Learning from Prototypes

Structured methods for the analysis and design of information systems have largely focused on representations and control mechanisms for the outcomes of the design process. Prototyping methods are more sensitive to critiques during the designprocess itself but do not preserve knowledge about it explicitly. In this paper, a systems arc iitecture called REMAP is presented that accumulates design process knowledge to manage systems evolution. To accomplish this, REMAP acquires and maintains dependencies among the design decisions made during a prototyping process. It includes a model for learning general design rules from such dependencies which can be applied to prototype refinement, systems maintenance, and design re-use. Introduction ulated explicitly jy users or analysts. Second, when systems are developed in a piecemeal fashion following the The process of large systems development is often iteraprototyping idea, analysts apply analogies to transfer tive, involving continuous modifications to programs experience gainek! from one subsystem to "similar combefore a "satisfactory" design emerges. Designers have ponents" of andther. Unfortunately, current developattempted to use aprotoryping approach whereby a workment methodologies preserve none of these aspects of ing prototype system is assembled quickly on the basis of process knowledge, making the process of prototype an initial assessment of a a problem situation, and then refinement and transfer of experience ad-hoc and susrefined repeatedly in response to critiques from users or ceptible to error design personnel. While this approach may offer significant advantages over "structured" approaches in terms It appears that the systems development process would of earlier user involvement, a major drawback is that the benefit greatly if the dependencies among decisons could initial construction of the system and the process of sucbe represented e plicitly, and more importantly, if the cessive refinement can be haphazard, failing to take coggeneral basis for them could be extracted during the nizance of the rationales for the initial design decisions course of analysi6 and development. This could lead to a and for successive changes in these decisions. more systematic| modification of prototypes and improved maintenance of full-blown implementations. PerThis paper employs a case study in the oil industry to haps more importantly, this knwledge could be used to analyze these shortcomings in some depth, and presents identify analogous features of different systems prean artificial-intellignece based architecture called cisely, enabling the use of cumulative learning for subREMAP (REpresentation and MAintenance of Process sequent designs in the same general application area. knowledge) which enhances the iterative design procedure typical for the prototyping approach by the capaThe paper is orgapized as follows: Section 2 begins with bility of preserving knowledge about the design process, a brief description of the prototyping process; detailed and applying this knowledge in analogous design situareal-world exam#les are then used to show the need to tions. maintain process knowledge. A formal model of our approach is presented in section 3, along with an overThe case study has revealed several types of process view of a partial limplementation of the REMAP archiknowledge that appear to be central to systems developtecture. Section ]* provides a discussion relating the ment. First, the design process consists of a sequence of model to previous work in systems analysis and artificial interdependent design decisions. The dependencies intelligence. We conclude with a summary of possible among decisions are typically based on general applicaapplications whi :h may benefit from the REMAP tion-specific rules; however, these rules are seldom articapproach.