Introduction

Artificial Intelligence (AI) is a branch of computer science that deals with the development of intelligent computer programs which solve problems in a way that would be considered intelligent if done by a human (Waterman 5). Knowledge based systems are subsets of AI programs that utilize domain knowledge (in our case architectural knowledge) which is explicit, and separate from the program’s other knowledge (Galle and Kovacs, Logic of Worms 6). Knowledge representation is a central issue of AI, and described as "a set of syntactic, and semantic conventions that makes [it] possible to describe things" (Bench-Capon 11).  AI offers several means of representing knowledge such as logic programming, production rules, frames, and semantic networks (Bench-Capon 27).

This commentary bibliography aims at presenting some works dealing with knowledge representation issues for architectural systems. The first group of works discuss the use of AI in design. Several approaches to formalization of architectural knowledge are explored in the second section. The final sections present types of knowledge based systems for architectural systems, with some examples.
 

AI in Design

Flemming argues that AI is a means of understanding a problem itself. Besides solving it, and because of this property and the possibility of incremental growth in AI programs, AI is a helpful device for  ill-understood problems like architectural problems (1-5). Kalay also supports the use of AI in design, claiming that knowledge based systems will be capable of assisting designers at a much higher level of the design process than traditional CAD systems do (319-28). Mitchell claims that the use of AI for architectural design depends on a paradigm of CAD, evolved in 1980’s, of viewing design as a knowledge based activity. Moreover, he claims that knowledge based systems do not respond to architectural problems, because the knowledge that they contain will always remain incomplete (379-83). A conflicting view to this is explored by Zreik who claims that integrating plenty of knowledge within a system is not always necessary to produce "satisficing" solutions to the problems (409). Carrara and Kalay mention the ever growing complexities of the buildings, and the processes that lead to their design, construction, and management, and they claim that these complexities require the involvement of experts from different disciplines in the building design process (389-95). In order to cope with this situation, the building design process should be clearly defined, and methods of AI should be applied to increase the computer assistance to the professionals (396).
 

Formalization of Architectural Knowledge

A formalism is defined as "some conceptual idea that is developed, and articulated into a well-formed abstract system" (Bijl 137). Logan discusses the necessity for formalization of design knowledge models, and claims that such formalisms will provide a tool for further design research (158-70). He explores the issue of representing the structure of design problems, referring to the terms of AI, and cognitive psychology. The term "structure" in this sense, refers to the structure of relationships in a design activity (159). Logan claims that such relationships can be represented by AI devices such as production rules, and problem transformations, and AI in design research should  focus on understanding these relationships, rather than solving problems (160-170).

The most influential work on formalization of architectural design knowledge is the one by Akin who establishes an information processing model of design, called Design Information Processing System (DIPS), that claims to provide explicit descriptions of systems related to design process (55-83). Such systems account for the behavior exhibited by the problem solver during the design process (36-53). Akin infers three general categories of design knowledge from design behavior: problem solving, physical intuition, and inductive reasoning (32-36). Then, he offers a computer system, called Architectural Inference Maker (AIM), for the simulation of reasoning behavior (140-164). The main claim of the author is that, even the intuitive aspects of design can be modeled besides the rational aspects of it, and such a formalized design knowledge can be manipulated in computer environment. Moreover, it can be taught directly in design education as search methods, and study of uncertainty (172-76).

Landsdown’s approach to design is similar to that of Akin. He defines design as a transformation of an object (or a system) from an initial, incomplete state to a final complete one. Since the transformation is brought by the application of knowledge, design can be seen as an information processing concept (120-25). Furthermore, Landsdown discusses requirements for knowledge based systems, and claims that they require appropriate, and multiple representation methods for both domain and control knowledge; truth maintenance systems incorporated; and links to conventional databases (126-28).

Oxman and Oxman propose a structured multi-level model of architectural knowledge, and describe a model of design knowledge that is composed of four levels: syntactic, and formal elements, and operations; syntactic structures, and compositional operations; generic knowledge structures; design paradigms, and schemata (Computability, 172-74). The main claim of the authors is that knowledge of design should be structured, and there should be a meaningful relationship between levels, or types of knowledge (181). They forecast that future design systems of CAD will manipulate such a multilevel model of design knowledge with both top-down, and bottom-up operation (182-83).

