, despite its aspirations to go beyond the "Euclidean" prison-house, it led to "Euclidean" space pictures itself. Lissitsky's vision, on the other hand, implied something more: it presaged in a fascinating way the kind of work students would be turning out as standard practice on computers half a century later. In other words, it appears that the new system of generating descriptions of space, offered today by the computer, was in direct response to Lissitsky's critique of perspective.

One of the most fascinating capabilities that the computer offers is that, once it has helped us build perspective images on the basis of planar data, such as plans and sections, it then allows us to control them. Pictures are stored and recalled, cut and pasted, squeezed and expanded, and finally, if not restored, erased. In other words, computers instantiate the program Lissitsky envisaged, a new kind of painting beyond the confines of perspective. Lissitsky' s dream was, in fact, misunderstood by Panofsky who concentrated only on the end-product of a process that Lissitsky was discussing. The artist was referring to a dynamic process that departed from an initial perspective description of an object, what he called the "rotating stick," to arrive through an intermediate series of steps at a second transformed description, an "apparent circle" or "apparent cylinder." Thus, the representation system that Lissitsky envisioned was an open sys- tem, a dynamic system within which objects could be "parametrized," to used CAD terminology, a system which computers contain today.

The story of this new invention, its possible applications as well as its possible implications, is found in William J. Mitchell's new book, The Reconfigured Eye. Like Panofsky's book, which charted the development of perspective in an evolutionary manner, Mitchell's investigates the development of computer-based picture generation. In so doing, Mitchell sees photography as the necessary precondition for the emergence of electronic photography, still video, and digital camera - the basic distinction between photography, or, in his words, a "modern" representation system, and digital image synthesizing, a "post- modem" system lying in the "analog" (continuous) character of the former as opposed to "digital" (discrete) character of the latter.

Mitchell uses an architectural metaphor to express this difference: the continuous motion of rolling down a ramp versus the discrete sequence of steps down a staircase. In the first case, you cannot count your steps reliably. In the second, the discreteness of the stairs makes the counting very easy. Thus, computer-based spatial descriptions can be stored efficiently and effectively, but can also be very easily manipulated, much more so than their ancestors, the camera- based images.

Perspective was the presupposition of photography. But it was also the presupposition of computer-based pictorial representations. Without perspective, the mathematization of space necessary for computer representations would have been historically impossible. It is interesting to keep in mind that, at their inception, perspective descriptions were conceived as digitally constituted. The perspective system of representation was based on the ballistic paradigm, what has been called "the arrow in the eye." Rays are "shot" at, or from, the eye (de- pending on the assumed general theory of op- tics) piercing an intermediate plane. Since the act of shooting is discrete, the traces on the plane are assumed to be discontinuous points rather than lines. This is quite explicitly stated in Albrecht Durer's famous engraving of a man drawing a lute illustrating the second perspective apparatus (from The Painter's Manual, which was first published in German in 1525, and in 1977 by Abaris Books).

Not only does the illustration clearly show a picture made up of dots rather than a sweeping line, but the accompanying text makes the digital paradigm and its consequent procedure explicit. The description is generated by "marking the spots, moving from point to point" discretely until the whole object has been "scanned and its points transferred to the tablet" (a translation of "das die ganze lauten gar an die tafe1 punctirst," Durer's emphasis).
The Reconfigured Eye engagingly shows the computer's capability of storing massive information and of rearranging images endlessly - this, with today's desktop technology, which is accessible with comparatively modest budgets (according to Western standards). More pragmatic, not to mention more detailed and broader in its applications, is the Digital Design Media (Van Nostrand Reinhold, 1991), also by William Mitchell, along with Malcolm McCullough. Its tone is that of a textbook, and its structure and selection of references make it a helpful companion to a general-purpose computerized design course. It's important to note that both the text and diagrams in this book are more pleasant to peruse than those in the classics of this genre.

The Electronic Design Studio, edited by Mitchell, McCullough, and Patrick Purcell, is even closer to the classroom than Digital Design Media. This is a collection of papers initially presented at the CAAD Futures Conference in 1989 in Cambridge, Massachusetts. The thirty-three papers are all equally passionately engaged in bringing computer tools to the design studio. What's interesting is that all the participants share a common orientation toward architecture and computers. Unfortunately, the authors also share, to a high degree, a continued dedication to traditional CAD. With very few exceptions, current developments in artificial intelligence, computer vision, cognition, and "architectural knowledge" are given short shrift. The variety of contributions come together and form a kind of medium-sized orchestra on the topic. The virtuosity of the individuals is uneven, however, as is usual in such cases.

