The task of information gathering in a distributed setting can be viewed in general terms as either distributed processing or distributed problem solving (DPS). Distributed processing is characterized by complete independence of subproblems. Agents need nothing other than local information to arrive at a subproblem solution of the required quality that can be synthesized with other agent subproblem solutions to arrive at a global solution. Distributed problem solving, on the other hand, is characterized by the existence of interdependencies between subproblems assigned to the individual agents, leading to a need for them to cooperate extensively during problem solving. They rely on communication to detect and exploit these interdependencies between subproblems. At the start, agents have only partial and incomplete views of global solution requirements. In spite of this deficiency in information, they may arrive at partial and tentative results that may be exchanged by the agents working on subproblems that are interdependent, to reduce the uncertainty that surrounds local problem solving. That is, agents may exploit the interdependencies between subproblems to their benefit[Lesser1991][Lesser1990].
In this paper, the information and knowledge resources comprising a corporate memory of an organization are viewed as distributed case bases. Doing so lets us map techniques for distributed problem solving into tools for knowledge manipulation in corporate memories.
The first technique is based on a method proposed by Nagendra Prasad, Lesser and Lander[Nagendra Prasad, Lesser, &Lander1996] for retrieval from distributed case bases. This method can be seen as an instantiation of aspects of the cooperative information gathering(CIG)[Oates, Nagendra Prasad, &Lesser1994] approach to intelligent information gathering from networked information resources. This approach relies on the ``FA/C'' paradigm[Lesser1991] previously developed as a framework for distributed problem solving. Oates, Nagendra Prasad and Lesser[Oates, Nagendra Prasad, &Lesser1994] provide an extensive discussion as to why it is better to treat information gathering in a networked environment as distributed problem solving. In a CIG task, potentially useful constraints may exist between different pieces of information. The discovery and exploitation of such constraints is necessarily a dynamic and incremental process that occurs during problem-solving and entails communication of partial results among agents in a timely and selective manner, to augment each agent's local view with a more global view. Given the incomplete nature of the local views of the individual agents, another important aspect of CIG is the explicit recognition of the role of solution and control uncertainty. Coupled with the fact that resources and time for conducting a search are limited in real-life problems, this leads to the notion of satisficing search. Another aspect of CIG is the explicit recognition and exploitation (or avoidance) of redundancy, leading to increased robustness or decreased resource demands depending on the context and the structure of the domain.
The second technique is the Federated Peer Learning-based cooperative Case-based Reasoning [Plaza, Arcos, &Mart´n1996]. Two modes of cooperative case-based reasoning are discussed: DistCBR where an agent can delegate its authority to another peer agent to solve a problem and ColCBR where an agent maintains authority while exploiting the experience of a peer agent. While remote evaluation capability supports DistCBR, ColCBR mode is supported by remote programming (or mobile code) capability of the underlying representation and communication framework. These modes let an agent exploit the collective memory in a distributed environment in a lazy, on-demand way.
In the following section, we first discuss why some of the resources comprising a corporate memory can be viewed as case bases. We then briefly discuss the Negotiated Retrieval Algorithm and the Federated Peer Learning-based cooperative Case-based Reasoning modes. Readers interested in further details are urged to refer to [Plaza, Arcos, &Mart´n1996][Nagendra Prasad, Lesser, &Lander1996].