An `AI in Design' View of Design
David C. Brown & Dan L. Grecu
AI in Design research is concerned with the use of AI techniques to study design. This is done by making hypotheses about designing and producing models of design knowledge and activity. Using these hypotheses and models, computer programs (i.e., design systems) are built that can aid designers, and carry out some portion of the design process. Thus hypotheses about design are tested and refined using design systems.
This paper provides a set of questions that can be used for pedagogical purposes to analyse and categorize computational research into design. The following sections group the questions into related topics.
Design systems need to be characterized in order to be compared and contrasted. This is particularly useful for teaching about design. Every design system can be given an initial, general characterization in terms of a few key attributes. These features correspond to the first questions asked about every project and create a context for all other information about the system.
Different design systems work on different design problems. Defining the problem is essential for characterizing a system. The description of a design problem lies in a space with dimensions such as ``type of design'', the ``design target'', and the ``design theory''. The other dimensions of a design problem description are inherited from general problem-solving theories: the input, the output and the assumptions made about the problem.
Solving a design problem efficiently requires an adequate representation. AI researchers have identified knowledge types which structure information based on various needs. One of the major tasks of any design system developer is to identify all the pieces of knowledge used by the system and to map them onto the most appropriate representations provided by AI. A list of the knowledge and representations used by a system can be used in its characterization.
A description of the methodology brings together a set of diverse aspects concerned with how the system builds, and possibly improves, a solution to a design problem. The characterization of design problem solving cannot be reduced to just the particular method used. Many other issues are critical: knowing how and where to look for a design solution; what types of reasoning to use; and what goals to pursue besides that of finding a solution.
Design systems are assembled from functional modules into various structures (i.e., architectures). The architecture used strongly affects how, and how well, a system works. As more sophisticated development tools and components become available, system architectures tend to be increasingly complex and more dimensions are needed for their description.
System evaluation remains one of the most difficult aspects of the development of a design system. The primary criteria remain the cost of the resources used, as well as the satisfaction of the user with the quality of the results and with the ease of use of the system. The methods used to evaluate design systems are strongly indicative of the intentions of the developers and should be used as part of the characterization of each system.
In this paper we have provided a set of questions that can be used to analyse and categorize computational research into design; have introduced a selection from the core literature in AI in Design; and presented a perspective of design based on that literature.