SIBYL: A Tool for Managing Group Decision Rationale Jintae Lee Center for Coordination Science and Artificial Intelligence Laboratory, MIT E40-140, 1 Amherst St Cambridge, MA 02139 jin@ai.mit.edu We describe SIBYL, a system that supports group decision making by representing and managing the qualitative aspects of decision making processes: such as the alternatives, the goals to be satisfied, and the arguments evaluating the alternatives with respect to these goals. We use an example session with SIBYL to illustrate the language, called DRL, that SIBYL uses for representing these qualitative aspects, and the set of services that SIBYL provides using this language. We also compare SIBYL to other systems with similar objectives and discuss the additional benefits that SIBYL provides. In particular, we compare SIBYL to gIBIS, a well- known "tool for exploratory policy discussion", and claim that SIBYL is mainly a knowledge-based system which uses a semi-formal representation, whereas gIBIS is mainly a hypertext system with semantic types. We conclude with a design heuristic, drawn from our experience with SIBYL, for systems whose goal includes eliciting knowledge from people. Explicit representation of a decision rationale, i.e. the deliberations leading to a decision, can provide many potential benefits, especially in the context of group decision making. The knowledge that people bring to a decision becomes available for others to augment or respond. The representation of a decision making process serves as a documentary record of how the decision develops, which in turn can serve as a basis for learning and justification. [Yakemovic and Conklin 90] provides a good documentation of these benefits and their ramifications. In addition, if the representation is well-structured, the system can provide services, such as managing dependencies among what is represented, that help people make better decisions. To achieve these benefits, the language for representing decision rationales should allow people to express naturally what they need to express, and people should get rewarded for using the language in their decision making. Most existing languages for representing decisions, such as decision trees [Raiffa 68] and influence diagrams [Howard and Matheson 81], are not designed to represent the deliberation aspect of decision making, but only the results of such deliberations. The few whose goal is to represent the decision rationale, such as gIBIS [Conklin and Begeman 88], are not either expressive enough and/or do not provide enough services to reward the user , as we discuss below. In this paper, we describe a system, called SIBYL, which is designed to meet these requirements by providing a language structured for the task of decision making and the set of services that rewards the users of this language. We proceed as follows. First, we elaborate the motivations underlying SIBYL. We then describe SIBYL; we briefly describe the language, called DRL (Decision Representation Language), that SIBYL uses for representing decision processes, and illustrate its use in an example session with SIBYL. After describing SIBYL, we compare SIBYL to other systems with similar objectives, especially to gIBIS, a hypertext system whose goal is to record design rationales. We discuss how SIBYL is similar to gIBIS, how SIBYL extends gIBIS, and what we gain as a result. We conclude with a design heuristic, drawn from our experience with SIBYL, for systems which have to elicit knowledge from people to provide its services.