The expedition of expertise : designing an expert system for design

Abstract

Principal assumptions are made during the early stage of the design process, fixing 70% to 80% of total building costs and most of a building's architectural and environmental qualities. The demands of any one constraint impose a whole set of assumptions that often result in a design that is satisfactory along only a few dimensions. Ruling out many alternatives at a stroke, such assumptions relieve the architect from exhaustively reviewing variations by removing opportunities from consideration. Both the power and crudeness of assumptions increase as constraints multiply and conflict. Having the ability to quickly and thoroughly evaluate assumptions and their consequences would allow architects to intelligently challenge and reform those assumptions and, as a result, to explore a broader range of possibilities for any particular design. This is particularly important in complex projects where the architect's primary role may be to orchestrate experts. This role is not insignificant, for the experts' recommendations will necessarily be bounded by their own concerns and will often conflict. The architect must assign values to design consequences and must provide the assumptions that the experts will base their recommendations on. The architect, then, focuses on making assumptions and interpreting evaluations. But assumptions are of ten outside analysis; rules of thumb, based on experience, generally prevail. Knowledge- based computer expert systems are a promising path of research for the support of conscious and explicit assumption-making. The crucial question for this technology and the central topic of this thesis is how to structure knowledge for use in such a system. My primary goal is to offer a representation of the knowledge involved in window design (a simpler and somewhat isolable subset of building design), a representation comprehensive enough to be useful, but also flexible enough to support differing design processes and decision sequences.