Zreik classifies design knowledge into five types: controlled, and uncontrolled general knowledge, specialized knowledge, subjective knowledge, and unknown (396-402), and mentions that recent knowledge based systems have proved their efficiency only with all controlled, and some part of uncontrolled general knowledge (406). In order to improve the capability of such systems of dealing with all kinds of design knowledge, he suggests use of machine learning techniques (407-09).

Bijl discusses the close correspondence between formalisms, and their applications in CAD (128-36), and claims that a formalism becomes invalid when a designer's expectation of an application changes (137). Then, he proposes a formalism for a description of things that is claimed to be distinct from their possible applications. The system is called MoLe (Modeling objects  with Logic expressions). The formal structure of the system consists of a kind, for any kind of thing; a slot that names a part relation to the kind, and a filler that names a part of a kind (140-45). This knowledge representation scheme is very similar to the notion of structured object explained by Bench-Capon (79-87).

The articles by Kocabas, and Woods explore a broader scope of issues on representation, and categorization of knowledge that are also valid for discussion of knowledge representation for architectural systems. Kocabas claims that categorization of knowledge is essential for building large knowledge bases (111). He describes a methodology for organization of descriptive knowledge into several functional categories: logical propositions, mathematical statements, formal statements, grammatical (or meta) statements, theoretical-hypothetical-empirical statements, historical sentences, and factual statements (114-19). The author claims that common sense as well as scientific knowledge can be represented in this categorization system (119). Woods offers a simpler classification system composed of two components: factual knowledge, and rules for predicting changes over time (1323-29). He emphasizes that taxonomic classification structures can advance both the expressive adequacy, and notational efficacy ("the aspects of notation that support both computational efficiency, and conceptual activities such as knowledge acquisition, and learning") of knowledge representations for intelligent systems (1330-33). Moreover, he claims that interest in object-oriented programming marks the beginning of this trend (1334).
 

Systems

Rosenman, Gero, and Oxman identify three types of knowledge based systems: rule-based systems, case-based systems, and prototype-based systems (288-96). This classification is based on classification of architectural knowledge into two categories: compiled knowledge, and case knowledge (286-87).

Compiled knowledge refers to general knowledge which is represented in computer environment either in the form of rules as found in rule-based expert systems, or schemas such as design prototypes which collect several elements of general knowledge in one representation (Rosenman et al. 288). Rule-based systems make use of if (condition), then (action) statements that are called production rules (Waterman 63), and prototype-based systems mostly utilize frames as the main representation scheme (Oxman and Oxman, Computability 180). Rule-based systems are best covered by Waterman, and prototype-based systems are covered by Oxman and Oxman from an architectural theoretical point of view comparing the terms "archetype", "type", and "prototype" (Computability 175-85). Case knowledge, on the other hand is the knowledge in which actual experiences are stored. When a similar experience is faced, all the relevant parts of case knowledge are retrieved, and this knowledge is then applied to the situation at hand. This process is called case-based reasoning in CAD terminology (Rosenman et al 290). The same differentiation in knowledge based systems is also made by Maher et al (1), and Shih (285).
 

Expert Systems

Expert systems are computer programs that contain knowledge derived from a body of expertise, and they differ from algorithmic programs in that they have the capability of making inferences (McCullough 11). Such systems may be developed for the tasks where no systematic procedure exists, or they may use the knowledge that is already available in the form of diagrams, tables, formulas, codes, or standards (Hamilton and Harrison 5).

Waterman’s A Guide to Expert Systems seems to be an appropriate book for an introduction to expert systems. Although it does not include much information on expert systems for architectural design, it provides a general framework for understanding such systems. The main feature of expert systems is defined by Waterman as the body of knowledge that accumulates during system building. The knowledge is explicit, and organized to simplify decision making (16). The system can act as an information processing model of problem solving in the given domain, and it provides the desired answers for a given problem situation. Moreover, the system shows how the answers would change for new situations. This lets the user evaluate the potential effect of new data, and understand its relationship to the solution. Therefore expert systems act as real experts who may make mistakes, but learn from the experience (29-32).

