Design Rationale Capture as Knowledge Acquisition Tradeoffs in the Design of Interactive Tools

This paper introduces a panel to be held at the Knowledge Acquisition Track of the Machine Learning Workshop (ML91). This panel will focus on the problem of acquiring design rationale knowledge from humans for later reuse. The design of tools for design rationale capture reveals several fundamental issues for knowledge acquisition, such as the relationships among formality and expressiveness of representations, and kinds of automated support for elicitation and analysis of knowledge. This paper sets the background for discussion by identifying dimensions of a design space for design rationale tools, and then includes position statements from each panelist arguing for various positions in this space. 1 THE PROBLEM OF DESIGN RATIONALE CAPTURE For many engineering tasks, including redesign, verification, and diagnosis, it is important to know why an artifact was designed the way it was. Design rationale is a general term referring to the knowledge or reasoning underlying a design. Although there is wide agreement that design rationale is valuable knowledge, there are few means for effectively acquiring it. Technology for recording and making this knowledge easily accessible could be used to facilitate collaborative work among designers and to support communication with people who make decisions about manufacturing, operations procedures, diagnostic procedures, configuration, testing, and marketing. Yet today design rationale information is hard to elicit from designers, fragmented, easily lost, and generally not captured in machine-intelligible form. Understanding the rationale of a design often requires a broad range of knowledge, including the design requirements (including constraints and evaluation metrics), the structure of the artifact, the reasons for choosing particular components or implementation approaches, the assumptions about the context in which the artifact is to be used, and the institutional experience of designing and manufacturing similar artifacts. In this paper we frame design rationale capture as a knowledge acquisition problem, where the task is to acquire design knowledge in a form that can be reused with automated assistance (e.g., for information retrieval or generation of rationale explanations). Several approaches have been proposed to facilitate the capture and reuse of design rationale. The panelists will present a few exemplary approaches, and explore some of the tradeoffs. In particular, the discussion will focus on the degree of autonomy versus interactivity in these knowledge acquisition systems. 2 SOME APPROACHES TO DESIGN RATIONALE CAPTURE Although still in the research stage, a range of approaches have been proposed for providing automated assistance to capture and reuse design rationale knowledge. Some notable points in the space include: ➀ Provide electronic notebooks that encourage designers to record their design processes on line [Lakin et al., 1989]. 2 Elicit semistructured text about the design rationale, based on argument-style discussions among designers [Conklin and Begeman, 1988; Lee, 1990; McCall, 1986]. 3 Elicit semistructured arguments in the context of knowledge-based design critics [Fischer, McCall, and Morch, 1989].