Computers can not only describe and control objects in space geometrically in the tradition of perspective, but can also capture more complex, conceptual spatial aspects of the world by employing space-related cognitive structures. They can describe, explain, and predict compositional organizations in a sophisticated, non-reductive way, like an expert. In addition, they can describe, explain, and predict moving in space, causal chains of events-such as social interaction in buildings as constrained by spatial configurations-and the interdependence of multiple processes 'occurring within the environment. In doing so, they employ a design intelligence that is not embodied ("enminded" is perhaps a better word) by traditional CAD.2

From recent evidence, the situation seems to be changing drastically, and most CAAD researchers have moved quickly
into such advanced territories of computer applications. Follow-up applications are certainly eagerly awaited.

Ironically, The Reconfigured Eye, which is, by its genre, more identifiable as an essay of ideas (and a lavishly illustrated one at that), is the most desirable of all the books mentioned as an educational support tool. Moreover, it will no doubt be the most lasting contribution on such a swiftly changing subject. Mitchell's prognosis of future uses of computers is modest and pragmatic in comparison to the more wild speculations put forth by current publications such as Michael Benedikt's Cyberspace (see reviews on pages 17 and 19), a book that meditates on a grandiose and more utopian scale about the possibilities of such manifold image storage and facile manipulation of high-tech engines.

Despite their relatively conservative character, Mitchell's projections might still come up against serious difficulties in being implemented. As in the case of CAD, digital image technology will always be confined unless supplemented by a more advanced methodology.1n both cases, the computer's power to store and process information is highly overestimated. It isn't that computers cannot perform ambitious tasks such as those that CAD and digital image theoreticians imagine and aspire to; but such feats of performance cannot be founded on a reductive modeling of design thinking and an unchallenging theory of design intelligence, because of limits of memory and time, which constrain even the most advanced computers.

The Reconfigured Eye is explicit about such limits when digital image applications grow in size. The sheer "shrinking" of the information contained in the recorded images, however, is insufficient. The answer is to be found in neither software nor hardware improvements; rather, the place to search for this special intelligence is in the culture of architecture and in the nature of the mind. Culture and nature have already developed many efficient and effective strategies, through history as well as natural evolution, enviably mastering the tasks of hoarding huge amounts of images and reconfiguring them inside 'Our heads.

It is no accident, at this moment of challenge posed by the reconfiguring eye - with so much refocusing on design intelligence and rethinking about visual cognition now occurring - that Panofsky' s text on perspective has been remembered, translated, and commented upon.
Like Uccello, Mitchell is captured by the sweetness of the new representation system and its challenges. And one might speculate that he is drawn not only by its potentials but also by the fact that, through the new system, we see projected in the world the structure of our internal cognitive apparatus - as in the case of perspective.

But unlike those who wrote about perspective in the Renaissance, Mitchell's book is preoccupied at least half of the time by potential problems and dangers which the new system might bring about. His point is not only excellently argued, but also very well taken. How can we guarantee that an explosion of fake images, images that can be very easily and abundantly manufactured by today's machines, does not destroy the integrity of our culture in the near future? Plato was worried that perspective would populate the world with fake imitations of the Ideas. Mitchell is troubled with the proliferation of fake documents reconfigured through new technology emerging constantly during our lives. Another equally important question is how will we cope with problems of intellectual property without frustrating the enormous potentials of "reconfiguration"; or, if we opt for giving total license to re-creation, how will we reward the initial creators? These are some of the headaches that will come, along with the hedonistic moments, in the new machine-based fantasy of tomorrow. After all, the angel of change holds a sword with a double-edged blade.
NOTES
1. From the Winter 1960 of Daedalus: The Journal of the Academy of American Arts and Sciences, dedicated to "The Visual Arts Today"; Gyorgy Kepes was the guest editor.
2. The resulting limitations can be seen very clearly in Possible Palladian Villas by George Hersey and Richard Freedman (MIT Press, 1992)1.
PERSPECTIVE AS SYMBOLIC FORM, Erwin Panofsky, translated and introduced by Christopher S. Wood, Zone Books (distributed through MIT Press), 1991, 196 pp., illus., $24.95.
L'ORIGINE DE LA PERSPECTIVE, Hubert Damisch, Flarnmarion, 1987,416 pp., $60.00.
ELECTRONIC DESIGN STUDIO: ARCHITEC- TURAL KNOWLEDGE AND MEDIA IN THE COMPUTER ERA, Malcolm McCullough, William J. Mitchell, and Patrick Purcell, editors, MIT Press, 1991,505 pp., i1lus., $19.95.
THE RECONFIGURED EYE: VISUAL TRUTH IN THE POST -PHOTOGRAPffiC ERA, William J. Mitchell, MIT Press, 1992,273 pp., illus., $39.95.

