Modeling Creativity and Knowledge-Based Creative Design Book by John S. Gero, Mary Lou Maher; Lawrence Erlbaum Associates, 1993

Preface Design is one of the most significant of the purposeful acts of human beings. It creates a new, artificial world for us to inhabit and as such changes us all. The burgeoning ability to model and implement design processes as reasoning systems, making use of artificial intelligence research and techniques, has given new impetus to the study of design theory and methodology. There is general acceptance that design activities can be categorized as being either routine or nonroutine. The distinction being drawn by this categorization is to separate design processes, in which the functions (i.e., goals or requirements) are known as are the available structures and the processes which map function to structure, from those in which one or more of function, structure or mapping is not known a priori. This allows us to begin to address one of the fundamental issues in design: that of creativity. Dictionary definitions of creativity are too imprecise to provide either a formal or operational definition that is useful. An operational definition of creativity in design that has been found to be useful states that creativity involves the introduction of new variables into the design process, variables which were not originally considered by the designer or design system. Such a definition begs important questions related to the value of artifacts designed by such processes.

The problem-solving paradigm used in artificial intelligence appears to preclude the introduction of new variables because it was initially concerned with finding ways of achieving predefined goals from a specified initial state while passing through known states in some state-space representation. Because creative design involves exploration (i.e., finding new goals, new states, and new state transition processes), it would appear that the original ideas of problem solving are applicable only to routine design. However, the concept of problem solving can be readily extended to encompass the notions involved in exploration. Over the last decade research into design processes utilizing ideas and models drawn from artificial intelligence has resulted in a better understanding of design as a process, particularly routine design. Most of the current research activity directly or indirectly deals with routine design only. Not surprisingly many practicing designers state that the level of understanding represented by models of routine design is only of mild interest to them because of the lack of any ideas about creativity embodied in them. This book provides a set of chapters in the areas of modeling creativity and knowledge-based creative design which examines the potential role and form of computer-aided design that supports creativity. The objectives are to define the state-of-the-art of computational creativity in design, and to identify research directions. This book is published at a time when the field of computational creativity in design is still immature, hopefully it will assist the field in reaching maturity and influence the directions of growth.

1 Introduction John S. Gero and Mary Lou Maher
Modeling creativity and knowledge-based creative design is a topic that invokes both positive and negative responses from designers and researchers in artificial intelligence and design. The topic is a-difficult one given our current understanding of creativity and potential computational models of creativity. For some, the notion of understanding creativity is in itself a contradiction of terms. In this chapter we raise some of the issues in proposing that creativity can be modeled and that knowledge-based approaches have the potential to provide computational models of creativity.

1.1 Creative design Design distinguishes itself from other human activities in a variety of ways. One of the most important ways is that the resulting artifact is expected to be different, albeit even if only slightly, from previous artifacts. This places design within a social context because the differences are evaluated within that context. The magnitude and quality of these differences are commonly used to separate artifacts into two categories although the boundary between them is fuzzy and constantly changes. These two categories are labeled routine and nonroutine designs. The labels can be applied equally to the processes of design that produced them as to the artifacts or designs themselves. The interest in drawing this distinction lies in the implications it has for articulating computational processes that support design activity. The inference is that processes for routine design are likely to be different from those for nonroutine design. Routine designs may be defined as ones that are recognized as not being different from previously produced designs in their class in any substantive way. Thus, in structural engineering, designing a reinforced concrete beam for a given span and load subject to the normal goals and constraints generally will result in a rectangular cross-section of particular dimensions with certain reinforcement sizes and placement. Whereas another designer may produce a design that has different dimensions and reinforcement sizes and placement, the two designs will be recognized as being remarkably similar. Furthermore, the same designer designing another reinforced concrete beam for another span and load may produce a design with a rectangular crosssection of different dimensions and reinforcement sizes and placement and this design is also recognized as being similar to the previous design.

What makes these designs similar, it can be argued, is that they all exhibit the same properties but with different magnitudes. More formally, we state that these designers all chose to use the same design variables to work with and produce different values for those variables dependent on their perception of the situation. They may well have used similar processes as well to produce the values of the variables. Two processes are of interest. The first is concerned with the selection of the variables of interest and the second is concerned with producing values for those variables. The design variables with their values describe a design. Nonroutine designs may be defined as ones that are recognized as being different from previously produced designs in their class in some substantive sense. We describe how we understand and interpret these differences later...