A growing collection of communities worldwide is researching phenomena of human organizations using computational methods. There are pressing research issues in both the development and ongoing operations of organizations, and in organization theory and analysis, that are highly amenable to techniques for computational modeling, theory-building/testing, and experimentation. This short note is intended to provide a quick orientation and perspective on compmtafional ory nization research, a phenomenon and problematic that has been the focus of this emerging research community, both within and outside of AI, DAI, and multiagent systems arenas. Organizations and organization-level phenomena are already ubiquitous in modern life. New technologies and new social and institutional arrangements are emerging rapidly, changing our notions of what it means to be organized, how best to organize, how to form, change and stabilize organizations, and so on.
Open multi-agent systems can be defined as multi-agent systems which agents can enter and leave freely. In general, the reseaxch work on multi-agent systems has not dealt explicitly with problems brought about by systems openness, in particnlm', the problem of organization dynamics. By dynamics of organization we mean the set of transformations to which the organization of an open multiagent system can he submitted during the system's functioning, due to the mutual influence of: its functional constraints, the changes in its environment, the entrance and the departure of agents in its strncmre, or the conceiver's or the user's intervention. One of the general problems that must be faced by a representational framework for dynamical organizations is that of representing means and requirements for the maintainance of the organizational integrity of a system in view of changes in its organization. Crucial to this is the problem of having agents reliably acquiring, maintaining and reasoning about dynamical descriptions of the changing organization. Out work addresses basic issues concerned with the latter problem, in preparation for tackling the former.
In this paper, we address the following question: to what extent can organization design be treated as a routine design problem, with a well-defined space o] possibilities and explicit evaluation criteria? The kinds of organizations we are particularly interested in are computational organizations which consist of computational agents that cooperate with one another on their organizational tasks. We present a model of computational organization design that uses predictive knowledge about how exactly various taskenvironmental and organizational factors determine the performance of the organization. Introduction We have been investigating ways to endow a set of agents with the capability to cooperate intelligently under various task environments. Our specific approach has been to give the agents the ability to form and change their task organization under potentially complex, uncertain, and/or dynamic task environments. The process of changing an organization by one or more members of the organization in order to improve performance has been called organization self-design (OSD) in the distributed artificial intelligence (DAI) literature [Corkill, 1983, Corkill and Lesser, 1983, Gasser and Ishida, 1991]. In [So and Durfee, 1993], we have presented a model of OSD which involves an organization design and evaluation component using a definite performance criterion. In this presentation, we elaborate on our model, and address the question of to what extent can organization design be treated as a routine design problem, with a well defined space of possibilities and explicit evaluation criteria.
Notes on Organization Design from the Perspective of AI Les Gasser Short Paper for AAAI COD Symposium An AIoriented approach to organization modeling, analysis, and design entails building specific, computational models of things that we often (informally) call organizations". Such models will be formal in the sense that they are computational---that is, they will have some defined semantics in the sense that a computer will take specific actions when operating upon them in a design or analysis process [Gasser et al. 1993]. Presently, we seem to have a tradeoffbetween clearl pragmatically and computationally defining the concept of"organization" on the one hand, and capturing the richness of features of interesting (e.g. The aim of this paper is to begin to think through the needs a computational theory of organization that is clear, rich, and representative, and yet still useful for both computational analysis and design. Some well-known attempts to define organization in AI are direct, simple and mundane---these include approaches such as those used in the UMASS DVMT [Durfee et al. 1987] and our own OSD project [Ishida et al. 1992], in which "organization" is seen as a (reconfigurable) mapping of capabilities to "agents" (that is, functional specialization) coupled with a distribution of knowledge and control to exploit this specialization so as to mil imlze resource use, called "coordination" or "distributed, network-wide control."
The purpose of this paper is to identify some research issues related to applying a multiagent learning approach to knowledge management (KM) for a business organization. Although multiagent learning can not address the entire spectrum of KM research issues, we describe a potential approach to its application particularly in the context of a large business organization with diverse product offerings. We look at what multiagent learning is, what the research issues are for ontology-based KM in a large, diverse business organization, and how a distributed intelligent agent-based approach might address some of these issues.