Dialogue with Alexander Tzonis:

Artificial Intelligence for Intelligent Architects

Alexander Tzonis is the chair of architectural theory at the University of Technology at Delft, Holland, and is the director of Architectural Knowledge Systems, a multidisciplinary research' group. He attended graduate school at Yale University and taught at Harvard betweeil1967 and 1981, where he came into contact with the early research in artificial intelligence then being undertaken in Cambridge.
Tzonis coauthored, with Serge Chermayeff, The Shape of Community (Penguin, 1971) and soon after published Towards a Non-oppressive Environment (i press inc., 1972). He has a consuming interest in architectural theory and its documentation, and is the chief editor of the Garland Archives, a multivolume series which has published the complete archives of Le Corbusier, Louis Kahn, Mies van der Rohe, Walter Gropius, R. M. Schindler, and Alvar Aalto. (See DBR 18, "Architectural Publishing").
With Liane Lefaivre he has written numerous articles and books, including Classical Architecture: The Poetics of Order (MlT Press, 1986), The Roots of Architecture (SUN, 1984), and Architecture in Europe Since 1968: Between Memory and Invention (Rizzoli, 1992). He has edited a forthcoming book, in collaboration with Ian White, Automation-Based Creative Design (Elsevier, 1993). In 1990 he published his first novel, a murder mystery about problem-solving, computation, and morality, entitled Hermes and the Golden Thinking Machine (MlT Press, 1990).
The specific agenda of the doctorate program directed by Tzonis, known as Architectural Knowledge Systems: Artificial Intelligence for the Intelligent Architect, is the documentation and analysis of architectural knowledge using computational, cognitive methodologies.
DBR interviewed Professor Tzonis in Delft in
December, 1992.
DBR: Reading the program description of your research group, I see the largest number of the participants are preoccupied with computers. What relation does the supposedly automated sphere of knowledge have to the more humanistically based "architectural knowledge"?

