Journal of Intelligent Information Systems, 1, 35-55 (1992) 9 1992 Kluwer Academic Publishers, Boston. Manufactured in The Netherlands. The Roles of Artificial Intelligence in Information Systems GIO WIEDERHOLD Stanford University, Stanford, California Between the minds that plan and the hands that build there must be a mediator Intertitle from Fritz Lang: Metropolis, UFA 1926. Abstract. By classifying information processing tasks which are suitable for artificial intelligence approaches we determine an architectural structure for large systems. We visualize a three-layer architecture of private applications, mediating information servers, and an infrastructure which provides information resources. The base information resources are likely to use algorithmic techniques, since they will deal with many similar base objects. Their results are at higher level of abstraction, diverse, and fewer in number. The information servers must consider the scope, assumptions, and meaning of those intermediate results. Such processing will require techniques grounded in artificial intelligence concepts. Applications will need artificial intelligence techniques to augment the human interface and provide high-level decision support. Information processing in the intermediate layer is domain-specific and a module is constrained to a single ontology. Processing here is comprised of search and control of search, focusing, pruning, fusion, and other means of data reduction. There are also control tasks associated with effective resource management. Their results are then composable by higher-level applications, which have to solve problems involving multiple subtasks. The architecture presented here is a generalization of a server-client model. The mediating server modules will need a machine-friendly interface to support the application layer. The partitioning enhances maintainability, but raises questions of effectiveness and efficiency. Without new and composable structures we will be stuck with a mixture of obsolete large systems and isolated new applications. A formal partitioning provides a model where subproblems become accessible to research. Interoperation is now a distinct source of research problems. We identify some of these issues, and hope that composability of solutions will permit progress in building effective large systems. Keywords: knowledge bases, databases, integration abstraction, fusion, architecture, autonomous modules 1. Introduction We define information systems as computer-based systems which can access a variety of computer-acquired, -stored, or -generated base data, and select and process that data to provide specific information to aid planners and management in their decision making. While the source data are often voluminous, a modest