McCullough provides a portfolio of expert systems that make use of building regulations, and design codes as the main source of knowledge. He claims that there are two main reasons for studying building regulations within a knowledge based framework. One of them is that building design codes, and regulations represent a readily available knowledge compiled over a long period of time by domain experts, as a result many knowledge acquisition problems have been already handled. The second reason is that the structure of design codes corresponds well to available knowledge representation formalisms (15). Furthermore, the book covers arguments on building standards theory (16-23), and relevant technologies such as building product modeling, hypertext, and hypermedia (54-85). The author predicts that use of expert systems for building regulations and codes will make such documents more clear, consistent, and correct (21-22).

Hamilton and Harrison discuss the use of expert systems for the building services industry. Their work covers recent developments (4-10), and future trends in building services industry in relation to expert system development. They mention the rapid progress in computer based information sources, and forecast that expert systems as highly developed information sources will be wide spread in future, particularly for regulations, standards, and codes of practice; quality assurance, and staff availability issues (11- 34). They also mention some advantages of expert systems in this area such as time savings, consistency of practices, and freedom of the expert from the routine cases (3).

The article by Chun and Lai explains a recently developed expert system called "Intelligent Critic System for Architectural Design". This system encapsulates different types of design knowledge into independent critic modules. Each critic module contains a different type of knowledge such as building regulations, and interior design principles (625-38). The main focus of the article is the representation of expert knowledge that can be manipulated by computational means.
 

Prototype-based Systems

A prototype is defined as "an especially representative but possibly abstract example of a set of particular objects" (Galle and Kovacs, Logic of Worms 7). Design prototypes are schemata (separate modules of general knowledge) bringing together the knowledge containing a class of design experiences. Design prototypes contain knowledge on function, behavior, and structural properties of a class of design elements, and the necessary procedures for the generation of an instance (Rosenman et al. 289). Oxman and Oxman view prototype as a concept rich in potential for representing a large amount of knowledge, and they claim that integration of prototype grammars, and reasoning capabilities of knowledge based systems are promising for future CAD research (Computability 180). On the other hand, designing with design prototypes is computation oriented, and repetitive. Moreover, there is the difficulty of making generalizations with such systems (Rosenman et al. 290).

A series of three articles by Galle and Kovacs explores a prototype based approach to representation of "soft" architectural design knowledge. The first and final articles of this series have been studied in this commentary bibliography. In the first article, a schematic site plan prototype is analyzed. The representation scheme utilized is formal logic (Galle and Kovacs, Logic of Worms 5-27). The main claim of the authors is that prototype analysis based on logic programming can be applied to any other architectural idea. Since many of the predicates are re-usable, the underlying vocabulary can easily be a component of different types of layouts, and the authors claim that their object types, and predicates are part of a "A Useful Language of Architecture" (28-29). The other article from the series discusses a wide range of design ideas with the common theme of "schematic plaza design" (Galle and Kovacs, Logic of  Plaza 160-76). The arguments mentioned in the first article are concluded in this article. The main claim of the authors is that "soft" architectural design knowledge can be analyzed through prototype analysis based on logic programming, and represented in a form that is available to knowledge based systems (177).
 

Case-Based Systems

Case-based reasoning is a well-defined paradigm in AI. It is based on the assumption that the human reasons from specific experiences rather than by following general guidelines (Rosenman et al. 290). The problem solving approach of case based reasoning is based on the recall, and re-use of specific experiences (Maher et al. 2).

Shih  supports use of case-based systems claiming that architecture, due to its nature, rejects all attempts at describing it with general formalisms. He claims that there is a knowledge which is meaningful only at a specific time, and for specific designers. Although dealing with such type of knowledge is not practical, without it the search for a solution  can not be done efficiently (372). The work of Shih focuses on case-based adaptation in design, and discusses three concepts: case-based search, self organization, and direct translation. Case-based search is a localized search process searching for variations which provide required functionality (373). Self organization deals with context sensitive grammars for localized adaptation (376), and direct translation translates a case directly to another structure by some translation functions (380). The author claims that utilization of these concepts in case-based systems would give rise to CAD systems that support designers better than current ones (385).