AT: I strongly believe that no computational theory or methodology or techniques of architecture can be developed without deeply founded architectural knowledge. As to whether the reverse is true, this is a most interesting question. Up until very recently most people in architectural theory or computer-aided design worked independently. They stayed in one direction, looked exclusively at architectural theory, and tried to understand it independently of computational aspects, which they viewed with suspicion. Or they concentrated on computer applications and tried to expand and improve them, disregarding architectural theory, which, in turn, they eyed with misgivings. It seems that the situation is changing now. We realize that neither architectural theory nor computer techniques can develop in isolation. The "development of architectural theory" encompasses cognitive as well as historically rooted rule systems. The closer we get to understanding architectural practice, the more we realize that it instantiates preexisting rules and activates predefined systems, despite the fact that it appears to improvise in a freewheeling manner.
DBR: Of what use are computers 10 history and theory? It seems to me that they are better suited to the more mechanical demands of practice.
AT: We should use the computer as a tool for both, and we do. Computers are one of the means for studying and thinking in general, and design-thinking in particular. With computers, we can approximate a kind of replica of the mind, which, however crude and reductive, is still the best tool we have. Cognitive scientists cannot test the brain to confirm hypotheses about structure-function relations of the mind, but they can collaborate with artificial intelligence scientists and test their hypotheses, experimenting freely with the organization of machines and their performance. After completing an initial phase of investigation dealing with more general kinds of intelligence (such as vision, language, and movement in space), research in artificial intelligence has reached a second phase where more complex kinds of intelligence (such as spatial and visual thinking, and design) are now approachable.
DBR: How would a computer-aided historian proceed, and what are the potential consequences?
AT: Like many other researchers we try to find actual cases of professional practice, document their methods in use, and reconstruct their underlying methodology. Where our group differs significantly from others is that we try to carry out such reconstructions in historical cases of exceptionally creative, intelligent performance. A considerable part of our activities are devoted to historical documentation and compiling archives be- cause, without scientifically organized archives, an empirical study of creativity, dealing with highly contrasting cases and claiming a high degree of generality, would have been impossible.
DBR: Are you interested in the constants of what you call design intelligence, or in historical change?
AT: Both. What we find in our research is the fascinating dialectic between the architecture of the mind, which is more or less invariable, and the architecture of culture and society, which keeps on growing. The way social forces and social perceptions shape or are shaped by knowledge has its own constraints. It constitutes a structure that was referred to in the mid-1970s as "socio-formations." I believe we are going through a phase where we recognize an- other kind of formation which is subject to strong changes, and that's the formation that comes out of the mind, out of cognition, on which society builds and whole systems - belief systems, ideological systems - are built. On top of it you have society, economics, politics, cultural super- structures that are always shifting, changing, reflecting the totality of society rather than the single cell of an individual mind. Architectural knowledge is a partial case of such belief systems. Architecture is nothing but a component of human thinking - not only an application component, but a component that is deeply rooted in the very nature of the mind. Creating architecture is a very human activity, like creating language or music. At its most fundamental level it has to do with spatial thinking.
DBR: I would like this to be made more concrete. Let us come back to architectural practice for a moment. If I understand correctly, the premise of the research program at Delft, the Architectural Knowledge Systems, is that in order to achieve technical breakthroughs, you need theoretical groundwork. How does that relate to the "structure" of knowledge?
AT: I will give you an example. Imagine, for instance, the making of a CAAD drawing. It can be produced very elegantly, quickly, without mistakes, and employing people who are not necessarily talented at drawing; it results in a reliable plan of a building. Some- one from technology or a completely different area in the building profession walks in and introduces four kinds of requirements that necessitate the redrawing of the plans. These changes might be relatively trivial- adding a bit of diameter to columns here, cut- ting a wall there, changing the fenestration and so on. Now, as we look at a CAAD drawing, we can easily envisage the changes because the mind grasps them quickly. That is why we call them trivial: they require no special thinking or knowledge to be carried out. While we modify dimensions on one end of the paper, we must calculate their repercussions on other components of the building that are connected to the ones that change. Trivial as they are, these changes still take time and energy. They are disruptive and ex- pensive, whether the redrawing is done by hand on transparent paper, or by a machine.
One would have to go through more or less the same process of redrawing when using CAAD techniques because once automated redrawing gets more sophisticated, it is impossible. Why? Because, for CAAD, there is no meaning in the lines. The computer only knows vectors and their coordinates. These lines do not represent objects, which have properties, or identifiable components of architecture. But when you or I look at the drawing, we understand precisely those identifiable objects, parts, and components, which may also have acceptable shapes, tolerances, dimensions, and various properties-geo- metrical, natural, symbolic, aesthetic spatial, cost microclimatic, and so on. If we want to develop a truly sophisticated and highly practical "redrawing technology" for an architectural office, then we have to introduce into the computer this component of intelligence and knowledge. There's no way to introduce it except through a highly theoretical analysis of what it involves to draw and recognize shapes that have meaning - technical, aesthetic, symbolic, functional, and so on. Theoretical analyses and theoretical breakthroughs should thus become a prerequisite for the most practical, efficient technology to be introduced in an office.
DBR: Is this kind of trivial redrawing job the main area of architecture where computers are moving?
AT: Not really. I will give you another ex- ample that involves more sophisticated skills: all architects begin with a sketch. Then the problem arises of how to translate the sketch into an architectural drawing. What happens frequently in an office is the following: the architect in charge of developing a concept does the sketch, then it may be decided to translate the sketch into an architectural drawing to test the idea. Most of the time, another architect takes the sketch and redraws it into a more accurate architectural drawing. This is a trivial job, but at the same time, it requires a very sophisticated understanding of what is involved implicitly in the sketch. From the technical and economic point of view, it is desirable for a machine to do it. But a machine cannot do it unless we do develop a sophisticated theory of what is involved when we look at a sketch and read the implicit things within it about what is to be built. It might involve the smallest, most minute change of a line, but this line is charged with technology, with architectural history, with symbols and all the other knowledge-loaded aspects of design. Now, imagine the drawing made by the assistant confronted with another modified sketch by the concept architect. Imagine how simple it is for a knowledgeable viewer to understand what changes are needed in the architectural drawing, but look at how difficult, impossible for a CAAD-loaded machine to automatically carry out this "mechanical" job. It cannot even control basic stylistic rules or historical constraints implied in a sketch, not to mention issues of function.
DBR: What about shape grammars? Don't they do exactly that-link history with CAAD?
AT: If you go to a CAAD conference, you will find a lot of teachers who claim they do historical work in their computers because they use shape grammars, and actually do end up with drawings that look very historical. This truly could give both history and computers a dubious reputation. First, shape grammars do not capture the richness and complexity of the works of the past. The authors of such grammars are deceived by the apparent simplicity of some of the so-called historical styles. Take classicism, for ex- ample. It appears so elementary, either when you look at a Palladian villa or when you listen to a Mozart sonata.
Some people are under the false impression that any beginning piano student can play Mozart. Well, they may be "playing" Mozart, but they're not interpreting Mozart, because the real theory behind a Mozart piece is not only loaded with a whole overlay of rules and principles, but also requires a familiarity with the dialogue that Mozart's work has with a large number of precedents to which it constantly connects. Similarly, shape grammars deceive their users. They give them the impression that they are in control of the methodology of historical works, while in reality, very little is explained to them about it and no real effort has been made to reconstruct this methodology.
Second, the computer itself is used in ways that are inferior to its capability to capture the complexity of the mind. A rigorous analysis of style presupposes knowledge of architectural theory in all its depth and rich- ness, and the reconstruction of its categorical organization and principles. This in turn pre- supposes robust spatial semantics and an understanding of language, vision, and cognitive restraints. All that, to be handled computationally, requires computer programs and algorithms that are far more sophisticated than the ones employed by shape grammars. The irony is that such sophisticated tools turn out to be much easier to use.
DBR: Don't you think computers invite such simplistic attitudes?
AT: Computers are indeed very prone to reductionist interpretations, like any scientistic approach. What we are trying to do is exactly the opposite: we try to introduce difficulty. Our ally in this approach is reality. One can observe that all the reductive approaches are very successful for six months, and then re- veal their limitations. Ultimately, they are exhausted very quickly.
The issue of function is conspicuously absent from most discussions about CAAD and shape grammars in particular. The reason is obvious: neither paradigms were made to function. In other design domains, however, where artificial intelligence was applied, form and function cohabited from the very beginning. And this is how we have been approaching architectural design in our laboratory.
DBR: How is your program structured?
AT: The Delft program of Architectural Knowledge Systems is organized into three branches. The first is "architectural documentation," where those who are interested in developing a basis or substructure for architectural research develop architectural documentation by looking to the literature, criticism, poetry, legal documents, religious documents, and so on. The student puts it into a package that can then be subject to rigorous interpretation. In many cases, the documentation takes place outside the university. The common link is the Garland Architectural Archives project, for which I am the general editor.
Occasionally the research involves an even more basic aspect of documentation, such as identifying and structuring the sources of architectural discourse. This was the case with a study carried out by Dr. AI- Abed on early Arabic discourse of architecture.
The second branch is given to interpreting the architectural documents, and developing from them an explicit reconstructed theory. This involves the systematic analysis of the documents, which have an implicit theory, an explicit reconstructed theory, which is still precomputational.
The third phase is to turn this knowledge, or knowledge systems (because they're not empirical data any longer but implementational) into machine-based systems that have a descriptive and explanatory power or a predictive design power. In other words, through those systems you can design new objects.
DBR: How can you relate descriptions and explanations to predictions? How can you make design out of documentation?
AT: The fundamental departure in our approach is that creative design is based on precedent-in other words, that creating de- sign from a tabula rasa is impossible. While many have believed and still believe that the creative act happens on a clean slate, we are trying to show that it is an illusion. In fact, our work explains why a large number of utopian ex nihilo schemes led to disasters, with or without the use of computers.
This can be shown by investigating how architectural thinking works and by showing, on the cognitive level, how it is impossible to deliver satisfactory products unless operating under constraints. Looking at architectural history, we identify these constraints as precedents.
Accumulating all kinds of cases for their own sake doesn't work. To start with, you have no way of selecting them, sorting them, reusing them. For this you need some preexisting principles, some precedents. But they are of a different kind than those we get from experience. They are deeper and much more abstract and generic. They are "pre-wired," so to speak, in our design apparatus. Then we have another type of precedent rule: those that we extract from individual cases carried out in the past. Precedent, pre-wired principles make this ex- traction possible. Certainly, we always keep precedent cases as isolated objects as well, to use whenever needed for special, "non-generalizable" situations, for example.
History and cognition research need each other. An understanding of cognition is very important because it can demonstrate precisely why "precedentless" design is impossible. And this is where history is also very much needed. It can show that those incredible leaps of the imagination, pure creative acts, were in fact examples of recasting old information into new shapes-of rethinking. Analogical design thinking allows designers to cannibalize information, knowledge, solutions of the past into completely new meanings for the present. So we have a number of examples where we try, through case studies, to demonstrate this idea and to build systems out of them which can generate design.
DBR: How do you situate this work in relation to what has been done in the past, in places like Berkeley with Design Methods (D.M.) research or, on the other hand, with Christopher Alexander?
AT: We know a little more than D.M. researchers did about the complexity of design problems, and we also take cognition more into consideration, as well as the finitude of the human mind. Bringing the two things together, it becomes increasingly clear how it is impossible to solve problems with any degree of complexity ex nihilo, as the D.M. approach assumed. Designing a huge warehouse might be possible without precedent. Its complexity is superficial. But if you try to rearrange your bedroom ex nihilo, out of the sheer combinations of demands, it gets much more difficult. Of course, Christopher Alexander has already said this in his A Pattern Language. What reformed Alexander, a one-time advocate of ex nihilo design, was the theory of language patterns inspired by natural language pattern studies of the 1960s. Language pat- terns are, of course, based on precedents; the problem with Alexander is that he does not supply a methodology demonstrating what is and what is not a precedent, or how, once you have patterns, the patterns can be combined in any way. Or how patterns are sensitive to different contexts. In other words, you had a surface analysis rather than a coherent system. The more you are dealing with a closed community, which changes very little, the more these patterns are operational. But the more you are dealing with buildings that are generated by innovating programs or with alien environments, the less potent these patterns are. For these reasons, Alexander's theory lacks depth. But his initial intuition about precedents was actually correct.
DBR: Can you give us an example of any of the cases you have investigated involving your holy triad, history-theory, cognition, and computers?