Maher et al. explain their approach to case-based systems, and provide a discussion on several examples of such systems. Their problem solving approach of case-based reasoning relies on analogy. Analogy lets people recognize something that they have not encountered before by relating it to something they have. They consider analogy from the perspective of memory, and reminding, and they accept concept of memory organization as a guideline for computer representations (2-4). The work of Maher et al. seems to be a valuable reference as a comparative portfolio of case-based systems. The comparisons are made according to the way of dealing with the complexity of cases (partonomic hierarchy, multimedia representations, etc.), and the way generalized knowledge is handled (geometric constraints, heuristic rules, etc.) (5-17).

Guena and Zreik’s approach to case-based systems is similar to that of Maher et al. They propose a system architecture based on reasoning through analogy with past cases, or situations (256). There are three main mechanisms within the system: an analogy mechanism that collects hypotheses about the variables; an exploration mechanism that searches through the solution space; and a generalizing mechanism that looks at experiences, and memorizes only what is needed to collect hypotheses (257-69). The main claim of the authors is that design learning is experiential, and by the help of analogy mechanism of the system they simulate it (270).

Oxman and Oxman propose a model for an electronic library of design precedents called PRECEDENTS. The model is composed of distinct chunks of knowledge called design stories. A formalism for a design story is proposed which represents the linkage between design issue, concept, and form in design stories (Precedents 275-285). Stories are structured in the memory according to a semantic network (276).

Kuhn and Herzog propose a method of representing, and retrieving design cases that is based on Wittgenstein’s language-game metaphor. They introduce the concept of language-game abstractions (LGA’s). A LGA combines precedent cases, the terms used for their descriptions, and the relations between these terms (69-79). Their main claim is that, use of language-games of architectural discourse avoids limitation of the scope of representation caused from the obligation of constructing a single consistent representation (81).
 

Conclusion

The use of AI for architectural systems seems to be a very promising, and an ever developing area of research in computer aided architectural design. It can be claimed that further work will be much more developed than the current ones, due to the rapid developments in computer technology.
 

Bibliography

Akin, O. Psychology of Architectural Design. London: Pion, 1986. 158 - 170.

Bench - Capon T. J. M. Knowledge Representation: An Approach to Artificial Intelligence. The A. P. I. C. Ser. 32. London: Academic Press,1990.

Bijl, A. "Designing with Words and Pictures in a Logic Modeling Environment." Computer Aided Architectural Design Futures. Proc. International Conference on Computer Aided Architectural Design Futures, Sept. 1985, Technical U. Delft, Delft, Netherlands. Ed. Alan Pipes. London: Butterworths, 1986. 128-145.

Carrara G. , and Y. E. Kalay. "Past, Present, Future: Process, and Knowledge in Architectural Design" Knowledge Based Computer Aided Architectural Design. Eds. Carrara Gianfranco, and Yehuda E. Kalay. Amsterdam: Elsevier, 1994. 389-396.

Chun, H. W. and Lai M. K. "Intelligent Critic System for Architectural Design" IEEE Transactions on Knowledge and Data Engineering 9.4 (1997): 625-638.

Flemming, U. "Artificial Intelligence and Design: A Midterm Review" Knowledge Based Computer Aided Architectural Design. Eds. Carrara Gianfranco, and Yehuda E. Kalay. Amsterdam:Elsevier, 1994. 1-24.

Galle, P., and L. B. Kovacs. "The Logic of Worms: A Study in Architectural Knowledge Representation." Environment and Planning B: Planning and Design 19 (1992): 5 - 31.

 - - - . "The Logic of Plaza Space: Representing Design Knowledge on Shape and Function." Environment and Planning B: Planning and Design 21 (1994): 159 - 177.

Guena, F. , and K. Zreik. "Analogy, Exploration and Generalization: Three Activities for Knowledge Based Architectural Systems." CAAD Futures ‘93. Proc. of the Fifth International Conference on Computer Aided Architectural Design Futures, Jul. 1993, Pittsburgh, PA. Amsterdam: Elsevier, 1993. 255 - 272.