AT: One of the most exciting cases has been the analysis and reconstruction of the conception of the Unite d'Habitation by Le Corbusier. The idea of design through precedent follows a model we research in which you have new programs, you have precedent, and then you have a frame through which they can be combined. And out of that, you have the development of new solutions; you need the two components, plus the theory contexts within which it is possible to marry new questions with old answers, and rephrase, reorganize the old solutions.
Here, for example, is a case study of creativity, reconstructing what went on in the mind of the designer. We take the ship, bottle rack, and hut used by Le Corbusier as sources for the Unite d 'Habitation. He's con- fronted with a program, he wakes up tortured by a number of questions, but instead of trying to develop answers ex nihilo, he turns to his memory and asks, Do any of those questions remind me of something I already know? Slowly, from inside his conscious memory emerge very concrete, integral objects. Now he focuses on each and extracts from it the part that is relative to the solution. For example, out of the Swiss hut, he ex- tracts the component of the piloti, which will be applicable to the Unite d'Habitation, and he uses several aspects of it that obviously ignore material, scale, number, configuration, but provide something, a fundamental gestalt of the design solution. At the same time, he pulls a similar extraction out of the bottle rack, which contains the interdependence of bottle to bottle rack the way the units to the structure will. Finally, he chops off the top of the ocean liner. The astounding ability of the mind is that, out of those fragments, it assembles and is then able to "make them new," to adopt a phrase from Ezra Pound, to synthesize them into an unprecedented whole. This appears crazy and incredible unless we try to interpret it; then we discover that there are certain fundamental structures of how objects are put together, of how space is put together, or how functions are interrelated inside forms that permit the re-gluing of those pieces into the new whole. So here you have the theory and the knowledge behind every object, a kind of gestalt.

Another case we have been investigating is the triangular bastion by Leonardo da Vinci in the 15th century, where sciagraphia (the art of drawing shadows) is rethought and renewed, to serve as a precedent for developing the algorithm for drawing the plan of the optimal fortification. Another of the studies in our group analyzes the way experts work today in the context of Beijing's old city in developing new housing. The method is reconstructed cognitively and computationally.
DBR: Do you have a more synthetic project
AT: Yes - an Architectural Thesaurus, an organized, intelligent memory whose structure reflects architectural knowledge constraints. It contains design principles capable of capturing large numbers of design rules drawn from precedent cases. The system contains the following modules: (1) Plan recognition: the development of methods for automatically recognizing architectural drawings; (2) Architectural program interpretation: the development of methods to analyze, parse, and accept constraining elements, normative or factual, from the discourse of architectural programs; (3) Analogy: the development of analogical design inference for searching and retrieving architectural precedents in the Thesaurus, matching architectural programmatic constraints of form-operation-performance. Analogy as a complex mechanism of inference is characteristic of most intelligent thinking.
DBR: But isn't this a project that should have been undertaken in collaboration with other groups, at least on the level of putting together all the data needed?
AT: It is an idea that we have been exploring.