Hamilton G. and  A. P. Harrison. Expert Systems for the Building Services Industry. London: BSRIA, 1988.

Kalay, Y. E. "Redefining the Role of Computer in Architecture: From Drafting Modelling Tools to Knowledge - Based Design Assistants" Computer Aided Design 17 (1985): 319 - 328.

Kocabas, S. "Functional Categorization of Knowledge" 1. Türk Yapay Zeka ve Yapay Sinir Aglari Sempozyumu, Jun. 1992, Bilkent U.  Eds. Kemal Oflazer et al. Ankara: Bilkent U. , 1992. 111 - 121.

Kuhn, C. , and M. Herzog. "Modelling the Representation of Architectural Design Cases." Automation Based Creative Design. Eds. Alexander Tzonis and Ian White. Amsterdam: Elsevier, 1994. 69 - 82.

Landsdown, J. "Requirements for Knowledge-based Systems in Design" Computer Aided Architectural Design Futures. International Conference on Computer Aided Architectural Design Futures, Sept. 1985, Technical U. Delft, Delft, Netherlands. Ed. Alan Pipes. London: Butterworths,1986. 120-128.

Logan, B. S. "Representing the Structure of Design Problems" Computer Aided Architectural Design Futures. International Conference on Computer Aided Architectural Design Futures, Sept. 1985, Technical U. Delft, Delft, Netherlands. Ed. Alan Pipes. London: Butterworths, 1986. 158 - 170.

Maher, M. L. , A. G. Silva Garza. , K. Bridge. , D. M. Zhang. "AI in Design: Case Based Reasoning in Design. " http://www.computer.org/pubs/expert/1997/features/x2034/x2034.htm (5 Nov. 1997)

McCullough, J. Knowledge Based Systems in Architecture. Stockholm: Swedish Council for Building Research, 1991.

Mitchell, W. J. "Three Paradigms for Computer Aided Design" Knowledge Based Computer Aided Architectural Design. Eds. Carrara Gianfranco, and Yehuda E. Kalay. Amsterdam: Elsevier, 1994. 389-396.

Oxman, R. , and R. Oxman. "Computability of Architectural Knowledge" The Electronic Design Studio: Architectural Knowledge and Media in the Computer Era. Eds. McCullough, Malcolm,William J.  Mitchell, and Patrick Purcell. Cambridge, Mass. : MIT Press,1990. 171-187.

- - - . "Precedents: Memory Structure in Design Case Libraries" CAAD Futures ‘93. Proc. of the Fifth International Conference on Computer Aided Architectural Design Futures, Jul. 1993 Pittsburgh, PA. Amsterdam: Elsevier, 1993. 273 - 287.

Rosenman, M. A. , J. S.  Gero, and R. Oxman, eds. "What’s in a Case: The Use of Case - bases, Knowledge - bases and Data - bases in Design." CAAD Futures ‘91. Proc. International Conference on Computer Aided Architectural Design Futures, Jul. 1991, Zurich. Ed. Gerhard Schmitt. Wiesbaden: Wieveg, 1992. 284 - 300.

Shih, S. "Case Based Representation and Adaptation in Design." CAAD Futures ‘91. Proc. International Conference on Computer Aided Architectural Design Futures, Jul. 1991, Zurich. Ed. Gerhard Schmitt. Wiesbaden: Wieveg, 1992. 372 - 385.

Waterman, D. A. A Guide to Expert Systems. Boston: Addison-Wesley, 1985.

Woods, W. A. "Important issues in Knowledge Representation." Proceedings IEEE, Institute of Electrical and Electronics Engineers, 74: 10 (1986): 1322 - 1334.

Zreik, K. "What Could Artificial Intelligence Know About the Knowledge Involved in Design Process?" CAAD Futures ‘91. Proc. International Conference on Computer Aided Architectural Design Futures, Jul. 1991, Zurich. Ed. Gerhard Schmitt. Wiesbaden: Wieveg, 1992. 284 - 386.